Direct Preference Optimization (DPO) has emerged as a simple and effective approach for aligning models with human preferences. However, existing DPO-based methods suffer from 3 key drawbacks: they rely on only a single positive-negative preference pair per question, restricting the diversity and richness of feedback; they often emphasize minimizing negative preference scores while neglecting to strengthen the positive preferences; and they depend on either human-annotated preferences or expert model outputs - both expensive and difficult to scale. Moreover, the deterministic ranking assumptions of recent Group-based preference optimization methods break down in open-ended tasks such as Visual Question Answering (VQA), where multiple answers can be equally plausible but differ subtly in relevance or specificity. Given this subtle variance in preferences, we propose to perform ranking over groups of preferences rather than relying on fine-grained ranking of individual ones, which is often noisy and subjective. To address these challenges, we introduce Self-Supervised Visual Preference Alignment via Differentiable Multi-Preference Multi-Group Ranking (SMPRO), a novel framework that (1) self-generates rich, diverse preference groups while eliminating the need for external annotations, (2) employs a fully differentiable ranking objective based on sorting networks to capture nuanced preference gradients across arbitrary numbers of preferences both within and across these groups, and (3) incorporates multiple positive preferences to enrich the positive preference group, capturing subtle distinctions among high-quality preferences. Extensive experiments across diverse visual tasks show that our approach achieves state-of-the-art performance in self-supervised setting. Specifically, our model surpasses existing baselines, achieving notable gains such as 82.4% on MM-Bench, 63.2% on MMStar, 94.6% on LLaVA-W, and 81.9% on AI2D. These results underscore the effectiveness of our approach in capturing richer preference signals and demonstrate its scalability for open-ended, ambiguous VQA tasks.

Improving the safety of vision-language models like CLIP via fine-tuning often comes at a steep price, causing significant drops in their generalization performance. We find this trade-off stems from rigid alignment strategies that force unsafe concepts toward single, predefined safe targets, disrupting the model's learned semantic structure. To address this, we propose a proximity-aware approach: redirecting unsafe concepts to their semantically closest safe alternatives to minimize representational change. We introduce SafeR-CLIP, a fine-tuning framework that applies this principle of minimal intervention. SafeR-CLIP successfully reconciles safety and performance, recovering up to 8.0% in zero-shot accuracy over prior methods while maintaining robust safety. To support more rigorous evaluation, we also contribute NSFWCaps, a new benchmark of 1,000 highly-aligned pairs for testing safety under distributional shift. Our work shows that respecting the geometry of pretrained representations is key to achieving safety without sacrificing performance.

Rohit Gupta, Jayakrishnan Unnikrishnan, Fan Fei, Sheng Liu, Son Tran, Mubarak Shah. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Image geolocalization, in which an AI model traditionally predicts the precise GPS coordinates of an image, is a challenging task with many downstream applications. However, the user cannot utilize the model to further their knowledge beyond the GPS coordinates; the model lacks an understanding of the location and the conversational ability to communicate with the user. In recent days, with the tremendous progress of large multimodal models (LMMs), both proprietary and open-source researchers have attempted to geolocalize images using LMMs. However, the issues remain unaddressed; beyond general tasks, for more specialized downstream tasks, such as geolocalization, LMMs struggle. In this work, we propose solving this problem by introducing a conversational model, GAEA, that provides information regarding the location of an image as the user requires. No large-scale dataset exists that enables the training of such a model. Thus, we propose GAEA-1.4M, a comprehensive dataset comprising over 800k images and approximately 1.4M question-answer pairs, constructed by leveraging OpenStreetMap (OSM) attributes and geographical context clues. For quantitative evaluation, we propose a diverse benchmark, GAEA-Bench, comprising 3.5k image-text pairs to evaluate conversational capabilities equipped with diverse question types. We consider 11 state- of-the-art open-source and proprietary LMMs and demonstrate that GAEA significantly outperforms the best open-source model, LLaVA-OneVision, by 18.2% and the best proprietary model, GPT-4o, by 7.2%. Our dataset, model and codes are available https://ucf-crcv.github.io/GAEA.

Artificial intelligence (AI) is transforming sports analytics by enabling automated, real-time understanding of soccer matches. Traditional approaches that rely on isolated data streams struggle to capture the full game context. We introduce SoccerChat, a multimodal conversational AI framework that fuses visual and textual information for comprehensive soccer video comprehension. Building on the SoccerNet dataset, we enrich it with jersey color annotations and automatic speech recognition (ASR) transcripts and curate a video-instruction dataset containing 48,677 question-answer (QA) pairs. Fine-tuning Qwen2- VL-7B- Instruct on this resource yields the Soc-cerChat model, which supports accurate event interpretation, classification, and referee assistance. Experiments across action classification and referee QA tasks demonstrate strong general event understanding and competitive officiating analysis. These results highlight the importance of multimodal integration for explainable, interactive, and trustworthy AI-driven sports ana-lytics. Links to the code, dataset, and model weights are available at https://2ithub.com/simula/SoccerChat.

Text-to-image diffusion models have demonstrated remarkable capabilities in creating images highly aligned with user prompts, yet their proclivity for memorizing training set images has sparked concerns about the originality of the generated images and privacy issues, potentially leading to legal complications for both model owners and users, particularly when the memorized images contain proprietary content. Although methods to mitigate these issues have been suggested, enhancing privacy often results in a significant decrease in the utility of the outputs, as indicated by text-alignment scores. To bridge the research gap, we introduce a novel method, PRSS, which refines the classifier-free guidance approach in diffusion models by integrating prompt re-anchoring (PR) to improve privacy and incorporating semantic prompt search (SS) to enhance utility. Extensive experiments across various privacy levels demonstrate that our approach consistently improves the privacyutility trade-off, establishing a new state-of-the-art.

Direct Preference Optimization (DPO) has been proposed as an effective and efficient alternative to reinforcement learning from human feedback (RLHF). In this paper, we propose a novel and enhanced version of DPO based on curriculum learning for text-to-image generation. Our method is divided into two training stages. First, a ranking of the examples generated for each prompt is obtained by employing a reward model. Then, increasingly difficult pairs of examples are sampled and provided to a text-to-image generative (diffusion or consistency) model. Generated samples that are far apart in the ranking are considered to form easy pairs, while those that are close in the ranking form hard pairs. In other words, we use the rank difference between samples as a measure of difficulty. The sampled pairs are split into batches according to their difficulty levels, which are gradually used to train the generative model. Our approach, Curriculum DPO, is compared against state-of-the-art fine-tuning approaches on nine benchmarks, outperforming the competing methods in terms of text alignment, aesthetics and human preference. Our code is available at https://github.com/CroitoruAlin/Curriculum-DPO.

Recent advances in Multi-Modal Large Language Models (M-LLMs) show promising results in video reasoning. Popular Multi-Modal Large Language Model (M-LLM) frameworks usually apply naive uniform sampling to reduce the number of video frames that are fed into an M-LLM, particularly for long context videos. However, it could lose crucial context in certain periods of a video, so that the downstream M-LLM may not have sufficient visual information to answer a question. To attack this pain point, we propose a light-weight M-LLM-based frame selection method that adaptively select frames that are more relevant to users’ queries. In order to train the proposed frame selector, we introduce two supervision signals (i) Spatial signal, where single frame importance score by prompting an M-LLM; (ii) Temporal signal, in which multiple frames selection by prompting Large Language Model (LLM) using the captions of all frame candidates. The selected frames are then digested by a frozen downstream video M-LLM for visual reasoning and question answering. Empirical results show that the proposed M-LLM video frame selector improves the performances various downstream video Large Language Model (video-LLM) across medium (ActivityNet, NExT-QA) and long (EgoSchema, LongVideoBench) context video question answering benchmarks.

Composed Image Retrieval (CIR) is a complex task that aims to retrieve images based on a multimodal query. Typical training data consists of triplets containing a reference image, a textual description of desired modifications, and the target image, which are expensive and time-consuming to acquire. The scarcity of CIR datasets has led to zero-shot approaches utilizing synthetic triplets or leveraging vision-language models (VLMs) with ubiquitous web-crawled image-caption pairs. However, these methods have significant limitations: synthetic triplets suffer from limited scale, lack of diversity, and unnatural modification text, while image-caption pairs hinder joint embedding learning of the multimodal query due to the absence of triplet data. Moreover, existing approaches struggle with complex and nuanced modification texts that demand sophisticated fusion and understanding of vision and language modalities. We present CoLLM, a one-stop framework that effectively addresses these limitations. Our approach generates triplets on-the-fly from image-caption pairs, enabling supervised training without manual annotation. We leverage Large Language Models (LLMs) to generate joint embeddings of reference images and modification texts, facilitating deeper multimodal fusion. Additionally, we introduce Multi-Text CIR (MTCIR), a large-scale dataset comprising 3.4M samples, and refine existing CIR benchmarks (CIRR and Fashion-IQ) to enhance evaluation reliability. Experimental results demonstrate that CoLLM achieves state-of-the-art performance across multiple CIR benchmarks and settings. MTCIR yields competitive results, with up to 15% performance improvement. Our refined benchmarks provide more reliable evaluation metrics for CIR models, contributing to the advancement of this important field. Project page is at collm-cvpr25.github.io.

Existing Large Multimodal Models (LMMs) generally focus on only a few regions and languages. As LMMs continue to improve, it is increasingly important to ensure they understand cultural contexts, respect local sensitivities, and support low-resource languages, all while effectively integrating corresponding visual cues. In pursuit of culturally diverse global multimodal models, our proposed All Languages Matter Benchmark (ALM-bench) represents the largest and most comprehensive effort to date for evaluating LMMs across 100 languages. ALM-bench challenges existing models by testing their ability to understand and reason about culturally diverse images paired with text in various languages, including many low-resource languages traditionally underrepresented in LMM research. The benchmark offers a robust and nuanced evaluation framework featuring various question formats, including true/false, multiple choice, and open-ended questions, which are further divided into short and long-answer categories. ALM-bench design ensures a comprehensive assessment of a model’s ability to handle varied levels of difficulty in visual and linguistic reasoning. To capture the rich tapestry of global cultures, ALM-bench carefully curates content from 13 distinct cultural aspects, ranging from traditions and rituals to famous personalities and celebrations. Through this, ALM-bench not only provides a rigorous testing ground for state-of-the-art open and closed-source LMMs but also highlights the importance of cultural and linguistic inclusivity, encouraging the development of models that can serve diverse global populations effectively. Our benchmark is publicly available at https://mbzuai-oryx.github.io/ALM-Bench/.

Bhuiyan Sanjid Shafique, Ashmal Vayani, Muhammad Maaz, Hanoona Abdul Rasheed, Dinura Dissanayake, Mohammed Irfan Kurpath, Yahya Hmaiti, Go Inoue, Jean Lahoud, Md. Safirur Rashid, Shadid Intisar Quasem, Maheen Fatima, Franco Vidal, Mykola Maslych, Ketan Pravin More, Sanoojan Baliah, Hasindri Watawana, Yuhao Li, Fabian Farestam, Leon Schaller, Roman Tymtsiv, Simon Weber, Hisham Cholakkal, Ivan Laptev, Shin’ichi Satoh, Michael Felsberg, Mubarak Shah, Salman Khan, Fahad Shahbaz Khan. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recent advancements in 3D Large Language Models (3DLLMs) have highlighted their potential in building general-purpose agents in the 3D real world, yet challenges remain due to the lack of high-quality robust instruction-following data, leading to limited discriminative power and generalization of 3DLLMs. In this paper, we introduce Robin3D, a powerful 3DLLM trained on large-scale instruction-following data generated by our novel data engine, Robust Instruction Generation (RIG) engine. RIG generates two key instruction data: 1) the Adversarial Instruction-following data, which features mixed negative and positive samples to enhance the model's discriminative understanding. 2) the Diverse Instruction-following data, which contains various instruction styles to enhance model's generalization. As a result, we construct 1 million instruction-following data, consisting of 344K Adversarial samples, 508K Diverse samples, and 165K benchmark training set samples. To better handle these complex instructions, Robin3D first incorporates Relation-Augmented Projector to enhance spatial understanding, and then strengthens the object referring and grounding ability through ID-Feature Bonding. Robin3D consistently outperforms previous methods across five widely-used 3D multimodal learning benchmarks, without the need for task-specific fine-tuning. Notably, we achieve a 7.8\% improvement in the grounding task (Multi3DRefer) and a 6.9\% improvement in the captioning task (Scan2Cap).

Timestamp prediction aims to determine when an image was captured using only visual information, supporting applications such as metadata correction, retrieval, and digital forensics. In outdoor scenarios, hourly estimates rely on cues like brightness, hue, and shadow positioning, while seasonal changes and weather inform date estimation. However, these visual cues significantly depend on geographic context, closely linking timestamp prediction to geo-localization. To address this interdependence, we introduce GT-Loc, a novel retrieval-based method that jointly predicts the capture time (hour and month) and geo-location (GPS coordinates) of an image. Our approach employs separate encoders for images, time, and location, aligning their embeddings within a shared high-dimensional feature space. Recognizing the cyclical nature of time, instead of conventional contrastive learning with hard positives and negatives, we propose a temporal metric-learning objective providing soft targets by modeling pairwise time differences over a cyclical toroidal surface. We present new benchmarks demonstrating that our joint optimization surpasses previous time prediction methods, even those using the ground-truth geo-location as an input during inference. Additionally, our approach achieves competitive results on standard geo-localization tasks, and the unified embedding space facilitates compositional and text-based image retrieval.

Composed video retrieval is a challenging task that strives to retrieve a target video based on a query video and a textual description detailing specific modifications. Standard retrieval frameworks typically struggle to handle the complexity of fine-grained compositional queries and variations in temporal understanding limiting their retrieval ability in the fine-grained setting. To address this issue, we introduce a novel dataset that captures both fine-grained and composed actions across diverse video segments, enabling more detailed compositional changes in retrieved video content. The proposed dataset, named Dense-WebVid-CoVR, consists of 1.6 million samples with dense modification text that is around seven times more than its existing counterpart. We further develop a new model that integrates visual and textual information through Cross-Attention (CA) fusion using grounded text encoder, enabling precise alignment between dense query modifications and target videos. The proposed model achieves state-of-the-art results surpassing existing methods on all metrics. Notably, it achieves 71.3\% Recall@1 in visual+text setting and outperforms the state-of-the-art by 3.4\%, highlighting its efficacy in terms of leveraging detailed video descriptions and dense modification texts. Our proposed dataset, code, and model are available at :https://github.com/OmkarThawakar/BSE-CoVR

In the image acquisition process, various forms of degradation, including noise, blur, haze, and rain, are frequently introduced. These degradations typically arise from the inherent limitations of cameras or unfavorable ambient conditions. To recover clean images from their degraded versions, numerous specialized restoration methods have been developed, each targeting a specific type of degradation. Recently, all-in-one algorithms have garnered significant attention by addressing different types of degradations within a single model without requiring the prior information of the input degradation type. However, most methods purely operate in the spatial domain and do not delve into the distinct frequency variations inherent to different degradation types. To address this gap, we propose an adaptive all-in-one image restoration network based on frequency mining and modulation. Our approach is motivated by the observation that different degradation types impact the image content on different frequency subbands, thereby requiring different treatments for each restoration task. Specifically, we first mine low- and high-frequency information from the input features, guided by the adaptively decoupled spectra of the degraded image. The extracted features are then modulated by a bidirectional operator to facilitate interactions between different frequency components. Finally, the modulated features are merged into the original input for a progressively guided restoration. With this approach, the model achieves adaptive reconstruction by accentuating the informative frequency subbands according to different input degradations. Extensive experiments demonstrate that the proposed method, AdaIR, achieves state-of-the-art performance on different image restoration tasks, including image denoising, dehazing, deraining, motion deblurring, and low-light image enhancement. The code is available at https://github.com/c-yn/AdaIR.

Text-to-image diffusion models have achieved unprecedented proficiency in generating realistic images. However, their inherent tendency to memorize and replicate training data during inference raises significant concerns, including potential copyright infringement. In response, various methods have been proposed to evaluate, detect, and mitigate memorization. Our analysis reveals that existing approaches significantly underperform in handling local memorization, where only specific image regions are memorized, compared to global memorization, where the entire image is replicated. Also, they cannot locate the local memorization regions, making it hard to investigate locally. To address these, we identify a novel "bright ending" (BE) anomaly in diffusion models prone to memorizing training images. BE refers to a distinct cross-attention pattern observed in text-to-image diffusion models, where memorized image patches exhibit significantly greater attention to the final text token during the last inference step than non-memorized patches. This pattern highlights regions where the generated image replicates training data and enables efficient localization of memorized regions. Equipped with this, we propose a simple yet effective method to integrate BE into existing frameworks, significantly improving their performance by narrowing the performance gap caused by local memorization. Our results not only validate the successful execution of the new localization task but also establish new state-of-the-art performance across all existing tasks, underscoring the significance of the BE phenomenon.

Existing self-supervised adversarial training (self-AT) methods rely on hand-crafted adversarial attack strategies for PGD attacks, which fail to adapt to the evolving learning dynamics of the model and do not account for instance-specific characteristics of images. This results in sub-optimal adversarial robustness and limits the alignment between clean and adversarial data distributions. To address this, we propose $\textit{ASTrA}$ ($\textbf{A}$dversarial $\textbf{S}$elf-supervised $\textbf{Tr}$aining with $\textbf{A}$daptive-Attacks), a novel framework introducing a learnable, self-supervised attack strategy network that autonomously discovers optimal attack parameters through exploration-exploitation in a single training episode. ASTrA leverages a reward mechanism based on contrastive loss, optimized with REINFORCE, enabling adaptive attack strategies without labeled data or additional hyperparameters. We further introduce a mixed contrastive objective to align the distribution of clean and adversarial examples in representation space. ASTrA achieves state-of-the-art results on CIFAR10, CIFAR100, and STL10 while integrating seamlessly as a plug-and-play module for other self-AT methods. ASTrA shows scalability to larger datasets, demonstrates strong semi-supervised performance, and is resilient to robust overfitting, backed by explainability analysis on optimal attack strategies. Project page for source code and other details at https://prakashchhipa.github.io/projects/ASTrA.

Bias in machine learning models can lead to unfair decision making, and while it has been well-studied in the image and text domains, it remains underexplored in action recognition. Action recognition models often suffer from background bias (i.e., inferring actions based on background cues) and foreground bias (i.e., relying on subject appearance), which can be detrimental to real-life applications such as autonomous vehicles or assisted living monitoring. While prior approaches have mainly focused on mitigating background bias using specialized augmentations, we thoroughly study both foreground and background bias. We propose ALBAR, a novel adversarial training method that mitigates foreground and background biases without requiring specialized knowledge of the bias attributes. Our framework applies an adversarial cross-entropy loss to the sampled static clip (where all the frames are the same) and aims to make its class probabilities uniform using a proposed entropy maximization loss. Additionally, we introduce a gradient penalty loss for regularization against the debiasing process. We evaluate our method on established background and foreground bias protocols, setting a new state-of-the-art and strongly improving combined debiasing performance by over 12% absolute on HMDB51. Furthermore, we identify an issue of background leakage in the existing UCF101 protocol for bias evaluation which provides a shortcut to predict actions and does not provide an accurate measure of the debiasing capability of a model. We address this issue by proposing more fine-grained segmentation boundaries for the actor, where our method also outperforms existing approaches.

In real-life scenarios, humans seek out objects in the 3D world to fulfill their daily needs or intentions. This inspires us to introduce 3D intention grounding, a new task in 3D object detection employing RGB-D, based on human intention, such as "I want something to support my back." Closely related, 3D visual grounding focuses on understanding human reference. To achieve detection based on human intention, it relies on humans to observe the scene, reason out the target that aligns with their intention ("pillow" in this case), and finally provide a reference to the AI system, such as "A pillow on the couch". Instead, 3D intention grounding challenges AI agents to automatically observe, reason and detect the desired target solely based on human intention. To tackle this challenge, we introduce the new Intent3D dataset, consisting of 44,990 intention texts associated with 209 fine-grained classes from 1,042 scenes of the ScanNet dataset. We also establish several baselines based on different language-based 3D object detection models on our benchmark. Finally, we propose IntentNet, our unique approach, designed to tackle this intention-based detection problem. It focuses on three key aspects: intention understanding, reasoning to identify object candidates, and cascaded adaptive learning that leverages the intrinsic priority logic of different losses for multiple objective optimization.

Dataset distillation has emerged as an effective strategy, significantly reducing training costs and facilitating more efficient model deployment. Recent advances have leveraged generative models to distill datasets by capturing the underlying data distribution. Unfortunately, existing methods require model fine-tuning with distillation losses to encourage diversity and representativeness. However, these methods do not guarantee sample diversity, limiting their performance. We propose a mode-guided diffusion model leveraging a pre-trained diffusion model without the need to fine-tune with distillation losses. Our approach addresses dataset diversity in three stages: Mode Discovery to identify distinct data modes, Mode Guidance to enhance intra-class diversity, and Stop Guidance to mitigate artifacts in synthetic samples that affect performance. We evaluate our approach on ImageNette, ImageIDC, ImageNet-100, and ImageNet-1K, achieving accuracy improvements of 4.4%, 2.9%, 1.6%, and 1.6%, respectively, over state-of-the-art methods. Our method eliminates the need for fine-tuning diffusion models with distillation losses, significantly reducing computational costs.

Understanding and explaining the behavior of machine learning models is essential for building transparent and trustworthy AI systems. We introduce DEXTER, a data-free framework that employs diffusion models and large language models to generate global, textual explanations of visual classifiers. DEXTER operates by optimizing text prompts to synthesize class-conditional images that strongly activate a target classifier. These synthetic samples are then used to elicit detailed natural language reports that describe class-specific decision patterns and biases. Unlike prior work, DEXTER enables natural language explanation about a classifier's decision process without access to training data or ground-truth labels. We demonstrate DEXTER's flexibility across three tasks—activation maximization, slice discovery and debiasing, and bias explanation—each illustrating its ability to uncover the internal mechanisms of visual classifiers. Quantitative and qualitative evaluations, including a user study, show that DEXTER produces accurate, interpretable outputs. Experiments on ImageNet, Waterbirds, CelebA, and FairFaces confirm that DEXTER outperforms existing approaches in global model explanation and class-level bias reporting. Code is available at https://github.com/perceivelab/dexter.

Composed Video Retrieval (CoVR) retrieves a target video given a query video and a modification text describing the intended change. Existing CoVR benchmarks emphasize appearance shifts or coarse event changes and therefore do not test the ability to capture subtle, fast-paced temporal differences. We introduce TF-CoVR, the first large-scale benchmark dedicated to temporally fine-grained CoVR. TF-CoVR focuses on gymnastics and diving, and provides 180K triplets drawn from FineGym and FineDiving datasets. Previous CoVR benchmarks, focusing on temporal aspect, link each query to a single target segment taken from the same video, limiting practical usefulness. In TF-CoVR, we instead construct each pair by prompting an LLM with the label differences between clips drawn from different videos; every pair is thus associated with multiple valid target videos (3.9 on average), reflecting real-world tasks such as sports-highlight generation. To model these temporal dynamics, we propose TF-CoVR-Base, a concise two-stage training framework: (i) pre-train a video encoder on fine-grained action classification to obtain temporally discriminative embeddings; (ii) align the composed query with candidate videos using contrastive learning. We conduct the first comprehensive study of image, video, and general multimodal embedding (GME) models on temporally fine-grained composed retrieval in both zero-shot and fine-tuning regimes. On TF-CoVR, TF-CoVR-Base improves zero-shot mAP@50 from 5.92 (LanguageBind) to 7.51, and after fine-tuning raises the state-of-the-art from 19.83 to 27.22.

With the recent exhibited strength of generative diffusion models, an open research question is if images generated by these models can be used to learn better visual representations. While this generative data expansion may suffice for easier visual tasks, we explore its efficacy on a more difficult discriminative task: clothes-changing person re-identification (CC-ReID). CC-ReID aims to match people appearing in non-overlapping cameras, even when they change their clothes across cameras. Not only that current CC-ReID models are constrained by the limited diversity of clothing in current CC-ReID datasets, but generating additional data that retains important personal features for accurate identification is a current challenge. To address this issue we propose DLCR, a novel data expansion frame-work that leverages pretrained diffusion and large language models (LLMs) to accurately generate diverse images of in-dividuals in varied attire. We generate additional data for five benchmark CC-ReID datasets (PRCC, CCVID, LaST, VC-Clothes, and LTCC) and increase their clothing diversity by 10 x, totaling over 2.1M generated images. DLCR employs diffusion-based text-guided inpainting, conditioned on clothing prompts constructed using LLMs, to generate synthetic data that only modifies a subject's clothes, while preserving their personally identifiable features. With this massive increase in data, we introduce two novel strategies - progressive learning and test-time prediction refinement - that reduce training time and boost CC-ReID performance. We validate our method through extensive ablations and experiments, showing massive improvements when training previous CC-ReID methods on our generated data. On the PRCC dataset, we obtain a large top-1 accuracy improvement of 11.3% by training CAL, a state-of-the-art (SOTA) method, with DLCR-generated data. We publicly release our code and generated data for each dataset here: https://github.com/CroitoruAlin/dlcr.

We present DiffExplainer , a novel framework that, leveraging language-vision models, enables multimodal global explainability. DiffExplainer employs diffusion models conditioned on optimized text prompts, synthesizing images that maximize class outputs and hidden features of a classifier, thus providing a visual tool for explaining decisions. Moreover, the analysis of generated visual descriptions allows for automatic identification of biases and spurious features, as opposed to traditional methods that often rely on manual intervention. The cross-modal transferability of language-vision models also enables the possibility to describe decisions in a more human-interpretable way, i.e., through text. We conduct comprehensive experiments demonstrating the effectiveness of DiffExplainer on (1) the generation of high-quality images explaining model decisions, surpassing existing activation maximization methods, and (2) the automated identification of biases and spurious features. • Leverages diffusion models to generate images explaining classifier decisions. • Enables bias and spurious feature detection without manual intervention. • Outperforms activation maximization methods in image quality and feature analysis. • Enables specific model analysis by the use of fixed prompts.

Vision systems that see and reason about the compositional nature of visual scenes are fundamental to understanding our world. The complex relations between objects and their locations, ambiguities, and variations in the real-world environment can be better described in human language, naturally governed by grammatical rules and other modalities such as audio and depth. The models learned to bridge the gap between such modalities and large-scale training data facilitate contextual reasoning, generalization, and prompt capabilities at test time. These models are referred to as foundation models. The output of such models can be modified through human-provided prompts without retraining, e.g., segmenting a particular object by providing a bounding box, having interactive dialogues by asking questions about an image or video scene or manipulating the robot's behavior through language instructions. In this survey, we provide a comprehensive review of such emerging foundation models, including typical architecture designs to combine different modalities (vision, text, audio, etc.), training objectives (contrastive, generative), pre-training datasets, fine-tuning mechanisms, and the common prompting patterns; textual, visual, and heterogeneous. We discuss the open challenges and research directions for foundation models in computer vision, including difficulties in their evaluations and benchmarking, gaps in their real-world understanding, limitations of contextual understanding, biases, vulnerability to adversarial attacks, and interpretability issues. We review recent developments in this field, covering a wide range of applications of foundation models systematically and comprehensively.

Understanding long-form videos requires precise temporal action segmentation. While existing studies typically employ multi-stage models that follow an iterative refinement process, we present a novel framework based on the denoising diffusion model that retains this core iterative principle. Within this framework, the model iteratively produces action predictions starting with random noise, conditioned on the features of the input video. To effectively capture three key characteristics of human actions, namely the position prior, the boundary ambiguity, and the relational dependency, we propose a cohesive masking strategy for the conditioning features. Moreover, a consistency gradient guidance technique is proposed, which maximizes the similarity between outputs with or without the masking, thereby enriching conditional information during the inference process. Extensive experiments are performed on four datasets, i.e., GTEA, 50Salads, Breakfast, and Assembly101. The results indicate that our proposed method outperforms or is on par with existing state-of-the-art techniques, underscoring the potential of generative approaches for action segmentation.

Self-supervised approaches for video have shown impressive results in video understanding tasks. However, unlike early works that leverage temporal self-supervision, current state-of-the-art methods primarily rely on tasks from the image domain (e.g., contrastive learning) that do not explicitly promote the learning of temporal features. We identify two factors that limit existing temporal self-supervision: 1) tasks are too simple, resulting in saturated training performance, and 2) we uncover shortcuts based on local appearance statistics that hinder the learning of high-level features. To address these issues, we propose 1) a more challenging reformulation of temporal self-supervision as frame-level (rather than clip-level) recognition tasks and 2) an effective augmentation strategy to mitigate shortcuts. Our model extends a representation of single video frames, pre-trained through contrastive learning, with a transformer that we train through temporal self-supervision. We demonstrate experimentally that our more challenging frame-level task formulations and the removal of shortcuts drastically improve the quality of features learned through temporal self-supervision. Our extensive experiments show state-of-the-art performance across 10 video understanding datasets, illustrating the generalization ability and robustness of our learned video representations. Project Page: https://daveishan.github.io/nms-webpage.

In this work, we present a novel approach to multi-view action recognition where we guide learned action representations to be separated from view-relevant information in a video. When trying to classify action instances captured from multiple viewpoints, there is a higher degree of difficulty due to the difference in background, occlusion, and visibility of the captured action from different camera angles. To tackle the various problems introduced in multi-view action recognition, we propose a novel configuration of learnable transformer decoder queries, in conjunction with two supervised contrastive losses, to enforce the learning of action features that are robust to shifts in viewpoints. Our disentangled feature learning occurs in two stages: the transformer decoder uses separate queries to separately learn action and view information, which are then further disentangled using our two contrastive losses. We show that our model and method of training significantly outperforms all other uni-modal models on four multi-view action recognition datasets: NTU RGB+D, NTU RGB+D 120, PKU-MMD, and N-UCLA. Compared to previous RGB works, we see maximal improvements of 1.5%, 4.8%, 2.2%, and 4.8% on each dataset, respectively. Our code can be found here: https://github.com/NyleSiddiqui/MultiView_Actions

Despite the commercial abundance of UAVs, aerial data acquisition remains challenging, and the existing Asia and North America-centric open-source UAV datasets are small-scale or low-resolution and lack diversity in scene contextuality. Additionally, the color content of the scenes, solar zenith angle, and population density of different geographies influence the data diversity. These factors conjointly render suboptimal aerial-visual perception of the deep neural network (DNN) models trained primarily on the ground view data, including the open-world foundational models. To pave the way for a transformative era of aerial detection, we present Multiview Aerial Visual RECognition or MAVREC, a video dataset where we record synchronized scenes from different perspectives - ground camera and drone-mounted camera. MAVREC consists of around 2.5 hours of industry-standard 2.7K resolution video sequences, more than 0.5 million frames, and 1.1 million annotated bounding boxes. This makes MAVREC the largest ground and aerial view dataset, and the fourth largest among all drone-based datasets across all modalities and tasks. Through our extensive benchmarking on MAVREC, we recognize that augmenting object detectors with ground view images from the corresponding geographical location is a superior pretraining strategy for aerial detection. Building on this strategy, we benchmark MAVREC with a curriculum-based semi-supervised object detection approach that leverages labeled (ground and aerial) and unlabeled (only aerial) images to enhance aerial detection.

We propose an efficient abnormal event detection model based on a lightweight masked auto-encoder (AE) applied at the video frame level. The novelty of the proposed model is threefold. First, we introduce an approach to weight tokens based on motion gradients, thus shifting the focus from the static background scene to the foreground objects. Second, we integrate a teacher decoder and a student decoder into our architecture, leveraging the discrepancy between the outputs given by the two decoders to improve anomaly detection. Third, we generate synthetic abnormal events to augment the training videos, and task the masked AE model to jointly reconstruct the original frames (without anomalies) and the corresponding pixel-level anomaly maps. Our design leads to an efficient and effective model, as demonstrated by the extensive experiments carried out on four benchmarks: Avenue, Shanghai Tech, UBnormal and UCSD Ped2. The empirical results show that our model achieves an excellent trade-off between speed and accuracy, obtaining competitive AUC scores, while processing 1655 FPS. Hence, our model is between 8 and 70 times faster than competing methods. We also conduct an ablation study to justify our design. Our code is freely available at: https://github.com/ristea/aed-mae.

In this paper, we propose VidLA, an approach for video-language alignment at scale. There are two major limitations of previous video-language alignment approaches. First, they do not capture both short-range and long-range temporal dependencies and typically employ complex hierarchical deep network architectures that are hard to integrate with existing pretrained image-text foundation models. To effectively address this limitation, we instead keep the network architecture simple and use a set of data tokens that operate at different temporal resolutions in a hierarchical manner, accounting for the temporally hierarchical nature of videos. By employing a simple two-tower architecture, we are able to initialize our video-language model with pretrained image-text foundation models, thereby boosting the final performance. Second, existing video-language alignment works struggle due to the lack of semantically aligned large-scale training data. To overcome it, we leverage recent LLMs to curate the largest video-language dataset to date with better visual grounding. Furthermore, unlike existing video-text datasets which only contain short clips, our dataset is enriched with video clips of varying durations to aid our temporally hierarchical data to-kens in extracting better representations at varying temporal scales. Overall, empirical results show that our proposed approach surpasses state-of-the-art methods on Multiple retrieval benchmarks, especially on longer videos, and performs competitively on classification benchmarks.

Composed video retrieval (CoVR) is a challenging problem in computer vision which has recently highlighted the integration of modification text with visual queries for more sophisticated video search in large databases. Existing works predominantly rely on visual queries combined with modification text to distinguish relevant videos. However, such a strategy struggles to fully preserve the rich query-specific context in retrieved target videos and only represents the target video using visual embedding. We introduce a novel CoVR framework that leverages detailed language descriptions to explicitly encode query-specific contextual information and learns discriminative embeddings of vision only, text only and vision-text for better alignment to accurately retrieve matched target videos. Our proposed framework can be flexibly employed for both composed video (CoVR) and image (CoIR) retrieval tasks. Experiments on three datasets show that our approach obtains state-of-the-art performance for both CovR and zero-shot CoIR tasks, achieving gains as high as around 7% in terms of recall@ K=1 score. Our code, detailed language descriptions for Web ViD-Co VR dataset are available at https://github.com/OmkarThawakar/composed-video-retrieval.

Malaria is a major health issue worldwide, and its diagnosis requires scalable solutions that can work effectively with low-cost microscopes (LCM). Deep learning-based methods have shown success in computer-aided diagnosis from microscopic images. However, these methods need annotated images that show cells affected by malaria parasites and their life stages. Annotating images from LCM significantly increases the burden on medical experts compared to annotating images from high-cost microscopes (HCM). For this reason, a practical solution would be trained on HCM images which should generalize well on LCM images during testing. While earlier methods adopted a multi-stage learning process, they did not offer an end-to-end approach. In this work, we present an end-to-end learning framework, named CodaMal (COntrastive Domain Adpation for MALaria). In order to bridge the gap between HCM (training) and LCM (testing), we propose a domain adaptive contrastive loss. It reduces the domain shift by promoting similarity between the representations of HCM and its corresponding LCM image, without imposing an additional annotation burden. In addition, the training objective includes object detection objectives with carefully designed augmentations, ensuring the accurate detection of malaria parasites. On the publicly available large-scale M5-dataset, our proposed method shows a significant improvement of $\mathbf{1 6 \%}$ over the state-of-the-art methods in terms of the mean average precision metric (mAP), provides $21 \times$ speed improvement during inference and requires only half of the learnable parameters used in prior methods. Our code is publicly available: https://daveishan.github.io/codamal-webpage/.

3D detection is a critical task that enables machines to identify and locate objects in three-dimensional space. It has a broad range of applications in several fields, including autonomous driving, robotics and augmented reality. Monocular 3D detection is attractive as it requires only a single camera, however, it lacks the accuracy and robustness required for real world applications. High resolution LiDAR on the other hand, can be expensive and lead to interference problems in heavy traffic given their active transmissions. We propose a balanced approach that combines the advantages of monocular and point cloud-based 3D detection. Our method requires only a small number of 3D points, that can be obtained from a low-cost, low-resolution sensor. Specifically, we use only 512 points, which is just 1% of a full LiDAR frame in the KITTI dataset. Our method reconstructs a complete 3D point cloud from this limited 3D information combined with a single image. The reconstructed 3D point cloud and corresponding image can be used by any multi-modal off-the-shelf detector for 3D object detection. By using the proposed network architecture with an off-the-shelf multi-modal 3D detector, the accuracy of 3D detection improves by 20% compared to the state-of-theart monocular detection methods and 6% to 9% compare to the baseline multi-modal methods on KITTI and JackRabbot datasets.

The application of Automatic Speech Recognition (ASR) technology in soccer enables sports analytics by extracting audio commentaries to provide insights into game events and facilitate automatic game understanding. This paper presents SoccerNet-Echoes, an extension of the SoccerNet dataset with automatically generated transcriptions of soccer game broadcasts. Generated using the Whisper model and translated with Google Translate into English when needed, these transcriptions enhance video content with textual information derived from game audio. SoccerNet-Echoes serves as a comprehensive resource for developing algorithms in action spotting, caption generation, and game summarization. Through a series of experiments, we demonstrate that combining modalities—audio, video, and text—yields mixed results on classification tasks. The combination of audio and video shows improved performance over individual modalities, while the addition of ASR text does not significantly enhance results. Additionally, our baseline summarization tasks indicate that ASR content enriches summaries, offering insights beyond event information. This multimodal dataset supports diverse applications, broadening the scope of research in sports analytics. The dataset is available at: https://github.com/SoccerNet/sn-echoes.

The recent introduction of Diffusion Transformers (DiTs) has demonstrated exceptional capabilities in image generation by using a different backbone architecture, departing from traditional U-Nets and embracing the scalable nature of transformers. Despite their advanced capabilities, the wide deployment of DiTs, particularly for real-time applications, is currently hampered by considerable computational demands at the inference stage. Post-training Quantization (PTQ) has emerged as a fast and data-efficient solution that can significantly reduce computation and memory footprint by using low-bit weights and activations. However, its applicability to DiTs has not yet been explored and faces non-trivial difficulties due to the unique design of DiTs. In this paper, we propose PTQ4DiT, a specifically designed PTQ method for DiTs. We discover two primary quantization challenges inherent in DiTs, notably the presence of salient channels with extreme magnitudes and the temporal variability in distributions of salient activation over multiple timesteps. To tackle these challenges, we propose Channel-wise Salience Balancing (CSB) and Spearmen's $\rho$-guided Salience Calibration (SSC). CSB leverages the complementarity property of channel magnitudes to redistribute the extremes, alleviating quantization errors for both activations and weights. SSC extends this approach by dynamically adjusting the balanced salience to capture the temporal variations in activation. Additionally, to eliminate extra computational costs caused by PTQ4DiT during inference, we design an offline re-parameterization strategy for DiTs. Experiments demonstrate that our PTQ4DiT successfully quantizes DiTs to 8-bit precision (W8A8) while preserving comparable generation ability and further enables effective quantization to 4-bit weight precision (W4A8) for the first time.

Text-to-image diffusion models have recently attracted the interest of many researchers, and inverting the diffusion process can play an important role in better understanding the generative process and how to engineer prompts in order to obtain the desired images. To this end, we study the task of predicting the prompt embedding given an image generated by a generative diffusion model. We consider a series of white-box and black-box models (with and without access to the weights of the diffusion network) to deal with the proposed task. We propose a novel learning framework comprising a joint prompt regression and multi-label vocabulary classification objective that generates improved prompts. To further improve our method, we employ a curriculum learning procedure that promotes the learning of image-prompt pairs with lower labeling noise ( i . e . that are better aligned). We conduct experiments on the DiffusionDB data set, predicting text prompts from images generated by Stable Diffusion. In addition, we make an interesting discovery: training a diffusion model on the prompt generation task can make the model generate images that are much better aligned with the input prompts, when the model is directly reused for text-to-image generation. Our code is publicly available for download at https://github.com/CroitoruAlin/Reverse-Stable-Diffusion . • We study the task of reversing the text-to-image generation process of diffusion models. • We propose a training pipeline comprising three novel components. • We demonstrate the usefulness of our pipeline on four underlying models. • We showcase a promising application of prompt generation models in text-to-image generation.

We propose a very fast frame-level model for anomaly detection in video, which learns to detect anomalies by distilling knowledge from multiple highly accurate object-level teacher models. To improve the fidelity of our student, we distill the low-resolution anomaly maps of the teachers by jointly applying standard and adversarial distillation, introducing an adversarial discriminator for each teacher to distinguish between target and generated anomaly maps. We conduct experiments on three benchmarks (Avenue, ShanghaiTech, UCSD Ped2), showing that our method is over 7 times faster than the fastest competing method, and between 28 and 62 times faster than object-centric models, while obtaining comparable results to recent methods. Our evaluation also indicates that our model achieves the best trade-off between speed and accuracy, due to its previously unheard-of speed of 1480 FPS. In addition, we carry out a comprehensive ablation study to justify our architectural design choices. Our code is freely available at: https://github.com/ristea/fast-aed.

We propose a novel self-supervised Video Object Segmentation (VOS) approach that strives to achieve better visual correspondence across frames and object-background discriminability for accurate object segmentation. Distinct from previous self-supervised VOS methods, our approach is based on a learning loss formulation that takes into account both object and background information to ensure object-background discriminability, rather than using only object appearance. The objective function comprises cutout-based reconstruction (cutout region represents part of a frame, whose pixels are replaced with some constant values) and tag prediction loss terms. The cutout-based reconstruction term utilizes a simple cutout scheme to learn the pixel-wise correspondence between the current and previous frames in order to reconstruct the original current frame with added cutout region in it. The introduced cutout patch guides the model to focus on the reappearance of scene parts , thereby implicitly equipping the model to address occlusion-based scenarios Next, the tag prediction term encourages object-background separability by grouping tags of all pixels in the cutout region that are similar, while separating them from the tags of the rest of the reconstructed pixels. Additionally, we introduce a zoom-in scheme that addresses the problem of small object segmentation by capturing fine structural information at multiple scales. Our proposed approach, termed CT-VOS, achieves state-of-the-art results on two challenging benchmarks: DAVIS-2017 and Youtube-VOS. A detailed ablation showcases the importance of the proposed loss formulation to effectively establish correspondences, object-background discriminability, and the impact of our zoom-in scheme to accurately segment small-sized objects.

Anomaly detection has recently gained increasing attention in the field of computer vision, likely due to its broad set of applications ranging from product fault detection on industrial production lines and impending event detection in video surveillance to finding lesions in medical scans. Regardless of the domain, anomaly detection is typically framed as a one-class classification task, where the learning is conducted on normal examples only. An entire family of successful anomaly detection methods is based on learning to reconstruct masked normal inputs (e.g. patches, future frames, etc.) and exerting the magnitude of the reconstruction error as an indicator for the abnormality level. Unlike other reconstruction-based methods, we present a novel self-supervised masked convolutional transformer block (SSMCTB) that comprises the reconstruction-based functionality at a core architectural level. The proposed self-supervised block is extremely flexible, enabling information masking at any layer of a neural network and being compatible with a wide range of neural architectures. In this work, we extend our previous self-supervised predictive convolutional attentive block (SSPCAB) with a 3D masked convolutional layer, a transformer for channel-wise attention, as well as a novel self-supervised objective based on Huber loss. Furthermore, we show that our block is applicable to a wider variety of tasks, adding anomaly detection in medical images and thermal videos to the previously considered tasks based on RGB images and surveillance videos. We exhibit the generality and flexibility of SSMCTB by integrating it into multiple state-of-the-art neural models for anomaly detection, bringing forth empirical results that confirm considerable performance improvements on five benchmarks: MVTec AD, BRATS, Avenue, ShanghaiTech, and Thermal Rare Event.

Bharadwaj et al. (2023) present a comments paper evaluating the classification accuracy of several state-of-the-art methods using EEG data averaged over random class samples. According to the results, some of the methods achieve above-chance accuracy, while the method proposed in (Palazzo et al. 2020), that is the target of their analysis, does not. In this rebuttal, we address these claims and explain why they are not grounded in the cognitive neuroscience literature, and why the evaluation procedure is ineffective and unfair.

Referring Video Object Segmentation (R-VOS) methods face challenges in maintaining consistent object segmentation due to temporal context variability and the presence of other visually similar objects. We propose an end-to-end R-VOS paradigm that explicitly models temporal instance consistency alongside the referring segmentation. Specifically, we introduce a novel hybrid memory that facilitates inter-frame collaboration for robust spatio-temporal matching and propagation. Features of frames with automatically generated high-quality reference masks are propagated to segment the remaining frames based on multi-granularity association to achieve temporally consistent R-VOS. Furthermore, we propose a new Mask Consistency Score (MCS) metric to evaluate the temporal consistency of video segmentation. Extensive experiments demonstrate that our approach enhances temporal consistency by a significant margin, leading to top-ranked performance on popular R-VOS benchmarks, i.e., Ref-YouTube-VOS (67.1%) and Ref-DAVIS17 (65.6%). The code is available athttps://github.com/bo-miao/HTR.

Deep learning models are susceptible to contrived adversarial examples, even in the decision-based black-box setting where the attacker has access to the model’s decisions only. Developing more efficient and practical attacks help in better understanding the limitations of deep models. It is important that attacks are crafted with limited queries to avoid suspicion. Since the required number of queries increase with dimensions, low-dimensional embeddings are attractive. This low query budget constraint is a bottleneck for learning-based and data-driven attacks which rely heavily on querying the model. We propose LSDAT, an image-agnostic non-data-driven decision-based black-box attack that exploits low-rank and sparse decomposition (LSD) of images to dramatically reduce the queries and improve fooling rates compared to existing methods. LSDAT crafts perturbations in the low-dimensional subspace formed by the sparse component of the input image and that of a target adversarial image to obtain query-efficiency. A viable perturbation is obtained by traversing the path between the input and adversarial sparse components. Theoretical analyses are provided to justify the functionality of LSDAT. Unlike other competitors (e.g., FFT), LSD works directly in the image domain to guarantee that non-$\ell _{2}$constraints, such as sparsity, are satisfied. LSDAT offers better control over the number of queries and is computationally efficient as it performs sparse decomposition of the input and adversarial images only once to generate all queries. Four variants of LSDAT are presented for different scenarios including a pure black-box attack where no queries are allowed. We demonstrate$\ell _{0}$,$\ell _{2}$and$\ell _{\infty} $bounded attacks with LSDAT to evince its efficiency compared to baseline attacks in diverse low-query budget scenarios. LSDAT obtains 15 to 20% improvement in fooling ResNet-50 while using far fewer queries than competing methods in a similar setting.

Recently attention-based networks have been successful for image restoration tasks. However, existing methods are either computationally expensive or have limited receptive fields, adding constraints to the model. They are also less resilient in spatial and contextual aspects and lack pixel-to-pixel correspondence, which may degrade feature representations. In this paper, we propose a novel and computationally efficient architecture Single Stage Adaptive Multi-Attention Network (SSAMAN) for image restoration tasks, particularly for image denoising and image deblurring. SSAMAN efficiently addresses computational challenges and expands receptive fields, enhancing robustness in spatial and contextual feature representation. Its Adaptive Multi-Attention Module (AMAM), which consists of Adaptive Pixel Attention Branch (APAB) and an Adaptive Channel Attention Branch (ACAB), uniquely integrates channel and pixel-wise dimensions, significantly improving sensitivity to edges, shapes, and textures. We perform extensive experiments and ablation studies to validate the performance of SSAMAN. Our model shows state-of-the-art results on various benchmarks, for example, on image denoising tasks, SSAMAN achieves a notable 40.08 dB PSNR on SIDD dataset, outperforming Restormer by 0.06 dB PSNR, with 41.02% less computational cost, and achieves a 40.05 dB PSNR on the DND dataset. For image deblurring, SSAMAN achieves 33.53 dB PSNR on GoPro dataset. Code and models are available at Github.

Geometric feature learning for 3-D surfaces is critical for many applications in computer graphics and 3-D vision. However, deep learning currently lags in hierarchical modeling of 3-D surfaces due to the lack of required operations and/or their efficient implementations. In this article, we propose a series of modular operations for effective geometric feature learning from 3-D triangle meshes. These operations include novel mesh convolutions, efficient mesh decimation, and associated mesh (un)poolings. Our mesh convolutions exploit spherical harmonics as orthonormal bases to create continuous convolutional filters. The mesh decimation module is graphics processing unit (GPU)-accelerated and able to process batched meshes on-the-fly, while the (un)pooling operations compute features for upsampled/downsampled meshes. We provide an open-source implementation of these operations, collectively termed Picasso. Picasso supports heterogeneous mesh batching and processing. Leveraging its modular operations, we further contribute a novel hierarchical neural network for perceptual parsing of 3-D surfaces, named PicassoNet++. It achieves highly competitive performance for shape analysis and scene segmentation on prominent 3-D benchmarks. The code, data, and trained models are available at https://github.com/EnyaHermite/Picasso.

Contrastive self-supervised learning (CSL) has managed to match or surpass the performance of supervised learning in image and video classification. However, it is still largely unknown if the nature of the representations induced by the two learning paradigms is similar. We investigate this under the lens of adversarial robustness. Our analysis of the problem reveals that CSL has intrinsically higher sensitivity to perturbations over supervised learning. We identify the uniform distribution of data representation over a unit hypersphere in the CSL representation space as the key contributor to this phenomenon. We establish that this is a result of the presence of false negative pairs in the training process, which increases model sensitivity to input perturbations. Our finding is supported by extensive experiments for image and video classification using adversarial perturbations and other input corruptions. We devise a strategy to detect and remove false negative pairs that is simple, yet effective in improving model robustness with CSL training. We close up to 68% of the robustness gap between CSL and its supervised counterpart. Finally, we contribute to adversarial learning by incorporating our method in CSL. We demonstrate an average gain of about 5% over two different state-of-the-art methods in this domain.

Clustered federated learning (FL) has been shown to produce promising results by grouping clients into clusters. This is especially effective in scenarios where separate groups of clients have significant differences in the distributions of their local data. Existing clustered FL algorithms are essentially trying to group together clients with similar distributions so that clients in the same cluster can leverage each other's data to better perform federated learning. However, prior clustered FL algorithms attempt to learn these distribution similarities indirectly during training, which can be quite time consuming as many rounds of federated learning may be required until the formation of clusters is stabilized. In this paper, we propose a new approach to federated learning that directly aims to efficiently identify distribution similarities among clients by analyzing the principal angles between the client data subspaces. Each client applies a truncated singular value decomposition (SVD) step on its local data in a single-shot manner to derive a small set of principal vectors, which provides a signature that succinctly captures the main characteristics of the underlying distribution. This small set of principal vectors is provided to the server so that the server can directly identify distribution similarities among the clients to form clusters. This is achieved by comparing the similarities of the principal angles between the client data subspaces spanned by those principal vectors. The approach provides a simple, yet effective clustered FL framework that addresses a broad range of data heterogeneity issues beyond simpler forms of Non-IIDness like label skews. Our clustered FL approach also enables convergence guarantees for non-convex objectives.

The recent growth in the consumption of online media by children during early childhood necessitates data-driven tools enabling educators to filter out appropriate educational content for young learners. This paper presents an approach for detecting educational content in online videos. We focus on two widely used educational content classes: literacy and math. For each class, we choose prominent codes (sub-classes) based on the Common Core Standards. For example, literacy codes include ‘letter names’, ‘letter sounds’, and math codes include ‘counting’, ‘sorting’. We pose this as a finegrained multilabel classification problem as videos can contain multiple types of educational content and the content classes can get visually similar (e.g., ‘letter names’vs ‘letter sounds’). We propose a novel class prototypes based supervised contrastive learning approach that can handle fine-grained samples associated with multiple labels. We learn a class prototype for each class and a loss function is employed to minimize the distances between a class prototype and the samples from the class. Similarly, distances between a class prototype and the samples from other classes are maximized. As the alignment between visual and audio cues are crucial for effective comprehension, we consider a multimodal transformer network to capture the interaction between visual and audio cues in videos while learning the embedding for videos. For evaluation, we present a dataset, APPROVE, employing educational videos from YouTube labeled with fine-grained education classes by education researchers. APPROVE consists of 193 hours of expert-annotated videos with 19 classes. The proposed approach outperforms strong baselines on APPROVE and other benchmarks such as Youtube-8M, and COIN. The dataset is available at https://nusci.csl.sri.com/project/APPROVE.

The pose-guided person image generation task requires synthesizing photorealistic images of humans in arbitrary poses. The existing approaches use generative adversarial networks that do not necessarily maintain realistic textures or need dense correspondences that struggle to handle complex deformations and severe occlusions. In this work, we show how denoising diffusion models can be applied for high-fidelity person image synthesis with strong sample diversity and enhanced mode coverage of the learnt data distribution. Our proposed Person Image Diffusion Model (PIDM) disintegrates the complex transfer problem into a series of simpler forward-backward denoising steps. This helps in learning plausible source-to-target transformation trajectories that result in faithful textures and undistorted appearance details. We introduce a ‘texture diffusion module’ based on cross-attention to accurately model the correspondences between appearance and pose information available in source and target images. Further, we propose ‘disentangled classifier-free guidance’ to ensure close resemblance between the conditional inputs and the synthesized output in terms of both pose and appearance information. Our extensive results on two large-scale benchmarks and a user study demonstrate the photorealism of our proposed approach under challenging scenarios. We also show how our generated images can help in downstream tasks. Code is available at https://github.com/ankanbhunia/PIDM.

Determining the exact latitude and longitude that a photo was taken is a useful and widely applicable task, yet it remains exceptionally difficult despite the accelerated progress of other computer vision tasks. Most previous approaches have opted to learn single representations of query images, which are then classified at different levels of geographic granularity. These approaches fail to exploit the different visual cues that give context to different hierarchies, such as the country, state, and city level. To this end, we introduce an end-to-end transformer-based architecture that exploits the relationship between different geographic levels (which we refer to as hierarchies) and the corresponding visual scene information in an image through hierarchical cross-attention. We achieve this by learning a query for each geographic hierarchy and scene type. Furthermore, we learn a separate representation for different environmental scenes, as different scenes in the same location are often defined by completely different visual features. We achieve state of the art accuracy on 4 standard geo-localization datasets : Im2GPS, Im2GPS3k, YFCC4k, and YFCC26k, as well as qualitatively demonstrate how our method learns different representations for different visual hierarchies and scenes, which has not been demonstrated in the previous methods. Above previous testing datasets mostly consist of iconic landmarks or images taken from social media, which makes the dataset a simple memory task, or makes it biased towards certain places. To address this issue we introduce a much harder testing dataset, Google-World-Streets-15k, comprised of images taken from Google Streetview covering the whole planet and present state of the art results. Our code can be found at https://github.com/AHKerrigan/GeoGuessNet.

Semi-Supervised Learning can be more beneficial for the video domain compared to images because of its higher an-notation cost and dimensionality. Besides, any video understanding task requires reasoning over both spatial and temporal dimensions. In order to learn both the static and motion related features for the semi-supervised action recognition task, existing methods rely on hard in-put inductive biases like using two-modalities (RGB and Optical-flow) or two-stream of different playback rates. Instead of utilizing unlabeled videos through diverse in-put streams, we rely on self-supervised video represen-tations, particularly, we utilize temporally-invariant and temporally-distinctive representations. We observe that these representations complement each other depending on the nature of the action. Based on this observation, we propose a student-teacher semi-supervised learning frame-work, TimeBalance, where we distill the knowledge from a temporally-invariant and a temporally-distinctive teacher. Depending on the nature of the unlabeled video, we dy-namically combine the knowledge of these two teach-ers based on a novel temporal similarity-based reweighting scheme. Our method achieves state-of-the-art performance on three action recognition benchmarks: UCF101, HMDB51, and Kinetics400. Code: https://github.com/DAVEISHAN/TimeBalance.

Answering questions about complex situations in videos requires not only capturing the presence of actors, objects, and their relations but also the evolution of these relationships over time. A situation hyper-graph is a representation that describes situations as scene sub-graphs for video frames and hyper-edges for connected sub-graphs and has been proposed to capture all such information in a compact structured form. In this work, we propose an architecture for Video Question Answering (VQA) that enables answering questions related to video content by predicting situation hyper-graphs, coined Situation Hyper-Graph based Video Question Answering (SHG- VQA). To this end, we train a situation hyper-graph decoder to implicitly identify graph representations with actions and object/human-object relationships from the input video clip. and to use cross-attention between the predicted situation hyper-graphs and the question embedding to predict the correct answer. The proposed method is trained in an end-to-end manner and optimized by a VQA loss with the cross-entropy function and a Hungarian matching loss for the situation graph prediction. The effectiveness of the proposed architecture is extensively evaluated on two challenging benchmarks: AGQA and STAR. Our results show that learning the underlying situation hyper-graphs helps the system to significantly improve its performance for novel challenges of video question-answering tasks11Code will be available at https://github.com/aurooj/SHG-VQA.

Weakly-supervised Temporal Action Localization (WTAL) attempts to localize the actions in untrimmed videos using only video-level supervision. Most recent works approach WTAL from a localization-by-classification perspective where these methods try to classify each video frame followed by a manually-designed post-processing pipeline to aggregate these per-frame action predictions into action snippets. Due to this perspective, the model lacks any explicit understanding of action boundaries and tends to focus only on the most discriminative parts of the video resulting in incomplete action localization. To address this, we present PivoTAL, Prior-driven Supervision for Weakly-supervised Temporal Action Localization, to approach WTAL from a localization-by-localization perspective by learning to localize the action snippets directly. To this end, PivoTAL leverages the underlying spatio-temporal regularities in videos in the form of action-specific scene prior, action snippet generation prior, and learnable Gaussian prior to supervise the localization-based training. PivoTAL shows significant improvement (of at least 3% avg mAP) over all existing methods on the benchmark datasets, THUMOS-14 and ActivitNet-v1.3.

Adopting contrastive image-text pretrained models like CLIP towards video classification has gained attention due to its cost-effectiveness and competitive performance. However, recent works in this area face a trade-off. Finetuning the pretrained model to achieve strong supervised performance results in low zero-shot generalization. Similarly, freezing the backbone to retain zero-shot capability causes significant drop in supervised accuracy. Because of this, recent works in literature typically train separate models for supervised and zero-shot action recognition. In this work, we propose a multimodal prompt learning scheme that works to balance the supervised and zero-shot performance under a single unified training. Our prompting approach on the vision side caters for three aspects: 1) Global video-level prompts to model the data distribution; 2) Local frame-level prompts to provide per-frame discriminative conditioning; and 3) a summary prompt to extract a condensed video representation. Additionally, we define a prompting scheme on the text side to augment the textual context. Through this prompting scheme, we can achieve state-of-the-art zero-shot performance on Kinetics-600, HMDB51 and UCF101 while remaining competitive in the supervised setting. By keeping the pretrained backbone frozen, we optimize a much lower number of parameters and retain the existing general representation which helps achieve the strong zero-shot performance. Our codes/models will be released at https://github.com/TalalWasim/Vita-Clip..

Visual Place Recognition (VPR) estimates the location of query images by matching them with images in a reference database. Conventional methods generally adopt aggregated CNN features for global retrieval and RANSAC-based geometric verification for reranking. However, RANSAC only employs geometric information but ignores other possible information that could be useful for reranking, e.g. local feature correlations, and attention values. In this paper, we propose a unified place recognition framework that handles both retrieval and reranking with a novel transformer model, named$R^{2}$Former. The proposed reranking module takes feature correlation, attention value, and$xy$coordinates into account, and learns to determine whether the image pair is from the same location. The whole pipeline is end-to-end trainable and the reranking module alone can also be adopted on other CNN or transformer backbones as a generic component. Remarkably,$R^{2}$Former significantly outperforms state-of-the-art methods on major VPR datasets with much less inference time and memory consumption. It also achieves the state-of-the-art on the hold-out MSLS challenge set and could serve as a simple yet strong solution for real-world large-scale applications. Experiments also show vision transformer tokens are comparable and sometimes better than CNN local features on local matching. The code is released at https://github.com/Jeff-Zilence/R2Former.

Video anomaly detection (VAD) without human monitoring is a complex computer vision task that can have a positive impact on society if implemented successfully. While recent advances have made significant progress in solving this task, most existing approaches overlook a critical real-world concern: privacy. With the increasing popularity of artificial intelligence technologies, it becomes crucial to implement proper AI ethics into their development. Privacy leakage in VAD allows models to pick up and amplify unnecessary biases related to people’s personal information, which may lead to undesirable decision making. In this paper, we propose TeD-SPAD, a privacy-aware video anomaly detection framework that destroys visual private information in a self-supervised manner. In particular, we propose the use of a temporally-distinct triplet loss to promote temporally discriminative features, which complements current weakly-supervised VAD methods. Using TeD-SPAD, we achieve a positive trade-off between privacy protection and utility anomaly detection performance on three popular weakly supervised VAD datasets: UCF-Crime, XD-Violence, and ShanghaiTech. Our proposed anonymization model reduces private attribute prediction by 32.25% while only reducing frame-level ROC AUC on the UCF-Crime anomaly detection dataset by 3.69%.

Temporal action segmentation is crucial for understanding long-form videos. Previous works on this task commonly adopt an iterative refinement paradigm by using multi-stage models. We propose a novel framework via denoising diffusion models, which nonetheless shares the same inherent spirit of such iterative refinement. In this framework, action predictions are iteratively generated from random noise with input video features as conditions. To enhance the modeling of three striking characteristics of human actions, including the position prior, the boundary ambiguity, and the relational dependency, we devise a unified masking strategy for the conditioning inputs in our framework. Extensive experiments on three benchmark datasets, i.e., GTEA, 50Salads, and Breakfast, are performed and the proposed method achieves superior or comparable results to state-of-the-art methods, showing the effectiveness of a generative approach for action segmentation. Code is at tinyurl.com/DiffAct.

Few-shot video action recognition is an effective approach to recognizing new categories with only a few labeled examples, thereby reducing the challenges associated with collecting and annotating large-scale video datasets. Existing methods in video action recognition rely on large labeled datasets from the same domain. However, this setup is not realistic as novel categories may come from different data domains that may have different spatial and temporal characteristics. This dissimilarity between the source and target domains can pose a significant challenge, rendering traditional few-shot action recognition techniques ineffective. To address this issue, in this work, we propose a novel cross-domain few-shot video action recognition method that leverages self-supervised learning and curriculum learning to balance the information from the source and target domains. To be particular, our method employs a masked autoencoder-based self-supervised training objective to learn from both source and target data in a self-supervised manner. Then a progressive curriculum balances learning the discriminative information from the source dataset with the generic information learned from the target domain. Initially, our curriculum utilizes supervised learning to learn class discriminative features from the source data. As the training progresses, we transition to learning target-domain-specific features. We propose a progressive curriculum to encourage the emergence of rich features in the target domain based on class discriminative supervised features in the source domain. We evaluate our method on several challenging benchmark datasets and demonstrate that our approach outperforms existing cross-domain few-shot learning techniques. Our code is available at https://github.com/Sarinda251/CDFSL-V

Self-supervised learning on large-scale multi-modal datasets allows learning semantically meaningful embeddings in a joint multi-modal representation space without relying on human annotations. These joint embeddings enable zero-shot cross-modal tasks like retrieval and classification. However, these methods often struggle to generalize well on out-of-domain data as they ignore the semantic structure present in modality-specific embeddings. In this context, we propose a novel Semantic-Structure-Preserving Consistency approach to improve generalizability by preserving the modality-specific relationships in the joint embedding space. To capture modality-specific semantic relationships between samples, we propose to learn multiple anchors and represent the multifaceted relationship between samples with respect to their relationship with these anchors. To assign multiple anchors to each sample, we propose a novel Multi-Assignment Sinkhorn-Knopp algorithm. Our experimentation demonstrates that our proposed approach learns semantically meaningful anchors in a self-supervised manner. Furthermore, our evaluation on MSR-VTT and YouCook2 datasets demonstrates that our proposed multi-anchor assignment based solution achieves state-of-the-art performance and generalizes to both in-and out-of-domain datasets. Code: https://github.com/Swetha5/Multi_Sinkhorn_Knopp

An oft-cited open problem of federated learning is the existence of data heterogeneity among clients. One pathway to understanding the drastic accuracy drop in federated learning is by scrutinizing the behavior of the clients’ deep models on data with different levels of "difficulty", which has been left unaddressed. In this paper, we investigate a different and rarely studied dimension of FL: ordered learning. Specifically, we aim to investigate how ordered learning principles can contribute to alleviating the heterogeneity effects in FL. We present theoretical analysis and conduct extensive empirical studies on the efficacy of orderings spanning three kinds of learning: curriculum, anti-curriculum, and random curriculum. We find that curriculum learning largely alleviates non-IIDness. Interestingly, the more disparate the data distributions across clients the more they benefit from ordered learning. We provide analysis explaining this phenomenon, specifically indicating how curriculum training appears to make the objective landscape progressively less convex, suggesting fast converging iterations at the beginning of the training procedure. We derive quantitative results of convergence for both convex and nonconvex objectives by modeling the curriculum training on federated devices as local SGD with locally biased stochastic gradients. Also, inspired by ordered learning, we propose a novel client selection technique that benefits from the real-world disparity in the clients. Our proposed approach to client selection has a synergic effect when applied together with ordered learning in FL.

Recent video recognition models utilize Transformer models for long-range spatio-temporal context modeling. Video transformer designs are based on self-attention that can model global context at a high computational cost. In comparison, convolutional designs for videos offer an efficient alternative but lack long-range dependency modeling. Towards achieving the best of both designs, this work proposes Video-FocalNet, an effective and efficient architecture for video recognition that models both local and global contexts. Video-FocalNet is based on a spatiotemporal focal modulation architecture that reverses the interaction and aggregation steps of self-attention for better efficiency. Further, the aggregation step and the interaction step are both implemented using efficient convolution and element-wise multiplication operations that are computationally less expensive than their self-attention counterparts on video representations. We extensively explore the design space of focal modulation-based spatiotemporal context modeling and demonstrate our parallel spatial and temporal encoding design to be the optimal choice. Video-FocalNets perform favorably well against the state-of-the-art transformer-based models for video recognition on five large-scale datasets (Kinetics-400, Kinetics-600, SS-v2, Diving-48, and ActivityNet-1.3) at a lower computational cost. Our code/models are released at https://github.com/TalalWasim/Video-FocalNets.

Missing/erroneous data is a major problem in today’s world. Collected seismic data sometimes contain gaps due to multitude of reasons like interference and sensor malfunction. Gaps in seismic waveforms hamper further signal processing to gain valuable information. Plethora of techniques are used for data reconstruction in other domains like image, video, audio, but translation of those methods to address seismic waveforms demands adapting them to lengthy sequence inputs, which is practically complex. Even if that is accomplished, high computational costs and inefficiency would still persist in these predominantly convolution-based reconstruction models. In this paper, we present a transformer-based deep learning model, Xi-Net, which utilizes multi-faceted time and frequency domain inputs for accurate waveform reconstruction. Xi-Net converts the input waveform to frequency domain, employs separate encoders for time and frequency domains, and one decoder for getting reconstructed output waveform from the fused features. 1D shifted-window transformer blocks form the elementary units of all parts of the model. To the best of our knowledge, this is the first transformer-based deep learning model for seismic waveform reconstruction. We demonstrate this model’s prowess by filling 0.5-1s random gaps in 120s waveforms, resembling the original waveform quite closely. The code, models can be found ${\color{red}{{\text{here}}}}$.

We propose a method for joint detection and tracking of multiple objects in 3D point clouds, a task conventionally treated as a two-step process comprising object detection followed by data association. Our method embeds both steps into a single end-to-end trainable network eliminating the dependency on external object detectors. Our model exploits temporal information employing multiple frames to detect objects and track them in a single network, thereby making it a utilitarian formulation for real-world scenarios. Computing affinity matrix by employing features similarity across consecutive point cloud scans forms an integral part of visual tracking. We propose an attention-based refinement module to refine the affinity matrix by suppressing erroneous correspondences. The module is designed to capture the global context in affinity matrix by employing self-attention within each affinity matrix and cross-attention across a pair of affinity matrices. Unlike competing approaches, our network does not require complex post-processing algorithms, and directly processes raw LiDAR frames to output tracking results. We demonstrate the effectiveness of our method on three tracking benchmarks: JRDB, Waymo, and KITTI. Experimental evaluations indicate the ability of our model to generalize well across datasets.

Drone-to-drone detection using visual feed has crucial applications, such as detecting drone collisions, detecting drone attacks, or coordinating flight with other drones. However, existing methods are computationally costly, follow non-end-to-end optimization, and have complex multi-stage pipelines, making them less suitable for real-time deployment on edge devices. In this work, we propose a simple yet effective framework, TransVisDrone, that provides an end-to-end solution with higher computational efficiency. We utilize CSPDarkNet-53 network to learn object-related spatial features and VideoSwin model to improve drone detection in challenging scenarios by learning spatio-temporal dependencies of drone motion. Our method achieves state-of-the-art performance on three challenging real-world datasets (Average [email protected]): NPS 0.95, FLDrones 0.75, and AOT 0.80, and a higher throughput than previous methods. We also demonstrate its deployment capability on edge devices and its usefulness in detecting drone-collision (encounter). Project: https://tusharsangam.github.io/TransVisDrone-project-page/

Recognizing and comprehending human actions and gestures is a crucial perception requirement for robots to interact with humans and carry out tasks in diverse domains, including service robotics, healthcare, and manufacturing. Event cameras, with their ability to capture fast-moving objects at a high temporal resolution, offer new opportunities compared to standard action recognition in RGB videos. However, previous research on event camera action recognition has primarily focused on sensor-specific network architectures and image encoding, which may not be suitable for new sensors and limit the use of recent advancement in transformer-based architectures. In this study, we employ using a computationally efficient model, namely the video transformer network (VTN), which initially acquires spatial embeddings per event-frame and then utilizes a temporal self-attention mechanism. This approach separates the spatial and temporal operations, resulting in VTN being more computationally efficient than other video transformers that process spatio-temporal volumes directly. In order to better adopt the VTN for the sparse and finegrained nature of event data, we design Event-Contrastive Loss$\left(\mathscr{L}_{E C}\right)$and event specific augmentations. Proposed$\left(\mathscr{L}_{E C}\right)$promotes learning fine-grained spatial cues in the spatial backbone of VTN by contrasting temporally misaligned frames. We evaluate our method on real-world action recognition of N-EPIC Kitchens dataset, and achieve state-of-the-art results on both protocols - testing in seen kitchen (74.9% accuracy) and testing in unseen kitchens (42.43% and 46.66% Accuracy). Our approach also takes less computation time compared to competitive prior approaches. We also evaluate our method on the standard DVS Gesture recognition dataset, achieving a competitive accuracy of 97.9% compared to prior work that uses dedicated architectures and image-encoding for the DVS dataset. These results demonstrate the potential of our framework EventTransAct for real-world applications of event-camera based action recognition. Project Page: https://tristandb8.github.io/EventTransAct_webpage/

Worldwide Geo-localization aims to pinpoint the precise location of images taken anywhere on Earth. This task has considerable challenges due to the immense variation in geographic landscapes. The image-to-image retrieval-based approaches fail to solve this problem on a global scale as it is not feasible to construct a large gallery of images covering the entire world. Instead, existing approaches divide the globe into discrete geographic cells, transforming the problem into a classification task. However, their performance is limited by the predefined classes and often results in inaccurate localizations when an image's location significantly deviates from its class center. To overcome these limitations, we propose GeoCLIP, a novel CLIP-inspired Image-to-GPS retrieval approach that enforces alignment between the image and its corresponding GPS locations. GeoCLIP's location encoder models the Earth as a continuous function by employing positional encoding through random Fourier features and constructing a hierarchical representation that captures information at varying resolutions to yield a semantically rich high-dimensional feature suitable to use even beyond geo-localization. To the best of our knowledge, this is the first work employing GPS encoding for geo-localization. We demonstrate the efficacy of our method via extensive experiments and ablations on benchmark datasets. We achieve competitive performance with just 20% of training data, highlighting its effectiveness even in limited-data settings. Furthermore, we qualitatively demonstrate geo-localization using a text query by leveraging the CLIP backbone of our image encoder. The project webpage is available at: https://vicentevivan.github.io/GeoCLIP

Denoising diffusion models represent a recent emerging topic in computer vision, demonstrating remarkable results in the area of generative modeling. A diffusion model is a deep generative model that is based on two stages, a forward diffusion stage and a reverse diffusion stage. In the forward diffusion stage, the input data is gradually perturbed over several steps by adding Gaussian noise. In the reverse stage, a model is tasked at recovering the original input data by learning to gradually reverse the diffusion process, step by step. Diffusion models are widely appreciated for the quality and diversity of the generated samples, despite their known computational burdens, i.e., low speeds due to the high number of steps involved during sampling. In this survey, we provide a comprehensive review of articles on denoising diffusion models applied in vision, comprising both theoretical and practical contributions in the field. First, we identify and present three generic diffusion modeling frameworks, which are based on denoising diffusion probabilistic models, noise conditioned score networks, and stochastic differential equations. We further discuss the relations between diffusion models and other deep generative models, including variational auto-encoders, generative adversarial networks, energy-based models, autoregressive models and normalizing flows. Then, we introduce a multi-perspective categorization of diffusion models applied in computer vision. Finally, we illustrate the current limitations of diffusion models and envision some interesting directions for future research.

Transformer models have achieved outstanding results on a variety of language tasks, such as text classification, ma- chine translation, and question answering. This success in the field of Natural Language Processing (NLP) has sparked interest in the computer vision community to apply these models to vision and multi-modal learning tasks. However, visual data has a unique structure, requiring the need to rethink network designs and training methods. As a result, Transformer models and their variations have been suc- cessfully used for image recognition, object detection, seg- mentation, image super-resolution, video understanding, image generation, text-image synthesis, and visual question answering, among other applications.

Most existing deep neural networks are static, which means they can only perform inference at a fixed complexity. But the resource budget can vary substantially across different devices. Even on a single device, the affordable budget can change with different scenarios, and repeatedly training networks for each required budget would be incredibly expensive. Therefore, in this work, we propose a general method called MutualNet to train a single network that can run at a diverse set of resource constraints. Our method trains a cohort of model configurations with various network widths and input resolutions. This mutual learning scheme not only allows the model to run at different width-resolution configurations but also transfers the unique knowledge among these configurations, helping the model to learn stronger representations overall. MutualNet is a general training methodology that can be applied to various network structures (e.g., 2D networks: MobileNets, ResNet, 3D networks: SlowFast, X3D) and various tasks (e.g., image classification, object detection, segmentation, and action recognition), and is demonstrated to achieve consistent improvements on a variety of datasets. Since we only train the model once, it also greatly reduces the training cost compared to independently training several models. Surprisingly, MutualNet can also be used to significantly boost the performance of a single network, if dynamic resource constraints are not a concern. In summary, MutualNet is a unified method for both static and adaptive, 2D and 3D networks. Code and pre-trained models are available at https://github.com/taoyang1122/MutualNet.

Adversarial susceptibility of neural image captioning is still under-explored due to the complex multi-model nature of the task. We introduce a GAN-based adversarial attack to effectively fool encoder-decoder based image captioning frameworks. Unique to our attack is the systematic disruption of the internal representation of an image at the encoder stage which allows control over the captions generated at the decoder stage. We cause the desired disruption with an input perturbation that promotes similarity between the features of the input image with a target image of our choice. The target image provides a convenient handle to control the incorrect captions in our method. We do not assume any knowledge of the decoder module, which makes our attack ‘gray-box’. Moreover, our attack remains agnostic to the type of decoder module, thereby proving effective for RNNs as well as Transformers as the language models. This makes our attack highly pragmatic.

Dense captioning methods generally detect events in videos first and then generate captions for the individual events. Events are localized solely based on the visual cues while ignoring the associated linguistic information and context. Whereas end-to-end learning may implicitly take guidance from language, these methods still fall short of the power of explicit modeling. In this paper, we propose aVisual-Semantic Embedding (ViSE) Frameworkthat models the word(s)-context distributional properties over the entire semantic space and computes weights for all then-gramssuch that higher weights are assigned to the more informativen-grams. The weights are accounted for in learning distributed representations of all the captions to construct a semantic space. To perform the contextualization of visual information and the constructed semantic space in a supervised manner, we designVisual-Semantic Joint Modeling Network (VSJM-Net). The learnedViSEembeddings are then temporally encoded with aHierarchical Descriptor Transformer (HDT). For caption generation, we exploit a transformer architecture to decode the input embeddings into natural language descriptions. Experiments on the large-scale ActivityNet Captions dataset and YouCook-II dataset demonstrate the efficacy of our method.

Detecting abnormal events in video is commonly framed as a one-class classification task, where training videos contain only normal events, while test videos encompass both normal and abnormal events. In this scenario, anomaly detection is an open-set problem. However, some studies assimilate anomaly detection to action recognition. This is a closed-set scenario that fails to test the capability of systems at detecting new anomaly types. To this end, we propose UBnormal, a new supervised open-set benchmark composed of multiple virtual scenes for video anomaly detection. Unlike existing data sets, we introduce abnormal events annotated at the pixel level at training time, for the first time enabling the use of fully-supervised learning methods for abnormal event detection. To preserve the typical open-set formulation, we make sure to include dis-joint sets of anomaly types in our training and test collections of videos. To our knowledge, UBnormal is the first video anomaly detection benchmark to allow a fair head-to-head comparison between one-class open-set models and supervised closed-set models, as shown in our experiments. Moreover, we provide empirical evidence showing that UB-normal can enhance the performance of a state-of-the-art anomaly detection framework on two prominent data sets, Avenue and ShanghaiTech. Our benchmark is freely available at https://github.com/lilygeorgescu/UBnormal.

We propose a novel one-step transformer-based person search framework, PSTR, that jointly performs person detection and re-identification (re-id) in a single architecture. PSTR comprises a person search-specialized (PSS) module that contains a detection encoder-decoder for person detection along with a discriminative re-id decoder for person re-id. The discriminative re-id decoder utilizes a multi-level supervision scheme with a shared decoder for discriminative re-id feature learning and also comprises a part attention block to encode relationship between different parts of a person. We further introduce a simple multi-scale scheme to support re-id across person instances at different scales. PSTR jointly achieves the diverse objectives of object-level recognition (detection) and instance-level matching (re-id). To the best of our knowledge, we are the first to propose an end-to-end one-step transformer-based person search framework. Experiments are performed on two popular benchmarks: CUHK-SYSU and PRW. Our extensive ablations reveal the merits of the proposed contributions. Further, the proposed PSTR sets a new state-of-the-art on both benchmarks. On the challenging PRW benchmark, PSTR achieves a mean average precision (mAP) score of 56.5%. The source code is available at https://github.com/JialeCao001/PSTR.

Visual private information leakage is an emerging key is-sue for the fast growing applications of video understanding like activity recognition. Existing approaches for mitigating privacy leakage in action recognition require privacy labels along with the action labels from the video dataset. However, annotating frames of video dataset for privacy la-bels is not feasible. Recent developments of self-supervised learning (SSL) have unleashed the untapped potential of the unlabeled data. For the first time, we present a novel training framework which removes privacy information from in-put video in a self-supervised manner without requiring pri-vacy labels. Our training framework consists of three main components: anonymization function, self-supervised pri-vacy removal branch, and action recognition branch. We train our framework using a minimax optimization strategy to minimize the action recognition cost function and max-imize the privacy cost function through a contrastive self-supervised loss. Employing existing protocols of known-action and privacy attributes, our framework achieves a competitive action-privacy trade-off to the existing state-of-the-art supervised methods. In addition, we introduce a new protocol to evaluate the generalization of learned the anonymization function to novel-action and privacy at-tributes and show that our self-supervised framework out-performs existing supervised methods. Code available at: https://github.com/DAVEISHAN/SPAct

Open-world object detection (OWOD) is a challenging computer vision problem, where the task is to detect a known set of object categories while simultaneously identifying unknown objects. Additionally, the model must incrementally learn new classes that become known in the next training episodes. Distinct from standard object detection, the OWOD setting poses significant challenges for generating quality candidate proposals on potentially unknown objects, separating the unknown objects from the background and detecting diverse unknown objects. Here, we introduce a novel end-to-end transformer-based framework, OW-DETR, for open-world object detection. The proposed OW-DETR comprises three dedicated components namely, attention-driven pseudo-labeling, novelty classification and objectness scoring to explicitly address the aforementioned OWOD challenges. Our OW-DETR explicitly encodes multi-scale contextual information, possesses less inductive bias, enables knowledge transfer from known classes to the unknown class and can better discriminate between unknown objects and background. Comprehensive experiments are performed on two benchmarks: MS-COCO and PASCAL VOC. The extensive ablations reveal the merits of our proposed contributions. Further, our model out-performs the recently introduced OWOD approach, ORE, with absolute gains ranging from 1.8% to 3.3% in terms of unknown recall on MS-COCO. In the case of incremental object detection, OW-DETR outperforms the state-of-the-art for all settings on PASCAL VOC. Our code is available at https://github.com/akshitac8/OW-DEtr.

Supervised deep learning methods require a large repository of annotated data; hence, label noise is inevitable. Training with such noisy data negatively impacts the generalization performance of deep neural networks. To combat label noise, recent state-of-the-art methods employ some sort of sample selection mechanism to select a possibly clean subset of data. Next, an off-the-shelf semi-supervised learning method is used for training where rejected samples are treated as unlabeled data. Our comprehensive analysis shows that current selection methods disproportionately select samples from easy (fast learnable) classes while rejecting those from relatively harder ones. This creates class imbalance in the selected clean set and in turn, deteriorates performance under high label noise. In this work, we propose UNICON, a simple yet effective sample selection method which is robust to high label noise. To address the disproportionate selection of easy and hard samples, we introduce a Jensen-Shannon divergence based uniform selection mechanism which does not require any probabilistic modeling and hyperparameter tuning. We complement our selection method with contrastive learning to further combat the memorization of noisy labels. Extensive experimentation on multiple benchmark datasets demonstrates the effectiveness of UNICON; we obtain an 11.4% improvement over the current state-of-the-art on CIFAR100 dataset with a 90% noise rate. Our code is publicly available.11https://github.com/nazmul-karim170/UNICON-Noisy-Label

Anomaly detection is commonly pursued as a one-class classification problem, where models can only learn from normal training samples, while being evaluated on both normal and abnormal test samples. Among the successful approaches for anomaly detection, a distinguished category of methods relies on predicting masked information (e.g. patches, future frames, etc.) and leveraging the reconstruction error with respect to the masked information as an abnormality score. Different from related methods, we propose to integrate the reconstruction-based functionality into a novel self-supervised predictive architectural building block. The proposed self-supervised block is generic and can easily be incorporated into various state-of-the-art anomaly detection methods. Our block starts with a convolutional layer with dilated filters, where the center area of the receptive field is masked. The resulting activation maps are passed through a channel attention module. Our block is equipped with a loss that minimizes the reconstruction error with respect to the masked area in the receptive field. We demonstrate the generality of our block by integrating it into several state-of-the-art frameworks for anomaly detection on image and video, providing empirical evidence that shows considerable performance improvements on MVTec AD, Avenue, and ShanghaiTech. We release our code as open source at: https://github.com/ristea/sspcab.

The dominant CNN-based methods for cross-view image geo-localization rely on polar transform and fail to model global correlation. We propose a pure transformer-based approach (TransGeo) to address these limitations from a different perspective. TransGeo takes full advantage of the strengths of transformer related to global information modeling and explicit position information encoding. We further leverage the flexibility of transformer input and propose an attention-guided non-uniform cropping method, so that uninformative image patches are removed with negligible drop on performance to reduce computation cost. The saved computation can be reallocated to increase resolution only for informative patches, resulting in performance improvement with no additional computation cost. This “attend and zoom-in” strategy is highly similar to human behavior when observing images. Remarkably, TransGeo achieves state-of-the-art results on both urban and rural datasets, with significantly less computation cost than CNN-based methods. It does not rely on polar transform and infers faster than CNN-based methods. Code is available at https://github.com/Jeff-Zilence/TransGeo2022.

Unlike traditional optoelectronic satellite imaging, Synthetic Aperture Radar (SAR) allows remote sensing applications to operate under all weather conditions. This makes it uniquely valuable for detecting ships/vessels involved in illegal, unreported, and unregulated (IUU) fishing. While recent work has shown significant improvement in this domain, detecting small objects using noisy point annotations remains an unexplored area. In order to meet the unique challenges of this problem, we propose a progressive training methodology that utilizes two different spatial sampling strategies. Firstly, we use stochastic sampling of background points to reduce the impact of class imbalance and missing labels, and secondly, during the refinement stage, we use hard negative sampling to improve the model. Experimental results on the challenging xView3 dataset show that our method outperforms conventional small object localization methods in a large, noisy dataset of SAR images.Source code for our method can be found at: https://github.com/manupillai308/DeepSAR

Adversarial attacks perturb data to modify a model’s prediction. These perturbations can be crafted in a white-box or black-box setting, depending on whether the target model architecture/weights are known or unknown. Compromised architecture and weights of a model makes it vulnerable to the more powerful white-box attacks. In this work, we determine if a deep model is compromised by distinguishing white-box from black-box adversarial attacks. The proposed method utilizes the internal representations of the target model and a proxy model to increase the detector efficacy. Additionally, it employs a spatial smoothing module to control the strength of white-box attacks relative to black-box attacks, and a proxy module to aid in measuring the transferability of the attack. Both modules work in tandem to increase the contrast of the internal representations between white-box and black-box attacks for better discrimination. We perform a detailed ablation of our method to showcase the importance of the different modules, and show that the spatial smoothing and proxy defense techniques enable our framework to significantly outperform the simple classification baseline on common vision datasets.

Video Instance Segmentation is a fundamental computer vision task that deals with segmenting and tracking object instances across a video sequence. Most existing methods typically accomplish this task by employing a multi-stage top-down approach that usually involves separate networks to detect and segment objects in each frame, followed by associating these detections in consecutive frames using a learned tracking head. In this work, however, we introduce a simple end-to-end trainable bottom-up approach to achieve instance mask predictions at the pixel-level granularity, instead of the typical region-proposals-based approach. Unlike contemporary frame-based models, our network pipeline processes an input video clip as a single 3D volume to incorporate temporal information. The central idea of our formulation is to solve the video instance segmentation task as a tag assignment problem, such that generating distinct tag values (here each tag could be any arbitrary value between 0 and 1) corresponding to different objects essentially separates individual object instances across the video sequence. To this end, we propose a novel spatio-temporal tagging loss that allows for sufficient separation of different objects as well as necessary identification of different instances of the same object. Furthermore, we present a tag-based attention module that improves instance tags, while concurrently learning instance propagation within a video. Evaluations demonstrate that our method provides competitive results on YouTube-VIS and DAVIS’19 datasets, and has minimum run-time compared to other state-of-the-art performance methods.

Video generation is one of the most challenging tasks in Machine Learning and Computer Vision fields of study. In this paper, we tackle the text to video generation problem, which is a conditional form of video generation. Humans can listen/read natural language sentences, and can imagine or visualize what is being described; therefore, we believe that video generation from natural language sentences will have an important impact on Artificial Intelligence. Video generation is relatively a new popular task in Computer Vision (CV), which is far from being solved. The majority of recent works deal with synthetic datasets or real datasets with very limited types of objects, scenes and motions. To the best of our knowledge, this is the very first work on the text (free-form sentences) to video generation on more realistic video datasets like Actor and Action Dataset (A2D) or UCF101. We tackle the complicated problem of video generation by regressing the latent representations of the first and last frames and employing a context-aware interpolation method to build the latent representations of in-between frames. We propose a stacking "upPooling" block to sequentially generate RGB frames out of each latent representations and progressively increase the resolution. Moreover, our proposed Discriminator encodes videos based on single and multiple frames.

Multiple object tracking in 3D point clouds has applications in mobile robots and autonomous driving. This is a challenging problem due to the sparse nature of the point clouds and the added difficulty of annotation in 3D for supervised learning. To overcome these challenges, we propose a neural network architecture that learns effective object features and their affinities in a self supervised fashion for multiple object tracking in 3D point clouds captured with LiDAR sensors. For self supervision, we use two approaches. First, we generate two augmented LiDAR frames from a single real frame by applying translation, rotation and cutout to the objects. Second, we synthesize a LiDAR frame using CAD models or primitive geometric shapes and then apply the above three augmentations to them. Hence, the ground truth object locations and associations are known in both frames for self supervision. This removes the need to annotate object associations in real data, and additionally the need for training data collection and annotation for object detection in synthetic data. To the best of our knowledge, this is the first self supervised multiple object tracking method for 3D data. Our model achieves state of the art results.

We propose Differentiable Temporal Logic (DTL), a model-agnostic framework that introduces temporal constraints to deep networks. DTL treats the outputs of a network as a truth assignment of a temporal logic formula, and computes a temporal logic loss reflecting the consistency between the output and the constraints. We propose a comprehensive set of constraints, which are implicit in data annotations, and incorporate them with deep networks via DTL. We evaluate the effectiveness of DTL on the temporal action segmentation task and observe improved performance and reduced logical errors in the output of different task models. Furthermore, we provide an extensive analysis to visualize the desirable effects of DTL.

Deep visual models are known to be vulnerable to adversarial attacks. The last few years have seen numerous techniques to compute adversarial inputs for these models. However, there are still under-explored avenues in this critical research direction. Among those is the estimation of adversarial textures for 3D models in an end-to-end optimization scheme. In this paper, we propose such a scheme to generate adversarial textures for 3D models that are highly transferable and invariant to different camera views and lighting conditions. Our method makes use of neural rendering with explicit control over the model texture and background. We ensure transferability of the adversarial textures by employing an ensemble of robust and non-robust models. Our technique utilizes 3D models as a proxy to simulate closer to real-life conditions, in contrast to conventional use of 2D images for adversarial attacks. We show the efficacy of our method with extensive experiments.

Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR.

Abnormal event detection in video is a complex computer vision problem that has attracted significant attention in recent years. The complexity of the task arises from the commonly-adopted definition of an abnormal event, that is, a rarely occurring event that typically depends on the surrounding context. Following the standard formulation of abnormal event detection as outlier detection, we propose a background-agnostic framework that learns from training videos containing only normal events. Our framework is composed of an object detector, a set of appearance and motion auto-encoders, and a set of classifiers. Since our framework only looks at object detections, it can be applied to different scenes, provided that normal events are defined identically across scenes and that the single main factor of variation is the background. This makes our method background agnostic, as we rely strictly on objects that can cause anomalies, and not on the background. To overcome the lack of abnormal data during training, we propose an adversarial learning strategy for the auto-encoders. We create a scene-agnostic set of out-of-domain pseudo-abnormal examples, which are correctly reconstructed by the auto-encoders before applying gradient ascent on the pseudo-abnormal examples. We further utilize the pseudo-abnormal examples to serve as abnormal examples when training appearance-based and motion-based binary classifiers to discriminate between normal and abnormal latent features and reconstructions. Furthermore, to ensure that the auto-encoders focus only on the main object inside each bounding box image, we introduce a branch that learns to segment the main object. We compare our framework with the state-of-the-art methods on four benchmark data sets, using various evaluation metrics. Compared to existing methods, the empirical results indicate that our approach achieves favorable performance on all data sets. In addition, we provide region-based and track-based annotations for two large-scale abnormal event detection data sets from the literature, namely ShanghaiTech and Subway.

introduction Share on Introduction to the Special Issue on Fine-Grained Visual Recognition and Re-Identification Authors: Shiliang Zhang Peking University Peking UniversityView Profile , Guorong Li University of Chinese Academy of Sciences University of Chinese Academy of SciencesView Profile , Weigang Zhang Harbin Institute of Technology Harbin Institute of TechnologyView Profile , Qingming Huang University of Chinese Academy of Sciences University of Chinese Academy of SciencesView Profile , Tiejun Huang Peking University Peking UniversityView Profile , Mubarak Shah University of Central Florida University of Central FloridaView Profile , Nicu Sebe University of Trento University of TrentoView Profile Authors Info & Claims ACM Transactions on Multimedia Computing, Communications, and ApplicationsVolume 18Issue 1sFebruary 2022 Article No.: 24pp 1–3https://doi.org/10.1145/3505280Online:25 January 2022Publication History 0citation169DownloadsMetricsTotal Citations0Total Downloads169Last 12 Months169Last 6 weeks22 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

Anomaly detection in video is a challenging computer vision problem. Due to the lack of anomalous events at training time, anomaly detection requires the design of learning methods without full supervision. In this paper, we approach anomalous event detection in video through self-supervised and multi-task learning at the object level. We first utilize a pre-trained detector to detect objects. Then, we train a 3D convolutional neural network to produce discriminative anomaly-specific information by jointly learning multiple proxy tasks: three self-supervised and one based on knowledge distillation. The self-supervised tasks are: (i) discrimination of forward/backward moving objects (arrow of time), (ii) discrimination of objects in consecutive/intermittent frames (motion irregularity) and (iii) reconstruction of object-specific appearance information. The knowledge distillation task takes into account both classification and detection information, generating large prediction discrepancies between teacher and student models when anomalies occur. To the best of our knowledge, we are the first to approach anomalous event detection in video as a multi-task learning problem, integrating multiple self-supervised and knowledge distillation proxy tasks in a single architecture. Our lightweight architecture outperforms the state-of-the-art methods on three benchmarks: Avenue, ShanghaiTech and UCSD Ped2. Additionally, we perform an ablation study demonstrating the importance of integrating self-supervised learning and normality-specific distillation in a multi-task learning setting.

The problem of grounding VQA tasks has seen an increased attention in the research community recently, with most attempts usually focusing on solving this task by using pretrained object detectors. However, pre-trained object detectors require bounding box annotations for detecting relevant objects in the vocabulary, which may not always be feasible for real-life large-scale applications. In this paper, we focus on a more relaxed setting: the grounding of relevant visual entities in a weakly supervised manner by training on the VQA task alone. To address this problem, we propose a visual capsule module with a query-based selection mechanism of capsule features, that allows the model to focus on relevant regions based on the textual cues about visual information in the question. We show that integrating the proposed capsule module in existing VQA systems significantly improves their performance on the weakly supervised grounding task. Overall, we demonstrate the effectiveness of our approach on two state-of-the-art VQA systems, stacked NMN and MAC, on the CLEVR-Answers benchmark, our new evaluation set based on CLEVR scenes with groundtruth bounding boxes for objects that are relevant for the correct answer, as well as on GQA, a real world VQA dataset with compositional questions. We show that the systems with the proposed capsule module consistently outperform the respective baseline systems in terms of answer grounding, while achieving comparable performance on VQA task.1

In many real-world problems, collecting a large number of labeled samples is infeasible. Few-shot learning (FSL) is the dominant approach to address this issue, where the objective is to quickly adapt to novel categories in presence of a limited number of samples. FSL tasks have been predominantly solved by leveraging the ideas from gradient-based meta-learning and metric learning approaches. However, recent works have demonstrated the significance of powerful feature representations with a simple embedding network that can outperform existing sophisticated FSL algorithms. In this work, we build on this insight and propose a novel training mechanism that simultaneously enforces equivariance and invariance to a general set of geometric transformations. Equivariance or invariance has been employed standalone in the previous works; however, to the best of our knowledge, they have not been used jointly. Simultaneous optimization for both of these contrasting objectives allows the model to jointly learn features that are not only independent of the input transformation but also the features that encode the structure of geometric transformations. These complementary sets of features help generalize well to novel classes with only a few data samples. We achieve additional improvements by incorporating a novel self-supervised distillation objective. Our extensive experimentation shows that even without knowledge distillation our proposed method can outperform current state-of-the-art FSL methods on five popular benchmark datasets.

Real-world videos contain many complex actions with inherent relationships between action classes. In this work, we propose an attention-based architecture that models these action relationships for the task of temporal action localization in untrimmed videos. As opposed to previous works that leverage video-level co-occurrence of actions, we distinguish the relationships between actions that occur at the same time-step and actions that occur at different time-steps (i.e. those which precede or follow each other). We define these distinct relationships as action dependencies. We propose to improve action localization performance by modeling these action dependencies in a novel attention-based Multi-Label Action Dependency (MLAD) layer. The MLAD layer consists of two branches: a Cooccurrence Dependency Branch and a Temporal Dependency Branch to model co-occurrence action dependencies and temporal action dependencies, respectively. We observe that existing metrics used for multi-label classification do not explicitly measure how well action dependencies are modeled, therefore, we propose novel metrics that consider both co-occurrence and temporal dependencies between action classes. Through empirical evaluation and extensive analysis, we show improved performance over state-of-the-art methods on multi-label action localization benchmarks (MultiTHUMOS and Charades) in terms of f-mAP and our proposed metric. Code is publicly available at https://github.com/ptirupat/MLAD.

The goal of out-of-distribution (OOD) detection is to handle the situations where the test samples are drawn from a different distribution than the training data. In this paper, we argue that OOD samples can be detected more easily if the training data is embedded into a low-dimensional space, such that the embedded training samples lie on a union of 1-dimensional subspaces. We show that such embedding of the in-distribution (ID) samples provides us with two main advantages. First, due to compact representation in the feature space, OOD samples are less likely to occupy the same region as the known classes. Second, the first singular vector of ID samples belonging to a 1-dimensional subspace can be used as their robust representative. Motivated by these observations, we train a deep neural network such that the ID samples are embedded onto a union of 1-dimensional subspaces. At the test time, employing sampling techniques used for approximate Bayesian inference in deep learning, input samples are detected as OOD if they occupy the region corresponding to the ID samples with probability 0. Spectral components of the ID samples are used as robust representative of this region. Our method does not have any hyperparameter to be tuned using extra information and it can be applied on different modalities with minimal change. The effectiveness of the proposed method is demonstrated on different benchmark datasets, both in the image and video classification domains.

We propose a novel transformer-based styled handwritten text image generation approach, HWT, that strives to learn both style-content entanglement as well as global and local style patterns. The proposed HWT captures the long and short range relationships within the style examples through a self-attention mechanism, thereby encoding both global and local style patterns. Further, the proposed transformer-based HWT comprises an encoder-decoder attention that enables style-content entanglement by gathering the style features of each query character. To the best of our knowledge, we are the first to introduce a transformer-based network for styled handwritten text generation.Our proposed HWT generates realistic styled handwritten text images and outperforms the state-of-the-art demonstrated through extensive qualitative, quantitative and human-based evaluations. The proposed HWT can handle arbitrary length of text and any desired writing style in a few-shot setting. Further, our HWT generalizes well to the challenging scenario where both words and writing style are unseen during training, generating realistic styled handwritten text images. Code is available at: https://github.com/ankanbhunia/Handwriting-Transformers

Multi-label zero-shot learning (ZSL) is a more realistic counter-part of standard single-label ZSL since several objects can co-exist in a natural image. However, the occurrence of multiple objects complicates the reasoning and requires region-specific processing of visual features to preserve their contextual cues. We note that the best existing multi-label ZSL method takes a shared approach towards attending to region features with a common set of attention maps for all the classes. Such shared maps lead to diffused attention, which does not discriminatively focus on relevant locations when the number of classes are large. Moreover, mapping spatially-pooled visual features to the class semantics leads to inter-class feature entanglement, thus hampering the classification. Here, we propose an alternate approach towards region-based discriminability-preserving multi-label zero-shot classification. Our approach maintains the spatial resolution to preserve region-level characteristics and utilizes a bi-level attention module (BiAM) to enrich the features by incorporating both region and scene context information. The enriched region-level features are then mapped to the class semantics and only their class predictions are spatially pooled to obtain image-level predictions, thereby keeping the multi-class features disentangled. Our approach sets a new state of the art on two large-scale multi-label zero-shot benchmarks: NUS-WIDE and Open Images. On NUS-WIDE, our approach achieves an absolute gain of 6.9% mAP for ZSL, compared to the best published results. Source code is available at https://github.com/akshitac8/BiAM.

In this paper, we address the problem of video geo-localization by proposing a Geo-Temporal Feature Learning (GTFL) Network to simultaneously learn the discriminative features for the query video frames and the gallery images for estimating the geo-spatial trajectory of a query video. Based on a transformer encoder architecture, our GTFL model encodes query and gallery data separately, via two dedicated branches. The proposed GPS Loss and Clip Triplet Loss exploit the geographical and temporal proximity between the frames and the clips to jointly learn the query and the gallery features. We also propose a deep learning approach to trajectory smoothing by predicting the outliers in the estimated GPS positions and learning the offsets to smooth the trajectory. We build a large dataset from four different regions of USA; New York, San Francisco, Berkeley and Bay Area using BDD driving videos as query, and by collecting corresponding Google StreetView (GSV) Images for gallery. Extensive evaluations of proposed method on this new dataset are provided . Code and dataset details is publicly available at https://github.com/kregmi/VTE.

Current face image retrieval solutions are limited, since they treat different facial attributes the same and cannot incorporate user’s preference for a subset of attributes in their search criteria. This paper introduces a new face image retrieval framework, where the input face query is augmented by both an adjustment vector that specifies the desired modifications to the facial attributes, and a preference vector that assigns different levels of importance to different attributes. For example, a user can ask for retrieving images similar to a query image, but with a different hair color, and no preference for absence/presence of eyeglasses in the results. To achieve this, we propose to disentangle the semantics, corresponding to various attributes, by learning a set of sparse and orthogonal basis vectors in the latent space of StyleGAN. Such basis vectors are then employed to decompose the dissimilarity between face images in terms of dissimilarity between their attributes, assign preference to the attributes, and adjust the attributes in the query. Enforcing sparsity on the basis vectors helps us to disentangle the latent space and adjust each attribute independently from other attributes, while enforcing orthogonality facilitates preference assignment and the dissimilarity decomposition. The effectiveness of our approach is illustrated by achieving state-of-the-art results for the face image retrieval task.

We address the problem of novel view video prediction; given a set of input video clips from a single/multiple views, our network is able to predict the video from a novel view. The proposed approach does not require any priors and is able to predict the video from wider angular distances, upto 45 degree, as compared to the recent studies predicting small variations in viewpoint. Moreover, our method relies only on RGB frames to learn a dual representation which is used to generate the video from a novel viewpoint. The dual representation encompasses a view-dependent and a global representation which incorporates complementary details to enable novel view video prediction. We demonstrate the effectiveness of our framework on two real world datasets: NTU- RGB+D and CMU Panoptic. A comparison with the State-of-the-art novel view video prediction methods shows an improvement of 26.1% in SSIM, 13.6% in PSNR, and 60% in FVD scores without using explicit priors from target views.

This paper proposes a novel approach for unsupervised sub-action learning in complex activities. The proposed method maps both visual and temporal representations to a latent space where the sub-actions are learnt discriminatively in an end-to-end fashion. To this end, we propose to learn sub-actions as latent concepts and a novel discriminative latent concept learning (DLCL) module aids in learning sub-actions. The proposed DLCL module lends on the idea of latent concepts to learn compact representations in the latent embedding space in an unsupervised way. The result is a set of latent vectors that can be interpreted as cluster centers in the embedding space. Our joint embedding learning with discriminative latent concept module is novel which eliminates the need for explicit clustering. We validate our approach on three benchmark datasets and show that the proposed combination of visual-temporal embedding and discriminative latent concepts allow to learn robust action representations in unsupervised setting.

The goal of zero-shot action recognition (ZSAR) is to classify action classes which were not previously seen during training. Traditionally, this is achieved by training a network to map, or regress, visual inputs to a semantic space where a nearest neighbor classifier is used to select the closest target class. We argue that this approach is sub-optimal due to the use of nearest neighbor on static semantic space and is ineffective when faced with multi-label videos - where two semantically distinct co-occurring action categories cannot be predicted with high confidence. To overcome these limitations, we propose a ZSAR framework which does not rely on nearest neighbor classification, but rather consists of a pairwise scoring function. Given a video and a set of action classes, our method predicts a set of confidence scores for each class independently. This allows for the prediction of several semantically distinct classes within one video input. Our evaluations show that our method not only achieves strong performance on three single-label action classification datasets (UCF-101, HMDB, and RareAct), but also outperforms previous ZSAR approaches on a challenging multi-label dataset (AVA) and a real-world surprise activity detection dataset (MEVA).

Attributes are semantically meaningful characteristics whose applicability widely crosses category boundaries. They are particularly important in describing and recognizing concepts for which no explicit training example is given, e.g., zero-shot learning. Additionally, since attributes are human describable, they can be used for efficient human-computer interaction. In this article, we propose to employ semantic segmentation to improve person-related attribute prediction. The core idea lies in the fact that many attributes describe local properties. In other words, the probability of an attribute to appear in an image is far from being uniform in the spatial domain. We build our attribute prediction model jointly with a deep semantic segmentation network. This harnesses the localization cues learned by the semantic segmentation to guide the attention of the attribute prediction to the regions where different attributes naturally show up. As a result of this approach, in addition to prediction, we are able to localize the attributes despite merely having access to image-level labels (weak supervision) during training. We first propose semantic segmentation-based pooling and gating, respectively denoted as SSP and SSG. In the former, the estimated segmentation masks are used to pool the final activations of the attribute prediction network, from multiple semantically homogeneous regions. This is in contrast to global average pooling which is agnostic with respect to where in the spatial domain activations occur. In SSG, the same idea is applied to the intermediate layers of the network. Specifically, we create multiple copies of the internal activations. In each copy, only values that fall within a certain semantic region are preserved while outside of that, activations are suppressed. This mechanism allows us to prevent pooling operation from blending activations that are associated with semantically different regions. SSP and SSG, while effective, impose heavy memory utilization since each channel of the activations is pooled/gated with all the semantic segmentation masks. To circumvent this, we propose Symbiotic Augmentation (SA), where we learn only one mask per activation channel. SA allows the model to either pick one, or combine (weighted superposition) multiple semantic maps, in order to generate the proper mask for each channel. SA simultaneously applies the same mechanism to the reverse problem by leveraging output logits of attribute prediction to guide the semantic segmentation task. We evaluate our proposed methods for facial attributes on CelebA and LFWA datasets, while benchmarking WIDER Attribute and Berkeley Attributes of People for whole body attributes. Our proposed methods achieve superior results compared to the previous works. Furthermore, we show that in the reverse problem, semantic face parsing significantly improves when its associated task is jointly learned, through our proposed Symbiotic Augmentation (SA), with facial attribute prediction. We confirm that when few training instances are available, indeed image-level facial attribute labels can serve as an effective source of weak supervision to improve semantic face parsing. That reaffirms the need to jointly model these two interconnected tasks.

This work presents a novel method of exploring human brain-visual representations, with a view towards replicating these processes in machines. The core idea is to learn plausible computational and biological representations by correlating human neural activity and natural images. Thus, we first propose a model, EEG-ChannelNet, to learn a brain manifold for EEG classification. After verifying that visual information can be extracted from EEG data, we introduce a multimodal approach that uses deep image and EEG encoders, trained in a siamese configuration, for learning a joint manifold that maximizes a compatibility measure between visual features and brain representations. We then carry out image classification and saliency detection on the learned manifold. Performance analyses show that our approach satisfactorily decodes visual information from neural signals. This, in turn, can be used to effectively supervise the training of deep learning models, as demonstrated by the high performance of image classification and saliency detection on out-of-training classes. The obtained results show that the learned brain-visual features lead to improved performance and simultaneously bring deep models more in line with cognitive neuroscience work related to visual perception and attention.

Multiple Object Tracking (MOT) plays an important role in solving many fundamental problems in video analysis and computer vision. Most MOT methods employ two steps: Object Detection and Data Association. The first step detects objects of interest in every frame of a video, and the second establishes correspondence between the detected objects in different frames to obtain their tracks. Object detection has made tremendous progress in the last few years due to deep learning. However, data association for tracking still relies on hand crafted constraints such as appearance, motion, spatial proximity, grouping etc. to compute affinities between the objects in different frames. In this paper, we harness the power of deep learning for data association in tracking by jointly modeling object appearances and their affinities between different frames in an end-to-end fashion. The proposed Deep Affinity Network (DAN) learns compact, yet comprehensive features of pre-detected objects at several levels of abstraction, and performs exhaustive pairing permutations of those features in any two frames to infer object affinities. DAN also accounts for multiple objects appearing and disappearing between video frames. We exploit the resulting efficient affinity computations to associate objects in the current frame deep into the previous frames for reliable on-line tracking. Our technique is evaluated on popular multiple object tracking challenges MOT15, MOT17 and UA-DETRAC. Comprehensive benchmarking under twelve evaluation metrics demonstrates that our approach is among the best performing techniques on the leader board for these challenges. The open source implementation of our work is available at https://github.com/shijieS/SST.git.

Augmenting neural networks with skip connections, as introduced in the so-called ResNet architecture, surprised the community by enabling the training of networks of more than 1,000 layers with significant performance gains. This paper deciphers ResNet by analyzing the effect of skip connections, and puts forward new theoretical results on the advantages of identity skip connections in neural networks. We prove that the skip connections in the residual blocks facilitate preserving the norm of the gradient, and lead to stable back-propagation, which is desirable from optimization perspective. We also show that, perhaps surprisingly, as more residual blocks are stacked, the norm-preservation of the network is enhanced. Our theoretical arguments are supported by extensive empirical evidence. Can we push for extra norm-preservation? We answer this question by proposing an efficient method to regularize the singular values of the convolution operator and making the ResNet's transition layers extra norm-preserving. Our numerical investigations demonstrate that the learning dynamics and the classification performance of ResNet can be improved by making it even more norm preserving. Our results and the introduced modification for ResNet, referred to as Procrustes ResNets, can be used as a guide for training deeper networks and can also inspire new deeper architectures.

Along with the success of deep neural network (DNN) models, rise the threats to the integrity of these models. A recent threat is the Trojan attack where an attacker interferes with the training pipeline by inserting triggers into some of the training samples and trains the model to act maliciously only for samples that contain the trigger. Since the knowledge of triggers is privy to the attacker, detection of Trojan networks is challenging. Existing Trojan detectors make strong assumptions about the types of triggers and attacks. We propose a detector that is based on the analysis of the intrinsic DNN properties; that are affected due to the Trojan insertion process. For a comprehensive analysis, we develop Odyssey, the most diverse dataset to date with over 3,000 clean and Trojan models. Odyssey covers a large spectrum of attacks; generated by leveraging the versatility in trigger designs and source to target class mappings. Our analysis results show that Trojan attacks affect the classifier margin and shape of decision boundary around the manifold of clean data. Exploiting these two factors, we propose an efficient Trojan detector that operates without any knowledge of the attack and significantly outperforms existing methods. Through a comprehensive set of experiments we demonstrate the efficacy of the detector on cross model architectures, unseen Triggers and regularized models.

We have recently seen great progress in image classification due to the success of deep convolutional neural networks and the availability of large-scale datasets. Most of the existing work focuses on single-label image classification. However, there are usually multiple tags associated with an image. The existing works on multi-label classification are mainly based on lab curated labels. Humans assign tags to their images differently, which is mainly based on their interests and personal tagging behavior. In this paper, we address the problem of personalized tag recommendation and propose an end-to-end deep network which can be trained on large-scale datasets. The user-preference is learned within the network in an unsupervised way where the network performs joint optimization for user-preference and visual encoding. A joint training of user-preference and visual encoding allows the network to efficiently integrate the visual preference with tagging behavior for a better user recommendation. In addition, we propose the use of adversarial learning, which enforces the network to predict tags resembling user-generated tags. We demonstrate the effectiveness of the proposed model on two different large-scale and publicly available datasets, YFCC100 M and NUS-WIDE. The proposed method achieves significantly better performance on both the datasets when compared to the baselines and other state-of-the-art methods. The code is publicly available at https://github.com/vyzuer/ALTReco.

Convolutional neural networks (CNNs) have become a key asset to most of fields in AI. Despite their successful performance, CNNs suffer from a major drawback. They fail to capture the hierarchy of spatial relation among different parts of an entity. As a remedy to this problem, the idea of capsules was proposed by Hinton. In this paper, we propose the SubSpace Capsule Network (SCN) that exploits the idea of capsule networks to model possible variations in the appearance or implicitly-defined properties of an entity through a group of capsule subspaces instead of simply grouping neurons to create capsules. A capsule is created by projecting an input feature vector from a lower layer onto the capsule subspace using a learnable transformation. This transformation finds the degree of alignment of the input with the properties modeled by the capsule subspace.We show that SCN is a general capsule network that can successfully be applied to both discriminative and generative models without incurring computational overhead compared to CNN during test time. Effectiveness of SCN is evaluated through a comprehensive set of experiments on supervised image classification, semi-supervised image classification and high-resolution image generation tasks using the generative adversarial network (GAN) framework. SCN significantly improves the performance of the baseline models in all 3 tasks.

Finding a small subset of data whose linear combination spans other data points, also called column subset selection problem (CSSP), is an important open problem in computer science with many applications in computer vision and deep learning. There are some studies that solve CSSP in a polynomial time complexity w.r.t. the size of the original dataset. A simple and efficient selection algorithm with a linear complexity order, referred to as spectrum pursuit (SP), is proposed that pursuits spectral components of the dataset using available sample points. The proposed non-greedy algorithm aims to iteratively find K data samples whose span is close to that of the first K spectral components of entire data. SP has no parameter to be fine tuned and this desirable property makes it problem-independent. The simplicity of SP enables us to extend the underlying linear model to more complex models such as nonlinear manifolds and graph-based models. The nonlinear extension of SP is introduced as kernel-SP (KSP). The superiority of the proposed algorithms is demonstrated in a wide range of applications.

Joint understanding of vision and natural language is a challenging problem with a wide range of applications in artificial intelligence. In this work, we focus on integration of video and text for the task of actor and action video segmentation from a sentence. We propose a capsule-based approach which performs pixel-level localization based on a natural language query describing the actor of interest. We encode both the video and textual input in the form of capsules, which provide a more effective representation in comparison with standard convolution based features. Our novel visual-textual routing mechanism allows for the fusion of video and text capsules to successfully localize the actor and action. The existing works on actor-action localization are mainly focused on localization in a single frame instead of the full video. Different from existing works, we propose to perform the localization on all frames of the video. To validate the potential of the proposed network for actor and action video localization, we extend an existing actor-action dataset (A2D) with annotations for all the frames. The experimental evaluation demonstrates the effectiveness of our capsule network for text selective actor and action localization in videos. The proposed method also improves upon the performance of the existing state-of-the art works on single frame-based localization.

Humans can continuously learn new knowledge as their experience grows. In contrast, previous learning in deep neural networks can quickly fade out when they are trained on a new task. In this paper, we hypothesize this problem can be avoided by learning a set of generalized parameters, that are neither specific to old nor new tasks. In this pursuit, we introduce a novel meta-learning approach that seeks to maintain an equilibrium between all the encountered tasks. This is ensured by a new meta-update rule which avoids catastrophic forgetting. In comparison to previous meta-learning techniques, our approach is task-agnostic. When presented with a continuum of data, our model automatically identifies the task and quickly adapts to it with just a single update. We perform extensive experiments on five datasets in a class-incremental setting, leading to significant improvements over the state of the art methods (e.g., a 21.3% boost on CIFAR100 with 10 incremental tasks). Specifically, on large-scale datasets that generally prove difficult cases for incremental learning, our approach delivers absolute gains as high as 19.1% and 7.4% on ImageNet and MS-Celeb datasets, respectively.

This paper introduces a new type of image enhancement problem. Compared to traditional image enhancement methods, which mostly deal with pixel-wise modifications of a given photo, our proposed task is to crop an image which is embedded within a photo and enhance the quality of the cropped image. We split our proposed approach into two deep networks: deep photo cropper and deep image enhancer. In the photo cropper network, we employ a spatial transformer to extract the embedded image. In the photo enhancer, we employ super-resolution to increase the number of pixels in the embedded image and reduce the effect of stretching and distortion of pixels. We use cosine distance loss between image features and ground truth for the cropper and the mean square loss for the enhancer. Furthermore, we propose a new dataset to train and test the proposed method. Finally, we analyze the proposed method with respect to qualitative and quantitative evaluations.

The existing research in action recognition is mostly focused on high-quality videos where the action is distinctly visible. In real-world surveillance environments, the actions in videos are captured at a wide range of resolutions. Most activities occur at a distance with a small resolution and recognizing such activities is a challenging problem. In this work, we focus on recognizing tiny actions in videos. We introduce a benchmark dataset, Tiny VIRAT, which contains natural low-resolution activities. The actions in Tiny VIRAT videos have multiple labels and they are extracted from surveillance videos which makes them realistic and more challenging. We propose a novel method for recognizing tiny actions in videos which utilizes a progressive generative approach to improve the quality of low-resolution actions. The proposed method also consists of a weakly trained attention mechanism which helps in focusing on the activity regions in the video. We perform extensive experiments to benchmark the proposed Tiny VIRAT dataset and observe that the proposed method significantly improves the action recognition performance over baselines. We also evaluate the proposed approach on synthetically resized action recognition datasets and achieve state-of-the-art results when compared with existing methods. The dataset and code are publicly available at https://github.com/UgurDemir/Tiny-VIRAT.

Accurate and fine-grained information about the extent of damage to buildings is essential for directing Humanitarian Aid and Disaster Response (HADR) operations in the immediate aftermath of any natural calamity. In recent years, satellite and UAV (drone) imagery has been used for this purpose, sometimes aided by computer vision algorithms. Existing Computer Vision approaches for building damage assessment typically rely on a two stage approach, consisting of building detection using an object detection model, followed by damage assessment through classification of the detected building tiles. These multi-stage methods are not end-to-end trainable, and suffer from poor overall results. We propose RescueNet, a unified model that can simultaneously segment buildings and assess the damage levels to individual buildings and can be trained end-to-end. In order to to model the composite nature of this problem, we propose a novel localization aware loss function, which consists of a Binary Cross Entropy loss for building segmentation, and a foreground only selective Categorical Cross-Entropy loss for damage classification, and show significant improvement over the widely used Cross-Entropy loss. RescueNet is tested on the large scale and diverse xBD dataset and achieves significantly better building segmentation and damage classification performance than previous methods and achieves generalization across varied geographical regions and disaster types.

Activity detection in security videos is a difficult problem due to multiple factors such as large field of view, presence of multiple activities, varying scales and viewpoints, and its untrimmed nature. The existing research in activity detection is mainly focused on datasets, such as UCF-101, JHMDB, THUMOS, and AVA, which partially address these issues. The requirement of processing security videos in real-time makes this even more challenging. In this work, we propose Gabriella, a real-time online system to perform activity detection on untrimmed security videos. The proposed method consists of three stages: tubelet extraction, activity classification, and online tubelet merging. For tubelet extraction, we propose a localization network which takes a video clip as input and spatio-temporally detects potential foreground regions at multiple scales to generate action tubelets. We propose a novel Patch-Dice loss to handle large variations in actor size. Our online processing of videos at a clip level drastically reduces the computation time in detecting activities. The detected tubelets are assigned activity class scores by the classification network and merged together using our proposed Tubelet-Merge Action-Split (TMAS) algorithm to form the final action detections. The TMAS algorithm efficiently connects the tubelets in an online fashion to generate action detections which are robust against varying length activities. We perform our experiments on the VIRAT and MEVA (Multiview Extended Video with Activities) datasets and demonstrate the effectiveness of the proposed approach in terms of speed (~100 fps) and performance with state-of-the-art results. More details about this work are available on our project webpage11https://www.crcv.ucf.edu/research/projects/gabriella-an-online-system-for-real-time-activity-detection-in-untrimmed-security-videos.

Mass utilization of body-worn cameras has led to a huge corpus of available egocentric video. Existing video summarization algorithms can accelerate browsing such videos by selecting (visually) interesting shots from them. Nonetheless, since the system user still has to watch the summary videos, browsing large video databases remain a challenge. Hence, in this work, we propose to generate a textual synopsis, consisting of a few sentences describing the most important events in a long egocentric videos. Users can read the short text to gain insight about the video, and more importantly, efficiently search through the content of a large video database using text queries. Since egocentric videos are long and contain many activities and events, using video-to-text algorithms results in thousands of descriptions, many of which are incorrect. Therefore, we propose a multi-task learning scheme to simultaneously generate descriptions for video segments and summarize the resulting descriptions in an end-to-end fashion. We Input a set of video shots and the network generates a text description for each shot. Next, visual-language content matching unit that is trained with a weakly supervised objective, identifies the correct descriptions. Finally, the last component of our network, called purport network, evaluates the descriptions all together to select the ones containing crucial information. Out of thousands of descriptions generated for the video, a few informative sentences are returned to the user. We validate our framework on the challenging UT Egocentric video dataset, where each video is between 3 to 5 hours long, associated with over 3000 textual descriptions on average. The generated textual summaries, including only 5 percent (or less) of the generated descriptions, are compared to groundtruth summaries in text domain using well-established metrics in natural language processing.1.

Batch Normalization (BN) is essential to effectively train state-of-the-art deep Convolutional Neural Networks (CNN). It normalizes the layer outputs during training using the statistics of each mini-batch. BN accelerates training procedure by allowing to safely utilize large learning rates and alleviates the need for careful initialization of the parameters. In this work, we study BN from the viewpoint of Fisher kernels that arise from generative probability models. We show that assuming samples within a mini-batch are from the same probability density function, then BN is identical to the Fisher vector of a Gaussian distribution. That means batch normalizing transform can be explained in terms of kernels that naturally emerge from the probability density function that models the generative process of the underlying data distribution. Consequently, it promises higher discrimination power for the batch-normalized mini-batch. However, given the rectifying non-linearities employed in CNN architectures, distribution of the layer outputs show an asymmetric characteristic. Therefore, in order for BN to fully benefit from the aforementioned properties, we propose approximating underlying data distribution not with one, but a mixture of Gaussian densities. Deriving Fisher vector for a Gaussian Mixture Model (GMM), reveals that batch normalization can be improved by independently normalizing with respect to the statistics of disentangled sub-populations. We refer to our proposed soft piecewise version of batch normalization as Mixture Normalization (MN). Through extensive set of experiments on CIFAR-10 and CIFAR-100, using both a 5-layers deep CNN and modern Inception-V3 architecture, we show that mixture normalization reduces required number of gradient updates to reach the maximum test accuracy of the batch-normalized model by ∼ 31%-47% across a variety of training scenarios. Replacing even a few BN modules with MN in the 48-layers deep Inception-V3 architecture is sufficient to not only obtain considerable training acceleration but also better final test accuracy. We show that similar observations are valid for 40 and 100-layers deep DenseNet architectures as well. We complement our study by evaluating the application of mixture normalization to the Generative Adversarial Networks (GANs), where "mode collapse" hinders the training process. We solely replace a few batch normalization layers in the generator with our proposed mixture normalization. Our experiments using Deep Convolutional GAN (DCGAN) on CIFAR-10 show that mixture-normalized DCGAN not only provides an acceleration of ∼ 58% but also reaches lower (better) "Fréchet Inception Distance" (FID) of 33.35 compared to 37.56 of its batch-normalized counterpart.

Animals search for food in their environment with a decision strategy which keeps them fit. Optimal foraging theory models this foraging behavior to determine the optimal decision strategy followed by animals. This theory has been successfully applied for humans as they search for information and is termed as information foraging. When people visit a tourist location, they follow a similar strategy to move from one spot to another, and collect information by capturing photographs. This behavior has similarities with the foraging behavior of animals which has been widely studied by the researchers. In this paper, we propose to employ optimal foraging theory to help tourists explore a location and capture photographs in an optimal way. We determine a decision strategy for tourist which provides a list of interesting spots to visit in a tourist location along with corresponding stay time. Finally, we solve an optimization problem to find a path through these spots which can be followed by tourists. The experimental results on a public dataset demonstrate the effectiveness of the proposed method.1Code available at https://github.com/vyzuer/foraging_theory.

Several recent studies have demonstrated the promise of deep visuomotor policies for robot manipulator control. Despite impressive progress, these systems are known to be vulnerable to physical disturbances, such as accidental or adversarial bumps that make them drop the manipulated object. They also tend to be distracted by visual disturbances such as objects moving in the robot's field of view, even if the disturbance does not physically prevent the execution of the task. In this paper, we propose an approach for augmenting a deep visuomotor policy trained through demonstrations with Task Focused visual Attention (TFA). The manipulation task is specified with a natural language text such as "move the red bowl to the left". This allows the visual attention component to concentrate on the current object that the robot needs to manipulate. We show that even in benign environments, the TFA allows the policy to consistently outperform a variant with no attention mechanism. More importantly, the new policy is significantly more robust: it regularly recovers from severe physical disturbances (such as bumps causing it to drop the object) from which the baseline policy, i.e. with no visual attention, almost never recovers. In addition, we show that the proposed policy performs correctly in the presence of a wide class of visual disturbances, exhibiting a behavior reminiscent of human selective visual attention experiments.

The goal of data selection is to capture the most structural information from a set of data. This paper presents a fast and accurate data selection method, in which the selected samples are optimized to span the subspace of all data. We propose a new selection algorithm, referred to as iterative projection and matching (IPM), with linear complexity w.r.t. the number of data, and without any parameter to be tuned. In our algorithm, at each iteration, the maximum information from the structure of the data is captured by one selected sample, and the captured information is neglected in the next iterations by projection on the null-space of previously selected samples. The computational efficiency and the selection accuracy of our proposed algorithm outperform those of the conventional methods. Furthermore, the superiority of the proposed algorithm is shown on active learning for video action recognition dataset on UCF-101; learning using representatives on ImageNet; training a generative adversarial network (GAN) to generate multi-view images from a single-view input on CMU Multi-PIE dataset; and video summarization on UTE Egocentric dataset.

In this work, we propose an end-to-end constrained clustering scheme to tackle the person re-identification (re-id) problem. Deep neural networks (DNN) have recently proven to be effective on person re-identification task. In particular, rather than leveraging solely a probe-gallery similarity, diffusing the similarities among the gallery images in an end-to-end manner has proven to be effective in yielding a robust probe-gallery affinity. However, existing methods do not apply probe image as a constraint, and are prone to noise propagation during the similarity diffusion process. To overcome this, we propose an intriguing scheme which treats person-image retrieval problem as a constrained clustering optimization problem, called deep constrained dominant sets (DCDS). Given a probe and gallery images, we re-formulate person re-id problem as finding a constrained cluster, where the probe image is taken as a constraint (seed) and each cluster corresponds to a set of images corresponding to the same person. By optimizing the constrained clustering in an end-to-end manner, we naturally leverage the contextual knowledge of a set of images corresponding to the given person-images. We further enhance the performance by integrating an auxiliary net alongside DCDS, which employs a multi-scale ResNet. To validate the effectiveness of our method we present experiments on several benchmark datasets and show that the proposed method can outperform state-of-the-art methods.

In this work we propose a capsule-based approach for semi-supervised video object segmentation. Current video object segmentation methods are frame-based and often require optical flow to capture temporal consistency across frames which can be difficult to compute. To this end, we propose a video based capsule network, CapsuleVOS, which can segment several frames at once conditioned on a reference frame and segmentation mask. This conditioning is performed through a novel routing algorithm for attention-based efficient capsule selection. We address two challenging issues in video object segmentation: 1) segmentation of small objects and 2) occlusion of objects across time. The issue of segmenting small objects is addressed with a zooming module which allows the network to process small spatial regions of the video. Apart from this, the framework utilizes a novel memory module based on recurrent networks which helps in tracking objects when they move out of frame or are occluded. The network is trained end-to-end and we demonstrate its effectiveness on two benchmark video object segmentation datasets; it outperforms current offline approaches on the Youtube-VOS dataset while having a run-time that is almost twice as fast as competing methods. The code is publicly available at https://github.com/KevinDuarte/CapsuleVOS.

The visual entities in cross-view (e.g. ground and aerial) images exhibit drastic domain changes due to the differences in viewpoints each set of images is captured from. Existing state-of-the-art methods address the problem by learning view-invariant images descriptors. We propose a novel method for solving this task by exploiting the generative powers of conditional GANs to synthesize an aerial representation of a ground-level panorama query and use it to minimize the domain gap between the two views. The synthesized image being from the same view as the reference (target) image, helps the network to preserve important cues in aerial images following our Joint Feature Learning approach. We fuse the complementary features from a synthesized aerial image with the original ground-level panorama features to obtain a robust query representation. In addition, we employ multi-scale feature aggregation in order to preserve image representations at different scales useful for solving this complex task. Experimental results show that our proposed approach performs significantly better than the state-of-the-art methods on the challenging CVUSA dataset in terms of top-1 and top-1% retrieval accuracies. Furthermore, we evaluate the generalization of the proposed method for urban landscapes on our newly collected cross-view localization dataset with geo-reference information.

Few-shot or one-shot learning of classifiers requires a significant inductive bias towards the type of task to be learned. One way to acquire this is by meta-learning on tasks similar to the target task. In this paper, we propose UMTRA, an algorithm that performs unsupervised, model-agnostic meta-learning for classification tasks. The meta-learning step of UMTRA is performed on a flat collection of unlabeled images. While we assume that these images can be grouped into a diverse set of classes and are relevant to the target task, no explicit information about the classes or any labels are needed. UMTRA uses random sampling and augmentation to create synthetic training tasks for meta-learning phase. Labels are only needed at the final target task learning step, and they can be as little as one sample per class. On the Omniglot and Mini-Imagenet few-shot learning benchmarks, UMTRA outperforms every tested approach based on unsupervised learning of representations, while alternating for the best performance with the recent CACTUs algorithm. Compared to supervised model-agnostic meta-learning approaches, UMTRA trades off some classification accuracy for a reduction in the required labels of several orders of magnitude.

This paper proposes a person-centric and online approach to the challenging problem of localization and prediction of actions and interactions in videos. Typically, localization or recognition is performed in an offline manner where all the frames in the video are processed together. This prevents timely localization and prediction of actions and interactions - an important consideration for many tasks including surveillance and human-machine interaction. In our approach, we estimate human poses at each frame and train discriminative appearance models using the superpixels inside the pose bounding boxes. Since the pose estimation per frame is inherently noisy, the conditional probability of pose hypotheses at current time-step (frame) is computed using pose estimations in the current frame and their consistency with poses in the previous frames. Next, both the superpixel and pose-based foreground likelihoods are used to infer the location of actors at each time through a Conditional Random Field enforcing spatio-temporal smoothness in color, optical flow, motion boundaries and edges among superpixels. The issue of visual drift is handled by updating the appearance models, and refining poses using motion smoothness on joint locations, in an online manner. For online prediction of action/interaction confidences, we propose an approach based on Structural SVM that operates on short video segments, and is trained with the objective that confidence of an action or interaction increases as time passes in a positive training clip. Lastly, we quantify the performance of both detection and prediction together, and analyze how the prediction accuracy varies as a time function of observed action/interaction at different levels of detection performance. Our experiments on several datasets suggest that despite using only a few frames to localize actions/interactions at each time instant, we are able to obtain competitive results to state-of-the-art offline methods.

In this work, we propose a tracker that differs from most existing multi-target trackers in two major ways. First, our tracker does not rely on a pre-trained object detector to get the initial object hypotheses. Second, our tracker's final output is the fine contours of the targets rather than traditional bounding boxes. Therefore, our tracker simultaneously solves three main problems: detection, data association and segmentation. This is especially important because the output of each of those three problems are highly correlated and the solution of one can greatly help improve the others. The proposed algorithm consists of two main components: structured learning and Lagrange dual decomposition. Our structured learning based tracker learns a model for each target and infers the best locations of all targets simultaneously in a video clip. The inference of our structured learning is achieved through a new Target Identity-aware Network Flow (TINF), where each node in the network encodes the probability of each target identity belonging to that node. The probabilities are obtained by training target specific models using a global structured learning technique. This is followed by proposed Lagrangian relaxation optimization to find the high quality solution to the network. This forms the first component of our tracker. The second component is Lagrange dual decomposition, which combines the structured learning tracker with a segmentation algorithm. For segmentation, multi-label Conditional Random Field (CRF) is applied to a superpixel based spatio-temporal graph in a segment of video, in order to assign background or target labels to every superpixel. We show how the multi-label CRF is combined with the structured learning tracker through our dual decomposition formulation. This leads to more accurate segmentation results and also helps better resolve typical difficulties in multiple target tracking, such as occlusion handling, ID-switch and track drifting. The experiments on diverse and challenging sequences show that our method achieves superior results compared to competitive approaches for detection, multiple target tracking as well as segmentation.

This paper presents a new approach for the challenging problem of geo-localization using image matching in a structured database of city-wide reference images with known GPS coordinates. We cast the geo-localization as a clustering problem of local image features. Akin to existing approaches to the problem, our framework builds on low-level features which allow local matching between images. For each local feature in the query image, we find its approximate nearest neighbors in the reference set. Next, we cluster the features from reference images using Dominant Set clustering, which affords several advantages over existing approaches. First, it permits variable number of nodes in the cluster, which we use to dynamically select the number of nearest neighbors for each query feature based on its discrimination value. Second, this approach is several orders of magnitude faster than existing approaches. Thus, we obtain multiple clusters (different local maximizers) and obtain a robust final solution to the problem using multiple weak solutions through constrained Dominant Set clustering on global image features, where we enforce the constraint that the query image must be included in the cluster. This second level of clustering also bypasses heuristic approaches to voting and selecting the reference image that matches to the query. We evaluate the proposed framework on an existing dataset of 102k street view images as well as a new larger dataset of 300k images, and show that it outperforms the state-of-the-art by 20 and 7 percent, respectively, on the two datasets.

Person re-identification is a challenging task mainly due to factors such as background clutter, pose, illumination and camera point of view variations. These elements hinder the process of extracting robust and discriminative representations, hence preventing different identities from being successfully distinguished. To improve the representation learning, usually local features from human body parts are extracted. However, the common practice for such a process has been based on bounding box part detection. In this paper, we propose to adopt human semantic parsing which, due to its pixel-level accuracy and capability of modeling arbitrary contours, is naturally a better alternative. Our proposed SPReID integrates human semantic parsing in person re-identification and not only considerably outperforms its counter baseline, but achieves state-of-the-art performance. We also show that, by employing a simple yet effective training strategy, standard popular deep convolutional architectures such as Inception-V3 and ResNet-152, with no modification, while operating solely on full image, can dramatically outperform current state-of-the-art. Our proposed methods improve state-of-the-art person re-identification on: Market-1501 [48] by ~17% in mAP and ~6% in rank-1, CUHK03 [24] by ~4% in rank-1 and DukeMTMC-reID [50] by ~24% in mAP and ~10% in rank-1.

Object detection in wide area motion imagery (WAMI) has drawn the attention of the computer vision research community for a number of years. WAMI proposes a number of unique challenges including extremely small object sizes, both sparse and densely-packed objects, and extremely large search spaces (large video frames). Nearly all state-of-the-art methods in WAMI object detection report that appearance-based classifiers fail in this challenging data and instead rely almost entirely on motion information in the form of background subtraction or frame-differencing. In this work, we experimentally verify the failure of appearance-based classifiers in WAMI, such as Faster R-CNN and a heatmap-based fully convolutional neural network (CNN), and propose a novel two-stage spatio-temporal CNN which effectively and efficiently combines both appearance and motion information to significantly surpass the state-of-the-art in WAMI object detection. To reduce the large search space, the first stage (ClusterNet) takes in a set of extremely large video frames, combines the motion and appearance information within the convolutional architecture, and proposes regions of objects of interest (ROOBI). These ROOBI can contain from one to clusters of several hundred objects due to the large video frame size and varying object density in WAMI. The second stage (FoveaNet) then estimates the centroid location of all objects in that given ROOBI simultaneously via heatmap estimation. The proposed method exceeds state-of-the-art results on the WPAFB 2009 dataset by 5-16% for moving objects and nearly 50% for stopped objects, as well as being the first proposed method in wide area motion imagery to detect completely stationary objects.

Surveillance videos are able to capture a variety of realistic anomalies. In this paper, we propose to learn anomalies by exploiting both normal and anomalous videos. To avoid annotating the anomalous segments or clips in training videos, which is very time consuming, we propose to learn anomaly through the deep multiple instance ranking framework by leveraging weakly labeled training videos, i.e. the training labels (anomalous or normal) are at video-level instead of clip-level. In our approach, we consider normal and anomalous videos as bags and video segments as instances in multiple instance learning (MIL), and automatically learn a deep anomaly ranking model that predicts high anomaly scores for anomalous video segments. Furthermore, we introduce sparsity and temporal smoothness constraints in the ranking loss function to better localize anomaly during training. We also introduce a new large-scale first of its kind dataset of 128 hours of videos. It consists of 1900 long and untrimmed real-world surveillance videos, with 13 realistic anomalies such as fighting, road accident, burglary, robbery, etc. as well as normal activities. This dataset can be used for two tasks. First, general anomaly detection considering all anomalies in one group and all normal activities in another group. Second, for recognizing each of 13 anomalous activities. Our experimental results show that our MIL method for anomaly detection achieves significant improvement on anomaly detection performance as compared to the state-of-the-art approaches. We provide the results of several recent deep learning baselines on anomalous activity recognition. The low recognition performance of these baselines reveals that our dataset is very challenging and opens more opportunities for future work. The dataset is available at: http://crcv.ucf.edu/projects/real-world.

Studying human brain signals has always gathered great attention from the scientific community. In Brain Computer Interface (BCI) research, for example, changes of brain signals in relation to specific tasks (e.g., thinking something) are detected and used to control machines. While extracting spatio-temporal cues from brain signals for classifying state of human mind is an explored path, decoding and visualizing brain states is new and futuristic. Following this latter direction, in this paper, we propose an approach that is able not only to read the mind, but also to decode and visualize human thoughts. More specifically, we analyze brain activity, recorded by an ElectroEncephaloGram (EEG), of a subject while thinking about a digit, character or an object and synthesize visually the thought item. To accomplish this, we leverage the recent progress of adversarial learning by devising a conditional Generative Adversarial Network (GAN), which takes, as input, encoded EEG signals and generates corresponding images. In addition, since collecting large EEG signals in not trivial, our GAN model allows for learning distributions with limited training data. Performance analysis carried out on three different datasets -- brain signals of multiple subjects thinking digits, characters, and objects -- show that our approach is able to effectively generate images from thoughts of a person. They also demonstrate that EEG signals encode explicitly cues from thoughts which can be effectively used for generating semantically relevant visualizations.

The recent advances in Deep Convolutional Neural Networks (DCNNs) have shown extremely good results for video human action classification, however, action detection is still a challenging problem. The current action detection approaches follow a complex pipeline which involves multiple tasks such as tube proposals, optical flow, and tube classification. In this work, we present a more elegant solution for action detection based on the recently developed capsule network. We propose a 3D capsule network for videos, called VideoCapsuleNet: a unified network for action detection which can jointly perform pixel-wise action segmentation along with action classification. The proposed network is a generalization of capsule network from 2D to 3D, which takes a sequence of video frames as input. The 3D generalization drastically increases the number of capsules in the network, making capsule routing computationally expensive. We introduce capsule-pooling in the convolutional capsule layer to address this issue and make the voting algorithm tractable. The routing-by-agreement in the network inherently models the action representations and various action characteristics are captured by the predicted capsules. This inspired us to utilize the capsules for action localization and the class-specific capsules predicted by the network are used to determine a pixel-wise localization of actions. The localization is further improved by parameterized skip connections with the convolutional capsule layers and the network is trained end-to-end with a classification as well as localization loss. The proposed network achieves state-of-the-art performance on multiple action detection datasets including UCF-Sports, J-HMDB, and UCF-101 (24 classes) with an impressive ~20% improvement on UCF-101 and ~15% improvement on J-HMDB in terms of v-mAP scores.

Multi-object tracking has been studied for decades. However, when it comes to tracking pedestrians in extremely crowded scenes, we are limited to only few works. This is an important problem which gives rise to several challenges. Pre-trained object detectors fail to localize targets in crowded sequences. This consequently limits the use of data-association based multi-target tracking methods which rely on the outcome of an object detector. Additionally, the small apparent target size makes it challenging to extract features to discriminate targets from their surroundings. Finally, the large number of targets greatly increases computational complexity which in turn makes it hard to extend existing multi-target tracking approaches to high-density crowd scenarios. In this paper, we propose a tracker that addresses the aforementioned problems and is capable of tracking hundreds of people efficiently. We formulate online crowd tracking as Binary Quadratic Programing. Our formulation employs target's individual information in the form of appearance and motion as well as contextual cues in the form of neighborhood motion, spatial proximity and grouping, and solves detection and data association simultaneously. In order to solve the proposed quadratic optimization efficiently, where state-of art commercial quadratic programing solvers fail to find the solution in a reasonable amount of time, we propose to use the most recent version of the Modified Frank Wolfe algorithm, which takes advantage of SWAP-steps to speed up the optimization. We show that the proposed formulation can track hundreds of targets efficiently and improves state-of-art results by significant margins on eleven challenging high density crowd sequences.

Recognizing human actions from unknown and unseen (novel) views is a challenging problem. We propose a Robust Non-Linear Knowledge Transfer Model (R-NKTM) for human action recognition from novel views. The proposed R-NKTM is a deep fully-connected neural network that transfers knowledge of human actions from any unknown view to a shared high-level virtual view by finding a set of non-linear transformations that connects the views. The R-NKTM is learned from 2D projections of dense trajectories of synthetic 3D human models fitted to real motion capture data and generalizes to real videos of human actions. The strength of our technique is that we learn a single R-NKTM for all actions and all viewpoints for knowledge transfer of any real human action video without the need for re-training or fine-tuning the model. Thus, R-NKTM can efficiently scale to incorporate new action classes. R-NKTM is learned with dummy labels and does not require knowledge of the camera viewpoint at any stage. Experiments on three benchmark cross-view human action datasets show that our method outperforms existing state-of-the-art.

Effective and efficient video retrieval has become a pressing need in the “big video” era. The objective of this work is to provide a principled model for computing the ranking scores of a video in response to one or more concepts, where the concepts could be directly supplied by users or inferred by the system from the user queries. Indeed, how to deal with multi-concept queries has become a central component in modern video retrieval systems that accept text queries. However, it has been long overlooked and simply implemented by weighted averaging of the corresponding concept detectors’ scores. Our approach, which can be considered as a latent ranking SVM, integrates the advantages of various recent works in text and image retrieval, such as choosing ranking over structured prediction, modeling inter-dependencies between querying concepts, and so on. Videos consist of shots, and we use latent variables to account for the mutually complementary cues within and across shots. Concept labels of shots are scarce and noisy. We introduce a simple and effective technique to make our model robust to outliers. Our approach gives superior performance when it is tested on not only the queries seen at training but also novel queries, some of which consist of more concepts than the queries used for training.

xWhat if we could effectively read the mind and transfer human visual capabilities to computer vision methods? In this paper, we aim at addressing this question by developing the first visual object classifier driven by human brain signals. In particular, we employ EEG data evoked by visual object stimuli combined with Recurrent Neural Networks (RNN) to learn a discriminative brain activity manifold of visual categories in a reading the mind effort. Afterward, we transfer the learned capabilities to machines by training a Convolutional Neural Network (CNN)-based regressor to project images onto the learned manifold, thus allowing machines to employ human brain-based features for automated visual classification. We use a 128-channel EEG with active electrodes to record brain activity of several subjects while looking at images of 40 ImageNet object classes. The proposed RNN-based approach for discriminating object classes using brain signals reaches an average accuracy of about 83%, which greatly outperforms existing methods attempting to learn EEG visual object representations. As for automated object categorization, our human brain-driven approach obtains competitive performance, comparable to those achieved by powerful CNN models and it is also able to generalize over different visual datasets.

In this paper, we address the problem of cross-view image geo-localization. Specifically, we aim to estimate the GPS location of a query street view image by finding the matching images in a reference database of geo-tagged birds eye view images, or vice versa. To this end, we present a new framework for cross-view image geo-localization by taking advantage of the tremendous success of deep convolutional neural networks (CNNs) in image classification and object detection. First, we employ the Faster R-CNN to detect buildings in the query and reference images. Next, for each building in the query image, we retrieve the k nearest neighbors from the reference buildings using a Siamese network trained on both positive matching image pairs and negative pairs. To find the correct NN for each query building, we develop an efficient multiple nearest neighbors matching method based on dominant sets. We evaluate the proposed framework on a new dataset that consists of pairs of street view and birds eye view images. Experimental results show that the proposed method achieves better geo-localization accuracy than other approaches and is able to generalize to images at unseen locations.

Deep learning has been demonstrated to achieve excellent results for image classification and object detection. However, the impact of deep learning on video analysis has been limited due to complexity of video data and lack of annotations. Previous convolutional neural networks (CNN) based video action detection approaches usually consist of two major steps: frame-level action proposal generation and association of proposals across frames. Also, most of these methods employ two-stream CNN framework to handle spatial and temporal feature separately. In this paper, we propose an end-to-end deep network called Tube Convolutional Neural Network (T-CNN) for action detection in videos. The proposed architecture is a unified deep network that is able to recognize and localize action based on 3D convolution features. A video is first divided into equal length clips and next for each clip a set of tube proposals are generated based on 3D Convolutional Network (ConvNet) features. Finally, the tube proposals of different clips are linked together employing network flow and spatio-temporal action detection is performed using these linked video proposals. Extensive experiments on several video datasets demonstrate the superior performance of T-CNN for classifying and localizing actions in both trimmed and untrimmed videos compared to state-of-the-arts.

Given a video and a description sentence with one missing word, “source sentence”, Video-Fill-In-the-Blank (VFIB) problem is to find the missing word automatically. The contextual information of the sentence, as well as visual cues from the video, are important to infer the missing word accurately. Since the source sentence is broken into two fragments: the sentence's left fragment (before the blank) and the sentence's right fragment (after the blank), traditional Recurrent Neural Networks cannot encode this structure accurately because of many possible variations of the missing word in terms of the location and type of the word in the source sentence. For example, a missing word can be the first word or be in the middle of the sentence and it can be a verb or an adjective. In this paper, we propose a framework to tackle the textual encoding: Two separate LSTMs (the LR and RL LSTMs) are employed to encode the left and right sentence fragments and a novel structure is introduced to combine each fragment with an external memory corresponding to the opposite fragments. For the visual encoding, end-to-end spatial and temporal attention models are employed to select discriminative visual representations to find the missing word. In the experiments, we demonstrate the superior performance of the proposed method on challenging VFIB problem. Furthermore, we introduce an extended and more generalized version of VFIB, which is not limited to a single blank. Our experiments indicate the generalization capability of our method in dealing with such more realistic scenarios.

Recent advancements in generative adversarial networks (GANs), using deep convolutional models, have supported the development of image generation techniques able to reach satisfactory levels of realism. Further improvements have been proposed to condition GANs to generate images matching a specific object category or a short text description. In this work, we build on the latter class of approaches and investigate the possibility of driving and conditioning the image generation process by means of brain signals recorded, through an electroencephalograph (EEG), while users look at images from a set of 40 ImageNet object categories with the objective of generating the seen images. To accomplish this task, we first demonstrate that brain activity EEG signals encode visually-related information that allows us to accurately discriminate between visual object categories and, accordingly, we extract a more compact class-dependent representation of EEG data using recurrent neural networks. Afterwards, we use the learned EEG manifold to condition image generation employing GANs, which, during inference, will read EEG signals and convert them into images. We tested our generative approach using EEG signals recorded from six subjects while looking at images of the aforementioned 40 visual classes. The results show that for classes represented by well-defined visual patterns (e.g., pandas, airplane, etc.), the generated images are realistic and highly resemble those evoking the EEG signals used for conditioning GANs, resulting in an actual reading-the-mind process.

This paper is the first to address the problem of unsupervised action localization in videos. Given unlabeled data without bounding box annotations, we propose a novel approach that: 1) Discovers action class labels and 2) Spatio-temporally localizes actions in videos. It begins by computing local video features to apply spectral clustering on a set of unlabeled training videos. For each cluster of videos, an undirected graph is constructed to extract a dominant set, which are known for high internal homogeneity and in-homogeneity between vertices outside it. Next, a discriminative clustering approach is applied, by training a classifier for each cluster, to iteratively select videos from the non-dominant set and obtain complete video action classes. Once classes are discovered, training videos within each cluster are selected to perform automatic spatio-temporal annotations, by first over-segmenting videos in each discovered class into supervoxels and constructing a directed graph to apply a variant of knapsack problem with temporal constraints. Knapsack optimization jointly collects a subset of supervoxels, by enforcing the annotated action to be spatio-temporally connected and its volume to be the size of an actor. These annotations are used to train SVM action classifiers. During testing, actions are localized using a similar Knapsack approach, where supervoxels are grouped together and SVM, learned using videos from discovered action classes, is used to recognize these actions. We evaluate our approach on UCF-Sports, Sub-JHMDB, JHMDB, THUMOS13 and UCF101 datasets. Our experiments suggest that despite using no action class labels and no bounding box annotations, we are able to get competitive results to the state-of-the-art supervised methods.

Semantic segmentation has been a long standing challenging task in computer vision. It aims at assigning a label to each image pixel and needs a significant number of pixel-level annotated data, which is often unavailable. To address this lack of annotations, in this paper, we leverage, on one hand, a massive amount of available unlabeled or weakly labeled data, and on the other hand, non-real images created through Generative Adversarial Networks. In particular, we propose a semi-supervised framework - based on Generative Adversarial Networks (GANs) - which consists of a generator network to provide extra training examples to a multi-class classifier, acting as discriminator in the GAN framework, that assigns sample a label y from the K possible classes or marks it as a fake sample (extra class). The underlying idea is that adding large fake visual data forces real samples to be close in the feature space, which, in turn, improves multiclass pixel classification. To ensure a higher quality of generated images by GANs with consequently improved pixel classification, we extend the above framework by adding weakly annotated data, i.e., we provide class level information to the generator. We test our approaches on several challenging benchmarking visual datasets, i.e. PASCAL, SiftFLow, Stanford and CamVid, achieving competitive performance compared to state-of-the-art semantic segmentation methods.

Reading the human mind has been a hot topic in the last decades, and recent research in neuroscience has found evidence on the possibility of decoding, from neuroimaging data, how the human brain works. At the same time, the recent rediscovery of deep learning combined to the large interest of scientific community on generative methods has enabled the generation of realistic images by learning a data distribution from noise. The quality of generated images increases when the input data conveys information on visual content of images. Leveraging on these recent trends, in this paper we present an approach for generating images using visually-evoked brain signals recorded through an electroencephalograph (EEG). More specifically, we recorded EEG data from several subjects while observing images on a screen and tried to regenerate the seen images. To achieve this goal, we developed a deep-learning framework consisting of an LSTM stacked with a generative method, which learns a more compact and noise-free representation of EEG data and employs it to generate the visual stimuli evoking specific brain responses.

This paper proposes a novel approach to tackle the challenging problem of 'online action localization' which entails predicting actions and their locations as they happen in a video. Typically, action localization or recognition is performed in an offline manner where all the frames in the video are processed together and action labels are not predicted for the future. This disallows timely localization of actions - an important consideration for surveillance tasks. In our approach, given a batch of frames from the immediate past in a video, we estimate pose and oversegment the current frame into superpixels. Next, we discriminatively train an actor foreground model on the superpixels using the pose bounding boxes. A Conditional Random Field with superpixels as nodes, and edges connecting spatio-temporal neighbors is used to obtain action segments. The action confidence is predicted using dynamic programming on SVM scores obtained on short segments of the video, thereby capturing sequential information of the actions. The issue of visual drift is handled by updating the appearance model and pose refinement in an online manner. Lastly, we introduce a new measure to quantify the performance of action prediction (i.e. online action localization), which analyzes how the prediction accuracy varies as a function of observed portion of the video. Our experiments suggest that despite using only a few frames to localize actions at each time instant, we are able to predict the action and obtain competitive results to state-of-the-art offline methods.

Scene labeling task is to segment the image into meaningful regions and categorize them into classes of objects which comprised the image. Commonly used methods typically find the local features for each segment and label them using classifiers. Afterwards, labeling is smoothed in order to make sure that neighboring regions receive similar labels. However, these methods ignore expressive connections between labels and non-local dependencies among regions. In this paper, we propose to use a sparse estimation of precision matrix (also called concentration matrix), which is the inverse of covariance matrix of data obtained by graphical lasso to find interaction between labels and regions. To do this, we formulate the problem as an energy minimization over a graph, whose structure is captured by applying sparse constraint on the elements of the precision matrix. This graph encodes (or represents) only significant interactions and avoids a fully connected graph, which is typically used to reflect the long distance associations. We use local and global information to achieve better labeling. We assess our approach on three datasets and obtained promising results.

This paper tackles the problem of spatio-temporal action localization in a video, without assuming the availability of multiple videos or any prior annotations. Action is localized by employing images downloaded from internet using action name. Given web images, we first dampen image noise using random walk and evade distracting backgrounds within images using image action proposals. Then, given a video, we generate multiple spatio-temporal action proposals. We suppress camera and background generated proposals by exploiting optical flow gradients within proposals. To obtain the most action representative proposals, we propose to reconstruct action proposals in the video by leveraging the action proposals in images. Moreover, we preserve the temporal smoothness of the video and reconstruct all proposal bounding boxes jointly using the constraints that push the coefficients for each bounding box toward a common consensus, thus enforcing the coefficient similarity across multiple frames. We solve this optimization problem using variant of two-metric projection algorithm. Finally, the video proposal that has the lowest reconstruction cost and is motion salient is used to localize the action. Our method is not only applicable to the trimmed videos, but it can also be used for action localization in untrimmed videos, which is a very challenging problem. We present extensive experiments on trimmed as well as untrimmed datasets to validate the effectiveness of the proposed approach.

The well-known word analogy experiments show that the recent word vectors capture fine-grained linguistic regularities in words by linear vector offsets, but it is unclear how well the simple vector offsets can encode visual regularities over words. We study a particular image-word relevance relation in this paper. Our results show that the word vectors of relevant tags for a given image rank ahead of the irrelevant tags, along a principal direction in the word vector space. Inspired by this observation, we propose to solve image tagging by estimating the principal direction for an image. Particularly, we exploit linear mappings and nonlinear deep neural networks to approximate the principal direction from an input image. We arrive at a quite versatile tagging model. It runs fast given a test image, in constant time w.r.t. the training set size. It not only gives superior performance for the conventional tagging task on the NUSWIDE dataset, but also outperforms competitive baselines on annotating images with previously unseen tags.

Effective and efficient video retrieval has become a pressing need in the "big video" era and how to deal with multi-concept queries is a central component. The objective of this work is to provide a principled model for calculating the ranking scores of video in response to multiple concepts. However, it has been long overlooked and simply implemented by weighted averaging the corresponding concept detectors' scores. Our approach, which can be considered as a latent ranking SVM, integrates the advantages of various recent works on text and image retrieval, such as choosing ranking over structured prediction and modeling inter-dependencies between querying concepts and the others. Videos consist of shots and we use latent variables to account for the mutually complementary cues within and across shots. We introduce a simple and effective way to make our model robust to outliers and scarce data. Our approach gives rise to superior performance when it is tested on not only the queries seen at training, but also novel queries, some of which consist of more concepts than the seen queries used for training.

We describe a system for content-based retrieval from large surveillance video archives, using behavior, action and appearance of objects. Objects are detected, tracked, and classified into broad categories. Their behavior and appearance are characterized by action detectors and descriptors, which are indexed in an archive. Queries can be posed as video exemplars, and the results can be refined through relevance feedback. The contributions of our system include the fusion of behavior and action detectors with appearance for matching; the improvement of query results through interactive query refinement (IQR), which learns a discriminative classifier online based on user feedback; and reasonable performance on low resolution, poor quality video. The system operates on video from ground cameras and aerial platforms, both RGB and IR. Performance is evaluated on publicly-available surveillance datasets, showing that subtle actions can be detected under difficult conditions, with reasonable improvement from IQR.

The availability of GIS (Geographical Information System) databases for many urban areas, provides a valuable source of information for improving the performance of many computer vision tasks. In this paper, we propose a method which leverages information acquired from GIS databases to perform semantic segmentation of the image alongside with geo-referencing each semantic segment with its address and geo-location. First, the image is segmented into a set of initial super-pixels. Then, by projecting the information from GIS databases, a set of priors are obtained about the approximate location of the semantic entities such as buildings and streets in the image plane. However, there are significant inaccuracies (misalignments) in the projections, mainly due to inaccurate GPS-tags and camera parameters. In order to address this misalignment issue, we perform data fusion such that it improves the segmentation and GIS projections accuracy simultaneously with an iterative approach. At each iteration, the projections are evaluated and weighted in terms of reliability, and then fused with the super-pixel segmentations. First segmentation is performed using random walks, based on the GIS projections. Then the global transformation which best aligns the projections to their corresponding semantic entities is computed and applied to the projections to further align them to the content of the image. The iterative approach continues until the projections and segments are well aligned.

Data association is the backbone to many multiple object tracking (MOT) methods. In this paper we formulate data association as a Generalized Maximum Multi Clique problem (GMMCP). We show that this is the ideal case of modeling tracking in real world scenario where all the pairwise relationships between targets in a batch of frames are taken into account. Previous works assume simplified version of our tracker either in problem formulation or problem optimization. However, we propose a solution using GMMCP where no simplification is assumed in either steps. We show that the NP hard problem of GMMCP can be formulated through Binary-Integer Program where for small and medium size MOT problems the solution can be found efficiently. We further propose a speed-up method, employing Aggregated Dummy Nodes for modeling occlusion and miss-detection, which reduces the size of the input graph without using any heuristics. We show that, using the speedup method, our tracker lends itself to real-time implementation which is plausible in many applications. We evaluated our tracker on six challenging sequences of Town Center, TUD-Crossing, TUD-Stadtmitte, Parking-lot 1, Parking-lot 2 and Parking-lot pizza and show favorable improvement against state of art.

In this paper we show that multiple object tracking (MOT) can be formulated in a framework, where the detection and data-association are performed simultaneously. Our method allows us to overcome the confinements of data association based MOT approaches; where the performance is dependent on the object detection results provided at input level. At the core of our method lies structured learning which learns a model for each target and infers the best location of all targets simultaneously in a video clip. The inference of our structured learning is done through a new Target Identity-aware Network Flow (TINF), where each node in the network encodes the probability of each target identity belonging to that node. The proposed Lagrangian relaxation optimization finds the high quality solution to the network. During optimization a soft spatial constraint is enforced between the nodes of the graph which helps reducing the ambiguity caused by nearby targets with similar appearance in crowded scenarios. We show that automatically detecting and tracking targets in a single framework can help resolve the ambiguities due to frequent occlusion and heavy articulation of targets. Our experiments involve challenging yet distinct datasets and show that our method can achieve results better than the state-of-art.

This paper presents an efficient approach for localizing actions by learning contextual relations, in the form of relative locations between different video regions. We begin by over-segmenting the videos into supervoxels, which have the ability to preserve action boundaries and also reduce the complexity of the problem. Context relations are learned during training which capture displacements from all the supervoxels in a video to those belonging to foreground actions. Then, given a testing video, we select a supervoxel randomly and use the context information acquired during training to estimate the probability of each supervoxel belonging to the foreground action. The walk proceeds to a new supervoxel and the process is repeated for a few steps. This "context walk" generates a conditional distribution of an action over all the supervoxels. A Conditional Random Field is then used to find action proposals in the video, whose confidences are obtained using SVMs. We validated the proposed approach on several datasets and show that context in the form of relative displacements between supervoxels can be extremely useful for action localization. This also results in significantly fewer evaluations of the classifier, in sharp contrast to the alternate sliding window approaches.

In this paper, we present a method to estimate a sequence of human poses in unconstrained videos. In contrast to the commonly employed graph optimization framework, which is NP-hard and needs approximate solutions, we formulate this problem into a unified two stage tree-based optimization problem for which an efficient and exact solution exists. Although the proposed method finds an exact solution, it does not sacrifice the ability to model the spatial and temporal constraints between body parts in the video frames, indeed it even models the symmetric parts better than the existing methods. The proposed method is based on two main ideas: 'Abstraction' and 'Association' to enforce the intra-and inter-frame body part constraints respectively without inducing extra computational complexity to the polynomial time solution. Using the idea of 'Abstraction', a new concept of 'abstract body part' is introduced to model not only the tree based body part structure similar to existing methods, but also extra constraints between symmetric parts. Using the idea of 'Association', the optimal tracklets are generated for each abstract body part, in order to enforce the spatiotemporal constraints between body parts in adjacent frames. Finally, a sequence of the best poses is inferred from the abstract body part tracklets through the tree-based optimization. We evaluated the proposed method on three publicly available video based human pose estimation datasets, and obtained dramatically improved performance compared to the state-of-the-art methods.

Selfies are now a global phenomenon. This massive number of self-portrait images taken and shared on social media is revolutionizing the way people introduce themselves and the circle of their friends to the world. While taking photos of oneself can be seen simply as recording personal memories, the urge to share them with other people adds an exclusive sensation to the selfies. Due to the Big Data nature of selfies, it is nearly impossible to analyze them manually. In this paper, we provide, to the best of our knowledge, the first selfie dataset for research purposes with more than 46,000 images. We address interesting questions about selfies, including how appearance of certain objects, concepts and attributes influences the popularity of selfies. We also study the correlation between popularity and sentiment in selfie images. In a nutshell, from a large scale dataset, we automatically infer what makes a selfie a good selfie. We believe that this research creates new opportunities for social, psychological and behavioral scientists to study selfies from a large scale point of view, a perspective that best fits the nature of the selfie phenomenon.

This paper presents a novel search paradigm that uses multiple images as input to perform semantic search of images. While earlier focuses on using single or multiple query images to retrieve images with views of the same instance, the proposed paradigm uses each query image to discover text-based descriptors that are leveraged to find the common concepts that are implicitly shared by all of the query images and retrieves images considering the found concepts. Our implementation uses high level visual features extracted from a deep convolutional network to retrieve images similar to each query input. These images have associated text previously generated by implicit crowdsourcing. A Bag of Words (BoW) textual representation of each query image is built from the associated text of the retrieved similar images. A learned vector space representation of English words extracted from a corpus of 100 billion words allows computing the conceptual similarity of words. The words that represent the input images are used to find new words that share conceptual similarity across all the input images. These new words are combined with the representations of the input images to obtain a BoW textual representation of the search, which is used to perform image retrieval. The retrieved images are re-ranked to enhance visual similarity with respect to any of the input images. Our experiments show that the concepts found are meaningful and that they retrieve correctly 72.43% of the images from the top 25, along with user ratings performed in the cases of study.

Human detection in dense crowds is an important problem, as it is a prerequisite to many other visual tasks, such as tracking, counting, action recognition or anomaly detection in behaviors exhibited by individuals in a dense crowd. This problem is challenging due to the large number of individuals, small apparent size, severe occlusions and perspective distortion. However, crowded scenes also offer contextual constraints that can be used to tackle these challenges. In this paper, we explore context for human detection in dense crowds in the form of a locally-consistent scale prior which captures the similarity in scale in local neighborhoods and its smooth variation over the image. Using the scale and confidence of detections obtained from an underlying human detector, we infer scale and confidence priors using Markov Random Field. In an iterative mechanism, the confidences of detection hypotheses are modified to reflect consistency with the inferred priors, and the priors are updated based on the new detections. The final set of detections obtained are then reasoned for occlusion using Binary Integer Programming where overlaps and relations between parts of individuals are encoded as linear constraints. Both human detection and occlusion reasoning in proposed approach are solved with local neighbor-dependent constraints, thereby respecting the inter-dependence between individuals characteristic to dense crowd analysis. In addition, we propose a mechanism to detect different combinations of body parts without requiring annotations for individual combinations. We performed experiments on a new and extremely challenging dataset of dense crowd images showing marked improvement over the underlying human detector.

We address the problem of classifying complex videos based on their content. A typical approach to this problem is performing the classification using semantic attributes, commonly termed concepts, which occur in the video. In this paper, we propose a contextual approach to video classification based on Generalized Maximum Clique Problem (GMCP) which uses the co-occurrence of concepts as the context model. To be more specific, we propose to represent a class based on the co-occurrence of its concepts and classify a video based on matching its semantic co-occurrence pattern to each class representation. We perform the matching using GMCP which finds the strongest clique of co-occurring concepts in a video. We argue that, in principal, the co-occurrence of concepts yields a richer representation of a video compared to most of the current approaches. Additionally, we propose a novel optimal solution to GMCP based on Mixed Binary Integer Programming (MBIP). The evaluations show our approach, which opens new opportunities for further research in this direction, outperforms several well established video classification methods.

While approaches based on bags of features excel at low-level action classification, they are ill-suited for recognizing complex events in video, where concept-based temporal representations currently dominate. This paper proposes a novel representation that captures the temporal dynamics of windowed mid-level concept detectors in order to improve complex event recognition. We first express each video as an ordered vector time series, where each time step consists of the vector formed from the concatenated confidences of the pre-trained concept detectors. We hypothesize that the dynamics of time series for different instances from the same event class, as captured by simple linear dynamical system (LDS) models, are likely to be similar even if the instances differ in terms of low-level visual features. We propose a two-part representation composed of fusing: (1) a singular value decomposition of block Hankel matrices (SSID-S) and (2) a harmonic signature (HS) computed from the corresponding eigen-dynamics matrix. The proposed method offers several benefits over alternate approaches: our approach is straightforward to implement, directly employs existing concept detectors and can be plugged into linear classification frameworks. Results on standard datasets such as NIST's TRECVID Multimedia Event Detection task demonstrate the improved accuracy of the proposed method.

A video captures a sequence and interactions of concepts that can be static, for instance, objects or scenes, or dynamic, such as actions. For large datasets containing hundreds of thousands of images or videos, it is impractical to manually annotate all the concepts, or all the instances of a single concept. However, a dictionary with visually-distinct elements can be created automatically from unlabeled videos which can capture and express the entire dataset. The downside to this machine-discovered dictionary is meaninglessness, i.e., its elements are devoid of semantics and interpretation. In this paper, we present an approach that leverages the strengths of semantic concepts and the machine-discovered DOVE by learning a relationship between them. Since instances of a semantic concept share visual similarity, the proposed approach uses soft-consensus regularization to learn the mapping that enforces instances from each semantic concept to have similar representations. The testing is performed by projecting the query onto the DOVE as well as new representations of semantic concepts from training, with non-negativity and unit summation constraints for probabilistic interpretation. We tested our formulation on TRECVID MED and SIN tasks, and obtained encouraging results.

Recent years have seen a major push for face recognition technology due to the large expansion of image sharing on social networks. In this paper, we consider the difficult task of determining parent-offspring resemblance using deep learning to answer the question "Who do I look like?" Although humans can perform this job at a rate higher than chance, it is not clear how they do it [2]. However, recent studies in anthropology [24] have determined which features tend to be the most discriminative. In this study, we aim to not only create an accurate system for resemblance detection, but bridge the gap between studies in anthropology with computer vision techniques. Further, we aim to answer two key questions: 1) Do offspring resemble their parents? and 2) Do offspring resemble one parent more than the other? We propose an algorithm that fuses the features and metrics discovered via gated autoencoders with a discriminative neural network layer that learns the optimal, or what we call genetic, features to delineate parent-offspring relationships. We further analyze the correlation between our automatically detected features and those found in anthropological studies. Meanwhile, our method outperforms the state-of-the-art in kinship verification by 3-10% depending on the relationship using specific (father-son, mother-daughter, etc.) and generic models.

The real world image databases such as Flickr are characterized by continuous addition of new images. The recent approaches for image annotation, i.e. the problem of assigning tags to images, have two major drawbacks. First, either models are learned using the entire training data, or to handle the issue of dataset imbalance, tag-specific discriminative models are trained. Such models become obsolete and require relearning when new images and tags are added to database. Second, the task of feature-fusion is typically dealt using ad-hoc approaches. In this paper, we present a weighted extension of Multi-view Non-negative Matrix Factorization (NMF) to address the aforementioned drawbacks. The key idea is to learn query-specific generative model on the features of nearest-neighbors and tags using the proposed NMF-KNN approach which imposes consensus constraint on the coefficient matrices across different features. This results in coefficient vectors across features to be consistent and, thus, naturally solves the problem of feature fusion, while the weight matrices introduced in the proposed formulation alleviate the issue of dataset imbalance. Furthermore, our approach, being query-specific, is unaffected by addition of images and tags in a database. We tested our method on two datasets used for evaluation of image annotation and obtained competitive results.

The number of GPS-tagged images available on the web is increasing at a rapid rate. The majority of such location tags are specified by the users, either through manual tagging or localization-chips embedded in the cameras. However, a known issue with user shared images is the unreliability of such GPS-tags. In this paper, we propose a method for addressing this problem. We assume a large dataset of GPS-tagged images which includes an unknown subset with contaminated tags is available. We develop a robust method for identification and refinement of this subset using the rest of the images in the dataset. In the proposed method, we form a large number of triplets of matching images and use them for estimating the location of the query image utilizing structure from motion. Some of the generated estimations may be inaccurate due to the noisy GPS-tags in the dataset. Therefore, we perform Random Walks on the estimations in order to identify the subset with the maximal agreement. Finally, we estimate the GPS-tag of the query utilizing the identified consistent subset using a weighted mean. We propose a new damping factor for Random Walks which conforms to the level of noise in the input, and consequently, robustifies Random Walks. We evaluated the proposed framework on a dataset of over 18k user-shared images; the experiments show our method robustly improves the accuracy of GPS-tags under diverse scenarios.

Complex event recognition is an expanding research area aiming to recognize entities of high-level semantics in videos. Typical approaches exploit the so-called “bags” of spatiotemporal features such as STIP, ISA and DTF-HOG; yet, more recently, the notion of concept has emerged as an alternative, intermediate representation with greater descriptive power, and “bags of concepts” have been used for recognition. In this paper we argue that concepts in an event tend to articulate over a discernible temporal structure and we exploit a temporal model using the scores of concept detectors as measurements. In addition, we propose several heuristics to improve the initialization of the model's latent states and take advantage of the time-sparsity of the concepts. Experimental results on videos from the challenging TRECVID MED 2012 dataset show that the proposed approach achieves an improvement in average precision of 8.92% over comparable bags of concepts, thus validating the use of temporal structure over concepts for complex event recognition.

Measuring the similarity of human actions in videos is a challenging task. Two critical factors that affect the performance include low-level feature representations and similarity metrics. However, finding the right feature representations and metrics is hard. In this paper, we describe a novel approach that jointly learns both of them from the data, while current approaches either only learn one or not learn at all. We propose a generative plus discriminative learning method based on gated auto encoders to simultaneously learn the features and their associated metrics. Our method differs from existing feature or metric learning methods in two ways: 1) while other methods treat feature learning and metric learning as independent tasks, we argue that they should be learned jointly since features and metrics are tightly inter-dependent; 2) our method learns more discriminative features than its purely generative counterparts.

There is a large variety of trackers, which have been proposed in the literature during the last two decades with some mixed success. Object tracking in realistic scenarios is a difficult problem, therefore, it remains a most active area of research in computer vision. A good tracker should perform well in a large number of videos involving illumination changes, occlusion, clutter, camera motion, low contrast, specularities, and at least six more aspects. However, the performance of proposed trackers have been evaluated typically on less than ten videos, or on the special purpose datasets. In this paper, we aim to evaluate trackers systematically and experimentally on 315 video fragments covering above aspects. We selected a set of nineteen trackers to include a wide variety of algorithms often cited in literature, supplemented with trackers appearing in 2010 and 2011 for which the code was publicly available. We demonstrate that trackers can be evaluated objectively by survival curves, Kaplan Meier statistics, and Grubs testing. We find that in the evaluation practice the F-score is as effective as the object tracking accuracy (OTA) score. The analysis under a large variety of circumstances provides objective insight into the strengths and weaknesses of trackers.

In this paper, we present a new framework for geo-locating an image utilizing a novel multiple nearest neighbor feature matching method using Generalized Minimum Clique Graphs (GMCP). First, we extract local features (e.g., SIFT) from the query image and retrieve a number of nearest neighbors for each query feature from the reference data set. Next, we apply our GMCP-based feature matching to select a single nearest neighbor for each query feature such that all matches are globally consistent. Our approach to feature matching is based on the proposition that the first nearest neighbors are not necessarily the best choices for finding correspondences in image matching. Therefore, the proposed method considers multiple reference nearest neighbors as potential matches and selects the correct ones by enforcing consistency among their global features (e.g., GIST) using GMCP. In this context, we argue that using a robust distance function for finding the similarity between the global features is essential for the cases where the query matches multiple reference images with dissimilar global features. Towards this end, we propose a robust distance function based on the Gaussian Radial Basis Function (G-RBF). We evaluated the proposed framework on a new data set of 102k street view images; the experiments show it outperforms the state of the art by 10 percent.

In this paper, we propose a discriminative representation of a video shot based on its camera motion and demonstrate how the representation can be used for high level multimedia tasks like complex event recognition. In our technique, we assume that a homography exists between a pair of subsequent frames in a given shot. Using purely image-based methods, we compute homography parameters that serve as coarse indicators of the ambient camera motion. Next, using Lie algebra, we map the homography matrices to an intermediate vector space that preserves the intrinsic geometric structure of the transformation. The mappings are stacked temporally to generate vector time-series per shot. To extract meaningful features from time-series, we propose an efficient linear dynamical system based technique. The extracted temporal features are further used to train linear SVMs as classifiers for a particular shot class. In addition to demonstrating the efficacy of our method on a novel dataset, we extend its applicability to recognize complex events in large scale videos under unconstrained scenarios. Our empirical evaluations on eight cinematographic shot classes show that our technique performs close to approaches that involve extraction of 3-D trajectories using computationally prohibitive structure from motion techniques.

We propose to leverage multiple sources of information to compute an estimate of the number of individuals present in an extremely dense crowd visible in a single image. Due to problems including perspective, occlusion, clutter, and few pixels per person, counting by human detection in such images is almost impossible. Instead, our approach relies on multiple sources such as low confidence head detections, repetition of texture elements (using SIFT), and frequency-domain analysis to estimate counts, along with confidence associated with observing individuals, in an image region. Secondly, we employ a global consistency constraint on counts using Markov Random Field. This caters for disparity in counts in local neighborhoods and across scales. We tested our approach on a new dataset of fifty crowd images containing 64K annotated humans, with the head counts ranging from 94 to 4543. This is in stark contrast to datasets used for existing methods which contain not more than tens of individuals. We experimentally demonstrate the efficacy and reliability of the proposed approach by quantifying the counting performance.

This paper presents an end-to-end video face recognition system, addressing the difficult problem of identifying a video face track using a large dictionary of still face images of a few hundred people, while rejecting unknown individuals. A straightforward application of the popular l1-minimization for face recognition on a frame-by-frame basis is prohibitively expensive, so we propose a novel algorithm Mean Sequence SRC (MSSRC) that performs video face recognition using a joint optimization leveraging all of the available video data and the knowledge that the face track frames belong to the same individual. By adding a strict temporal constraint to the l1-minimization that forces individual frames in a face track to all reconstruct a single identity, we show the optimization reduces to a single minimization over the mean of the face track. We also introduce a new Movie Trailer Face Dataset collected from 101 movie trailers on YouTube. Finally, we show that our method matches or outperforms the state-of-the-art on three existing datasets (YouTube Celebrities, YouTube Faces, and Buffy) and our unconstrained Movie Trailer Face Dataset. More importantly, our method excels at rejecting unknown identities by at least 8% in average precision.

We propose an approach to improve the detection performance of a generic detector when it is applied to a particular video. The performance of offline-trained objects detectors are usually degraded in unconstrained video environments due to variant illuminations, backgrounds and camera viewpoints. Moreover, most object detectors are trained using Haar-like features or gradient features but ignore video specific features like consistent color patterns. In our approach, we apply a Super pixel-based Bag-of-Words (BoW) model to iteratively refine the output of a generic detector. Compared to other related work, our method builds a video-specific detector using super pixels, hence it can handle the problem of appearance variation. Most importantly, using Conditional Random Field (CRF) along with our super pixel-based BoW model, we develop and algorithm to segment the object from the background. Therefore our method generates an output of the exact object regions instead of the bounding boxes generated by most detectors. In general, our method takes detection bounding boxes of a generic detector as input and generates the detection output with higher average precision and precise object regions. The experiments on four recent datasets demonstrate the effectiveness of our approach and significantly improves the state-of-art detector by 5-16% in average precision.

Deformable part models have achieved impressive performance for object detection, even on difficult image datasets. This paper explores the generalization of deformable part models from 2D images to 3D spatiotemporal volumes to better study their effectiveness for action detection in video. Actions are treated as spatiotemporal patterns and a deformable part model is generated for each action from a collection of examples. For each action model, the most discriminative 3D sub volumes are automatically selected as parts and the spatiotemporal relations between their locations are learned. By focusing on the most distinctive parts of each action, our models adapt to intra-class variation and show robustness to clutter. Extensive experiments on several video datasets demonstrate the strength of spatiotemporal DPMs for classifying and localizing actions.

We propose a novel approach to boost the performance of generic object detectors on videos by learning video-specific features using a deep neural network. The insight behind our proposed approach is that an object appearing in different frames of a video clip should share similar features, which can be learned to build better detectors. Unlike many supervised detector adaptation or detection-by-tracking methods, our method does not require any extra annotations or utilize temporal correspondence. We start with the high-confidence detections from a generic detector, then iteratively learn new video-specific features and refine the detection scores. In order to learn discriminative and compact features, we propose a new feature learning method using a deep neural network based on auto en-coders. It differs from the existing unsupervised feature learning methods in two ways: first it optimizes both discriminative and generative properties of the features simultaneously, which gives our features better discriminative ability, second, our learned features are more compact, while the unsupervised feature learning methods usually learn a redundant set of over-complete features. Extensive experimental results on person and horse detection show that significant performance improvement can be achieved with our proposed method.

In this paper, we propose a novel approach to extract primary object segments in videos in the `object proposal' domain. The extracted primary object regions are then used to build object models for optimized video segmentation. The proposed approach has several contributions: First, a novel layered Directed Acyclic Graph (DAG) based framework is presented for detection and segmentation of the primary object in video. We exploit the fact that, in general, objects are spatially cohesive and characterized by locally smooth motion trajectories, to extract the primary object from the set of all available proposals based on motion, appearance and predicted-shape similarity across frames. Second, the DAG is initialized with an enhanced object proposal set where motion based proposal predictions (from adjacent frames) are used to expand the set of object proposals for a particular frame. Last, the paper presents a motion scoring function for selection of object proposals that emphasizes high optical flow gradients at proposal boundaries to discriminate between moving objects and the background. The proposed approach is evaluated using several challenging benchmark videos and it outperforms both unsupervised and supervised state-of-the-art methods.

In this paper we propose a novel aesthetic model emphasizing psycho-visual statistics extracted from multiple levels in contrast to earlier approaches that rely only on descriptors suited for image recognition or based on photographic principles. At the lowest level, we determine dark-channel, sharpness and eye-sensitivity statistics over rectangular cells within a frame. At the next level, we extract Sentibank features (1,200 pre-trained visual classifiers) on a given frame, that invoke specific sentiments such as "colorful clouds", "smiling face" etc. and collect the classifier responses as frame-level statistics. At the topmost level, we extract trajectories from video shots. Using viewer's fixation priors, the trajectories are labeled as foreground, and background/camera on which statistics are computed. Additionally, spatio-temporal local binary patterns are computed that capture texture variations in a given shot. Classifiers are trained on individual feature representations independently. On thorough evaluation of 9 different types of features, we select the best features from each level -- dark channel, affect and camera motion statistics. Next, corresponding classifier scores are integrated in a sophisticated low-rank fusion framework to improve the final prediction scores. Our approach demonstrates strong correlation with human prediction on 1,000 broadcast quality videos released by NHK as an aesthetic evaluation dataset.

In this paper we investigate a new problem called visual business recognition. Automatic identification of businesses in images is an interesting task with plenty of potential applications especially for mobile device users. We propose a multimodal approach which incorporates business directories, textual information, and web images in a unified framework. We assume the query image is associated with a coarse location tag and utilize business directories for extracting an over complete list of nearby businesses which may be visible in the image. We use the name of nearby businesses as search keywords in order to automatically collect a set of relevant images from the web and perform image matching between them and the query. Additionally, we employ a text processing method customized for business recognition which is assisted by nearby business names; we fuse the information acquired from image matching and text processing in a probabilistic framework to recognize the businesses. We tested the proposed algorithm on a challenging set of user-uploaded and street view images with promising results for this new application.

Turbulence mitigation refers to the stabilization of videos with nonuniform deformations due to the influence of optical turbulence. Typical approaches for turbulence mitigation follow averaging or dewarping techniques. Although these methods can reduce the turbulence, they distort the independently moving objects, which can often be of great interest. In this paper, we address the novel problem of simultaneous turbulence mitigation and moving object detection. We propose a novel three-term low-rank matrix decomposition approach in which we decompose the turbulence sequence into three components: the background, the turbulence, and the object. We simplify this extremely difficult problem into a minimization of nuclear norm, Frobenius norm, and l1 norm. Our method is based on two observations: First, the turbulence causes dense and Gaussian noise and therefore can be captured by Frobenius norm, while the moving objects are sparse and thus can be captured by l1 norm. Second, since the object's motion is linear and intrinsically different from the Gaussian-like turbulence, a Gaussian-based turbulence model can be employed to enforce an additional constraint on the search space of the minimization. We demonstrate the robustness of our approach on challenging sequences which are significantly distorted with atmospheric turbulence and include extremely tiny moving objects.

This paper proposes a novel representation of articulated human actions and gestures and facial expressions. The main goals of the proposed approach are: 1) to enable recognition using very few examples, i.e., one or k-shot learning, and 2) meaningful organization of unlabeled datasets by unsupervised clustering. Our proposed representation is obtained by automatically discovering high-level subactions or motion primitives, by hierarchical clustering of observed optical flow in four-dimensional, spatial, and motion flow space. The completely unsupervised proposed method, in contrast to state-of-the-art representations like bag of video words, provides a meaningful representation conducive to visual interpretation and textual labeling. Each primitive action depicts an atomic subaction, like directional motion of limb or torso, and is represented by a mixture of four-dimensional Gaussian distributions. For one--shot and k-shot learning, the sequence of primitive labels discovered in a test video are labeled using KL divergence, and can then be represented as a string and matched against similar strings of training videos. The same sequence can also be collapsed into a histogram of primitives or be used to learn a Hidden Markov model to represent classes. We have performed extensive experiments on recognition by one and k-shot learning as well as unsupervised action clustering on six human actions and gesture datasets, a composite dataset, and a database of facial expressions. These experiments confirm the validity and discriminative nature of the proposed representation.

In this paper, we propose a novel method that exploits correlation between audio-visual dynamics of a video to segment and localize objects that are the dominant source of audio. Our approach consists of a two-step spatiotemporal segmentation mechanism that relies on velocity and acceleration of moving objects as visual features. Each frame of the video is segmented into regions based on motion and appearance cues using the QuickShift algorithm, which are then clustered over time using K-means, so as to obtain a spatiotemporal video segmentation. The video is represented by motion features computed over individual segments. The Mel-Frequency Cepstral Coefficients (MFCC) of the audio signal, and their first order derivatives are exploited to represent audio. The proposed framework assumes there is a non-trivial correlation between these audio features and the velocity and acceleration of the moving and sounding objects. The canonical correlation analysis (CCA) is utilized to identify the moving objects which are most correlated to the audio signal. In addition to moving-sounding object identification, the same framework is also exploited to solve the problem of audio-video synchronization, and is used to aid interactive segmentation. We evaluate the performance of our proposed method on challenging videos. Our experiments demonstrate significant increase in performance over the state-of-the-art both qualitatively and quantitatively, and validate the feasibility and superiority of our approach.

Single camera-based multiple-person tracking is often hindered by difficulties such as occlusion and changes in appearance. In this paper, we address such problems by proposing a robust part-based tracking-by-detection framework. Human detection using part models has become quite popular, yet its extension in tracking has not been fully explored. Our approach learns part-based person-specific SVM classifiers which capture the articulations of the human bodies in dynamically changing appearance and background. With the part-based model, our approach is able to handle partial occlusions in both the detection and the tracking stages. In the detection stage, we select the subset of parts which maximizes the probability of detection, which significantly improves the detection performance in crowded scenes. In the tracking stage, we dynamically handle occlusions by distributing the score of the learned person classifier among its corresponding parts, which allows us to detect and predict partial occlusions, and prevent the performance of the classifiers from being degraded. Extensive experiments using the proposed method on several challenging sequences demonstrate state-of-the-art performance in multiple-people tracking.

This paper presents a novel method for estimating the geospatial trajectory of a moving camera with unknown intrinsic parameters, in a city-scale urban environment. The proposed method is based on a three step process that includes: 1) finding the best visual matches of individual images to a dataset of geo-referenced street view images, 2) Bayesian tracking to estimate the frame localization and its temporal evolution, and 3) a trajectory reconstruction algorithm to eliminate inconsistent estimations. As a result of matching features in query image with the features in the reference geo-taged images, in the first step, we obtain a distribution of geolocated votes of matching features which is interpreted as the likelihood of the location (latitude and longitude) given the current observation. In the second step, Bayesian tracking framework is used to estimate the temporal evolution of frame geolocalization based on the previous state probabilities and current likelihood. Finally, once a trajectory is estimated, we perform a Minimum Spanning Trees (MST) based trajectory reconstruction algorithm to eliminate trajectory loops or noisy estimations. The proposed method was tested on sixty minutes of video, which included footage downloaded from YouTube and footage captured by random users in Orlando and Pittsburgh.

method is proposed for identifying five crowd behaviors (bottlenecks, fountainheads, lanes, arches, and blocking) in visual scenes. In the algorithm, a scene is overlaid by a grid of particles initializing a dynamical system defined by the optical flow. Time integration of the dynamical system provides particle trajectories that represent the motion in the scene; these trajectories are used to locate regions of interest in the scene. Linear approximation of the dynamical system provides behavior classification through the Jacobian matrix; the eigenvalues determine the dynamic stability of points in the flow and each type of stability corresponds to one of the five crowd behaviors. The eigenvalues are only considered in the regions of interest, consistent with the linear approximation and the implicated behaviors. The algorithm is repeated over sequential clips of a video in order to record changes in eigenvalues, which may imply changes in behavior. The method was tested on over 60 crowd and traffic videos.

Summary form only given. We present a concept for automatic construction site monitoring by taking into account 4D information (3D over time), that is acquired from highly-overlapping digital aerial images. On the one hand today's maturity of flying micro aerial vehicles (MAVs) enables a low-cost and an efficient image acquisition of high-quality data that maps construction sites entirely from many varying viewpoints. On the other hand, due to low-noise sensors and high redundancy in the image data, recent developments in 3D reconstruction workflows have benefited the automatic computation of accurate and dense 3D scene information. Having both an inexpensive high-quality image acquisition and an efficient 3D analysis workflow enables monitoring, documentation and visualization of observed sites over time with short intervals. Relating acquired 4D site observations, composed of color, texture, geometry over time, largely supports automated methods toward full scene understanding, the acquisition of both the change and the construction site's progress.

In this paper, we present an efficient alternative to the traditional vocabulary based on bag-of-visual words (BoW) used for visual classification tasks. Our representation is both conceptually and computationally superior to the bag-of-visual words: (1) We iteratively generate a Maximum Likelihood estimate of an image given a set of characteristic features in contrast to the BoW methods where an image is represented as a histogram of visual words, (2) We randomly sample a set of characteristic features instead of employing computation-intensive clustering algorithms used during the vocabulary generation step of BoW methods. Our performance compares favorably to the state-of-the-art on experiments over three challenging human action and a scene categorization dataset, demonstrating the universal applicability of our method.

In this paper, we present a novel approach to recognizing human actions from different views by view knowledge transfer. An action is originally modelled as a bag of visual-words (BoVW), which is sensitive to view changes. We argue that, as opposed to visual words, there exist some higher level features which can be shared across views and enable the connection of action models for different views. To discover these features, we use a bipartite graph to model two view-dependent vocabularies, then apply bipartite graph partitioning to co-cluster two vocabularies into visual-word clusters called bilingual-words (i.e., high-level features), which can bridge the semantic gap across view-dependent vocabularies. Consequently, we can transfer a BoVW action model into a bag-of-bilingual-words (BoBW) model, which is more discriminative in the presence of view changes. We tested our approach on the IXMAS data set and obtained very promising results. Moreover, to further fuse view knowledge from multiple views, we apply a Locally Weighted Ensemble scheme to dynamically weight transferred models based on the local distribution structure around each test example. This process can further improve the average recognition rate by about 7%.

We introduce a new large-scale video dataset designed to assess the performance of diverse visual event recognition algorithms with a focus on continuous visual event recognition (CVER) in outdoor areas with wide coverage. Previous datasets for action recognition are unrealistic for real-world surveillance because they consist of short clips showing one action by one individual [15, 8]. Datasets have been developed for movies [11] and sports [12], but, these actions and scene conditions do not apply effectively to surveillance videos. Our dataset consists of many outdoor scenes with actions occurring naturally by non-actors in continuously captured videos of the real world. The dataset includes large numbers of instances for 23 event types distributed throughout 29 hours of video. This data is accompanied by detailed annotations which include both moving object tracks and event examples, which will provide solid basis for large-scale evaluation. Additionally, we propose different types of evaluation modes for visual recognition tasks and evaluation metrics along with our preliminary experimental results. We believe that this dataset will stimulate diverse aspects of computer vision research and help us to advance the CVER tasks in the years ahead.

Several attempts have been lately proposed to tackle the problem of recovering the original image of an underwater scene using a sequence distorted by water waves. The main drawback of the state of the art [18] is that it heavily depends on modelling the waves, which in fact is ill-posed since the actual behavior of the waves along with the imaging process are complicated and include several noise components; therefore, their results are not satisfactory. In this paper, we revisit the problem by proposing a data-driven two-stage approach, each stage is targeted toward a certain type of noise. The first stage leverages the temporal mean of the sequence to overcome the structured turbulence of the waves through an iterative robust registration algorithm. The result of the first stage is a high quality mean and a better structured sequence; however, the sequence still contains unstructured sparse noise. Thus, we employ a second stage at which we extract the sparse errors from the sequence through rank minimization. Our method converges faster, and drastically outperforms state of the art on all testing sequences even only after the first stage.

This paper proposes a novel method for recognition and classification of events represented by Mixture distributions of location and flow. The main idea is to classify observed events into semantically meaningful groups even when motion is observed from distinct viewpoints. Events in the proposed framework are modeled as motion patterns, which are represented by mixtures of multivariate Gaussians, and are obtained by hierarchical clustering of optical flow in the four dimensional space (x, y, u, v). Such motion patterns observed from varying viewpoints, and in distinct locations or datasets, can be compared using different families of divergences between statistical distributions, given that a transformation between the views is known. One of the major contributions of this paper is to compare and match two motion pattern mixture distributions by estimating the similarity transformation between them, that minimizes their Kullback-Leibler (KL) divergence. The KL divergence between Gaussian mixtures is approximated by Monte Carlo sampling, and the minimization is accomplished by employing an iterative nonlinear least squares estimation method, which bears close resemblance to the Iterative Closest Point (ICP) algorithm. We present a robust framework for matching of high-dimensional, sampled point sets representing statistical distributions, by defining similarity measures between them, for global energy minimization. The proposed approach is tested for classification of events observed across several datasets, captured from both static and moving cameras, involving real world pedestrian as well as vehicular motion. Encouraging results are obtained which demonstrate the feasibility and validity of the proposed approach.

Recognition of human actions in a video acquired by a moving camera typically requires standard preprocessing steps such as motion compensation, moving object detection and object tracking. The errors from the motion compensation step propagate to the object detection stage, resulting in miss-detections, which further complicates the tracking stage, resulting in cluttered and incorrect tracks. Therefore, action recognition from a moving camera is considered very challenging. In this paper, we propose a novel approach which does not follow the standard steps, and accordingly avoids the aforementioned difficulties. Our approach is based on Lagrangian particle trajectories which are a set of dense trajectories obtained by advecting optical flow over time, thus capturing the ensemble motions of a scene. This is done in frames of unaligned video, and no object detection is required. In order to handle the moving camera, we propose a novel approach based on low rank optimization, where we decompose the trajectories into their camera-induced and object-induced components. Having obtained the relevant object motion trajectories, we compute a compact set of chaotic invariant features which captures the characteristics of the trajectories. Consequently, a SVM is employed to learn and recognize the human actions using the computed motion features. We performed intensive experiments on multiple benchmark datasets and two new aerial datasets called ARG and APHill, and obtained promising results.

These keynote speeches discuss the following: Finding it Now: Stream Mining in Real Time; Patterns of Motion: Discovery and Generalized Representation; and 3D Structure Reconstruction from Videos.

When the UAV goes to high altitudes such that the observed surface of the earth becomes planar, the structure and motion recovery of the earth's moving plane becomes ambiguous. This planar degeneracy has been pointed out very often in the literature; therefore, current navigation methods either completely fail or give many confusing solutions in such scenario. Interestingly, the horizon line in planar scenes is straight and distinctive; hence, easily detected. Therefore, we show in this paper that the horizon line provides two degrees of freedom that control the relative orientation between the camera coordinate system and the local surface of earth. The recovered degrees of freedom help linearize and disambiguate the planar flow, and therefore we obtain a unique solution for the UAV motion estimation. Unlike previous work which used the horizon to provide the roll angle and the pitch percentage and only employed them for flight stability, we extract the exact angles and directly use them to estimate the ego motion. Additionally, we propose a novel horizon detector based on the maximum a posteriori estimation of both motion and appearance features which outperforms the other detectors in planar scenarios. We thoroughly experimented on the proposed method against information from GPS and gyroscopes, and obtained promising results.

This article presents an interactive application that enables users to improve the visual aesthetics of their digital photographs using several novel spatial recompositing techniques. This work differs from earlier efforts in two important aspects: (1) it focuses on both photo quality assessment and improvement in an integrated fashion, (2) it enables the user to make informed decisions about improving the composition of a photograph. The tool facilitates interactive selection of one or more than one foreground objects present in a given composition, and the system presents recommendations for where it can be relocated in a manner that optimizes a learned aesthetic metric while obeying semantic constraints. For photographic compositions that lack a distinct foreground object, the tool provides the user with crop or expansion recommendations that improve the aesthetic appeal by equalizing the distribution of visual weights between semantically different regions. The recomposition techniques presented in the article emphasize learning support vector regression models that capture visual aesthetics from user data and seek to optimize this metric iteratively to increase the image appeal. The tool demonstrates promising aesthetic assessment and enhancement results on variety of images and provides insightful directions towards future research.

Human identity recognition is an important yet under-addressed problem. Previous methods were strictly limited to high quality photographs, where the principal techniques heavily rely on body details such as face detection. In this paper, we propose an algorithm to address the novel problem of human identity recognition over a set of unordered low quality aerial images. Assuming a user was able to manually locate a target in some images of the set, we find the target in each other query image by implementing a weighted voter-candidate formulation. In the framework, every manually located target is a voter, and the set of humans in a query image are candidates. In order to locate the target, we detect and align blobs of voters and candidates. Consequently, we use PageRank to extract distinguishing regions, and then match multiple regions of a voter to multiple regions of a candidate using Earth Mover Distance (EMD). This generates a robust similarity measure between every voter-candidate pair. Finally, we identify the candidate with the highest weighted vote as the target. We tested our technique over several aerial image sets that we collected, along with publicly available sets, and have obtained promising results.

We present a novel method for the discovery and statistical representation of motion patterns in a scene observed by a static camera. Related methods involving learning of patterns of activity rely on trajectories obtained from object detection and tracking systems, which are unreliable in complex scenes of crowded motion. We propose a mixture model representation of salient patterns of optical flow, and present an algorithm for learning these patterns from dense optical flow in a hierarchical, unsupervised fashion. Using low level cues of noisy optical flow, K-means is employed to initialize a Gaussian mixture model for temporally segmented clips of video. The components of this mixture are then filtered and instances of motion patterns are computed using a simple motion model, by linking components across space and time. Motion patterns are then initialized and membership of instances in different motion patterns is established by using KL divergence between mixture distributions of pattern instances. Finally, a pixel level representation of motion patterns is proposed by deriving conditional expectation of optical flow. Results of extensive experiments are presented for multiple surveillance sequences containing numerous patterns involving both pedestrian and vehicular traffic.

A novel method for crowd flow modeling and anomaly detection is proposed for both coherent and incoherent scenes. The novelty is revealed in three aspects. First, it is a unique utilization of particle trajectories for modeling crowded scenes, in which we propose new and efficient representative trajectories for modeling arbitrarily complicated crowd flows. Second, chaotic dynamics are introduced into the crowd context to characterize complicated crowd motions by regulating a set of chaotic invariant features, which are reliably computed and used for detecting anomalies. Third, a probabilistic framework for anomaly detection and localization is formulated. The overall work-flow begins with particle advection based on optical flow. Then particle trajectories are clustered to obtain representative trajectories for a crowd flow. Next, the chaotic dynamics of all representative trajectories are extracted and quantified using chaotic invariants known as maximal Lyapunov exponent and correlation dimension. Probabilistic model is learned from these chaotic feature set, and finally, a maximum likelihood estimation criterion is adopted to identify a query video of a scene as normal or abnormal. Furthermore, an effective anomaly localization algorithm is designed to locate the position and size of an anomaly. Experiments are conducted on known crowd data set, and results show that our method achieves higher accuracy in anomaly detection and can effectively localize anomalies.

We present an interactive application that enables users to improve the visual aesthetics of their digital photographs using spatial recomposition. Unlike earlier work that focuses either on photo quality assessment or interactive tools for photo editing, we enable the user to make informed decisions about improving the composition of a photograph and to implement them in a single framework. Specifically, the user interactively selects a foreground object and the system presents recommendations for where it can be moved in a manner that optimizes a learned aesthetic metric while obeying semantic constraints. For photographic compositions that lack a distinct foreground object, our tool provides the user with cropping or expanding recommendations that improve its aesthetic quality. We learn a support vector regression model for capturing image aesthetics from user data and seek to optimize this metric during recomposition. Rather than prescribing a fully-automated solution, we allow user-guided object segmentation and inpainting to ensure that the final photograph matches the user's criteria. Our approach achieves 86% accuracy in predicting the attractiveness of unrated images, when compared to their respective human rankings. Additionally, 73% of the images recomposited using our tool are ranked more attractive than their original counterparts by human raters.

Video Surveillance and Monitoring is very active area of research in Computer Vision. However, most of the current approaches assume that the observed scene is not crowded, and that reliable tracks of objects are available over longer durations. Therefore, these approaches are not extendable to more challenging surveillance videos of crowded environments like markets, subways, religious festivals, parades, concerts, football matches etc, where tracking of individual objects is very hard, if not impossible. We have proposed a framework, which views the flow of a high density crowd like the flow of a liquid, prompting the use of ideas and techniques often found in the study of hydrodynamics. Therefore, we treat interactions of people in the scene like moving particles in a liquid on three different length scales (macroscopic, mesoscopic, and microscopic); each scale corresponding to one of the three problems: tracking individuals, detection of abnormal behaviors, and segmentation of crowd motion.

We propose a set of kinematic features that are derived from the optical flow for human action recognition in videos. The set of kinematic features includes divergence, vorticity, symmetric and antisymmetric flow fields, second and third principal invariants of flow gradient and rate of strain tensor, and third principal invariant of rate of rotation tensor. Each kinematic feature, when computed from the optical flow of a sequence of images, gives rise to a spatiotemporal pattern. It is then assumed that the representative dynamics of the optical flow are captured by these spatiotemporal patterns in the form of dominant kinematic trends or kinematic modes. These kinematic modes are computed by performing Principal Component Analysis (PCA) on the spatiotemporal volumes of the kinematic features. For classification, we propose the use of multiple instance learning (MIL) in which each action video is represented by a bag of kinematic modes. Each video is then embedded into a kinematic-mode-based feature space and the coordinates of the video in that space are used for classification using the nearest neighbor algorithm. The qualitative and quantitative results are reported on the benchmark data sets.

In this paper, we present a systematic framework for recognizing realistic actions from videos “in the wild.” Such unconstrained videos are abundant in personal collections as well as on the web. Recognizing action from such videos has not been addressed extensively, primarily due to the tremendous variations that result from camera motion, background clutter, changes in object appearance, and scale, etc. The main challenge is how to extract reliable and informative features from the unconstrained videos. We extract both motion and static features from the videos. Since the raw features of both types are dense yet noisy, we propose strategies to prune these features. We use motion statistics to acquire stable motion features and clean static features. Furthermore, PageRank is used to mine the most informative static features. In order to further construct compact yet discriminative visual vocabularies, a divisive information-theoretic algorithm is employed to group semantically related features. Finally, AdaBoost is chosen to integrate all the heterogeneous yet complementary features for recognition. We have tested the framework on the KTH dataset and our own dataset consisting of 11 categories of actions collected from YouTube and personal videos, and have obtained impressive results for action recognition and action localization.

In this paper, we propose a novel approach for learning generic visual vocabulary. We use diffusion maps to automatically learn a semantic visual vocabulary from abundant quantized midlevel features. Each midlevel feature is represented by the vector of pointwise mutual information (PMI). In this midlevel feature space, we believe the features produced by similar sources must lie on a certain manifold. To capture the intrinsic geometric relations between features, we measure their dissimilarity using diffusion distance. The underlying idea is to embed the midlevel features into a semantic lower-dimensional space. Our goal is to construct a compact yet discriminative semantic visual vocabulary. Although the conventional approach using k-means is good for vocabulary construction, its performance is sensitive to the size of the visual vocabulary. In addition, the learnt visual words are not semantically meaningful since the clustering criterion is based on appearance similarity only. Our proposed approach can effectively overcome these problems by capturing the semantic and geometric relations of the feature space using diffusion maps. Unlike some of the supervised vocabulary construction approaches, and the unsupervised methods such as pLSA and LDA, diffusion maps can capture the local intrinsic geometric relations between the midlevel feature points on the manifold. We have tested our approach on the KTH action dataset, our own YouTube action dataset and the fifteen scene dataset, and have obtained very promising results.

In this paper we introduce a novel method to detect and localize abnormal behaviors in crowd videos using Social Force model. For this purpose, a grid of particles is placed over the image and it is advected with the space-time average of optical flow. By treating the moving particles as individuals, their interaction forces are estimated using social force model. The interaction force is then mapped into the image plane to obtain Force Flow for every pixel in every frame. Randomly selected spatio-temporal volumes of Force Flow are used to model the normal behavior of the crowd. We classify frames as normal and abnormal by using a bag of words approach. The regions of anomalies in the abnormal frames are localized using interaction forces. The experiments are conducted on a publicly available dataset from University of Minnesota for escape panic scenarios and a challenging dataset of crowd videos taken from the web. The experiments show that the proposed method captures the dynamics of the crowd behavior successfully. In addition, we have shown that the social force approach outperforms similar approaches based on pure optical flow.

In this paper, we propose a systematic framework for action recognition in unconstrained amateur videos. Inspired by the success of local features used in object and pose recognition, we extract local static features from the sampled frames to capture local pose shape and appearance. In addition, we extract spatiotemporal features (ST features), which have been successfully used in action recognition, to capture the local motions. In the action recognition phase, we use the Pyramid Match Kernel based on weighted similarities of multi-resolution histograms to match two videos within the same feature types. In order to handle complementary but heterogeneous features, i.e., static and motion features, we chose a multi-kernel classifier for feature fusion. To reduce the noise introduced by the background clutter, our system also tries to automatically find the rough region of interest/action. Preliminary tests on the KTH action dataset, UCF sports dataset, and a YouTube action dataset have shown promising results.

We use concepts from chaos theory in order to model nonlinear dynamical systems that exhibit deterministic behavior. Observed time series from such a system can be embedded into a higher dimensional phase space without the knowledge of an exact model of the underlying dynamics. Such an embedding warps the observed data to a strange attractor, in the phase space, which provides precise information about the dynamics involved. We extract this information from the strange attractor and utilize it to predict future observations. Given an initial condition, the predictions in the phase space are computed through kernel regression. This approach has the advantage of modeling dynamics without making any assumptions about the exact form (linear, polynomial, radial basis, etc.) of the mapping function. The predicted points are then warped back to the observed time series. We demonstrate the utility of these predictions for human action synthesis, and dynamic texture synthesis. Our main contributions are: multivariate phase space reconstruction for human actions and dynamic textures, a deterministic approach to model dynamics in contrast to the popular noise-driven approaches for dynamic textures, and video synthesis from kernel regression in the phase space. Experimental results provide qualitative and quantitative analysis of our approach on standard data sets.

Action recognition methods suffer from many drawbacks in practice, which include (1)the inability to cope with incremental recognition problems; (2)the requirement of an intensive training stage to obtain good performance; (3) the inability to recognize simultaneous multiple actions; and (4) difficulty in performing recognition frame by frame. In order to overcome all these drawbacks using a single method, we propose a novel framework involving the feature-tree to index large scale motion features using Sphere/Rectangle-tree (SR-tree). The recognition consists of the following two steps: 1) recognizing the local features by non-parametric nearest neighbor (NN), 2) using a simple voting strategy to label the action. The proposed method can provide the localization of the action. Since our method does not require feature quantization, the feature- tree can be efficiently grown by adding features from new training examples of actions or categories. Our method provides an effective way for practical incremental action recognition. Furthermore, it can handle large scale datasets due to the fact that the SR-tree is a disk-based data structure. We have tested our approach on two publicly available datasets, the KTH and the IXMAS multi-view datasets, and obtained promising results.

We propose a novel method for automatically discovering key motion patterns happening in a scene by observing the scene for an extended period. Our method does not rely on object detection and tracking, and uses low level features, the direction of pixel wise optical flow. We first divide the video into clips and estimate a sequence of flow-fields. Each moving pixel is quantized based on its location and motion direction. This is essentially a bag of words representation of clips. Once a bag of words representation is obtained, we proceed to the screening stage, using a measure called the `conditional entropy'. After obtaining useful words we apply Diffusion maps. Diffusion maps framework embeds the manifold points into a lower dimensional space while preserving the intrinsic local geometric structure. Finally, these useful words in lower dimensional space are clustered to discover key motion patterns. Diffusion map embedding involves diffusion time parameter which gives us ability to detect key motion patterns at different scales using multi-scale analysis. In addition, clips which are represented in terms of frequency of motion patterns can also be clustered to determine multiple dominant motion patterns which occur simultaneously, providing us further understanding of the scene. We have tested our approach on two challenging datasets and obtained interesting and promising results.

We propose a method of mining most informative features for the event recognition from photo collections. Our goal is to classify different event categories based on the visual content of a group of photos that constitute the event. Such photo groups are typical in a personal photo collection of different events. Visual features are extracted from the images, yet the features from individual images are often noisy and not all of them represent the distinguishing characteristics of an event. We employ the PageRank technique to mine the most informative features from the images that belong to the same event. Subsequently, we classify different event categories using the multiple images of the same event because we argue that they are more informative about the content of an event rather than any single image. We compare our proposed approach with the standard bag of features method (BOF) and observe considerable improvements in recognition accuracy.

The three articles in this special section are selected papers from the IEEE CS Conference on Computer Vision and Pattern Recognition that was held in Anchorage, AL, in June 2008.

Occlusion and lack of visibility in crowded and cluttered scenes make it difficult to track individual people correctly and consistently, particularly in a single view. We present a multi-view approach to solving this problem. In our approach we neither detect nor track objects from any single camera or camera pair; rather evidence is gathered from all the cameras into a synergistic framework and detection and tracking results are propagated back to each view. Unlike other multi-view approaches that require fully calibrated views our approach is purely image-based and uses only 2D constructs. To this end we develop a planar homographic occupancy constraint that fuses foreground likelihood information from multiple views, to resolve occlusions and localize people on a reference scene plane. For greater robustness this process is extended to multiple planes parallel to the reference plane in the framework of plane to plane homologies. Our fusion methodology also models scene clutter using the Schmieder and Weathersby clutter measure, which acts as a confidence prior, to assign higher fusion weight to views with lesser clutter. Detection and tracking are performed simultaneously by graph cuts segmentation of tracks in the space-time occupancy likelihood data. Experimental results with detailed qualitative and quantitative analysis, are demonstrated in challenging multi-view, crowded scenes.

We propose a novel method to model and learn the scene activity, observed by a static camera. The proposed model is very general and can be applied for solution of a variety of problems. The motion patterns of objects in the scene are modeled in the form of a multivariate nonparametric probability density function of spatiotemporal variables (object locations and transition times between them). Kernel Density Estimation is used to learn this model in a completely unsupervised fashion. Learning is accomplished by observing the trajectories of objects by a static camera over extended periods of time. It encodes the probabilistic nature of the behavior of moving objects in the scene and is useful for activity analysis applications, such as persistent tracking and anomalous motion detection. In addition, the model also captures salient scene features, such as the areas of occlusion and most likely paths. Once the model is learned, we use a unified Markov Chain Monte Carlo (MCMC)-based framework for generating the most likely paths in the scene, improving foreground detection, persistent labeling of objects during tracking, and deciding whether a given trajectory represents an anomaly to the observed motion patterns. Experiments with real-world videos are reported which validate the proposed approach.

This paper presents a new method for segmenting medical images by modeling interaction between neighboring structures. Compared to previously reported methods, the proposed approach enables simultaneous segmentation of multiple neighboring structures for improved robustness. During the segmentation process, the object contour evolution and shape prior estimates are influenced by the interactions between neighboring shapes consisting of attraction, repulsion, and competition. Instead of estimating the a priori shape of each structure independently, an interactive maximum a posteriori shape estimation method is used for estimating the shape priors by considering shape prior distribution, neighboring shapes, and image features. Energy functionals are then formulated to model the interaction and segmentation. With the proposed method, neighboring structures with similar intensities and/or textures, and blurred boundaries can be extracted simultaneously. Experimental results obtained on both synthetic data and medical images demonstrate that the introduced interaction between neighboring structures improves segmentation performance compared with other existing approaches.

This is a high level computer vision paper, which employs concepts from Natural Language Understanding in solving the video retrieval problem. Our main contribution is the utilization of the semantic word similarity measures (Lin and PMI-IR similarities) for video retrieval. In our approach, we use trained concept detectors, and the visual co-occurrence relations between such concepts. We propose two methods for content-based retrieval of videos: (1) A method for retrieving anewconcept(a concept which is not known to the system, and no annotation is available) using semantic word similarity and visual co-occurrence. (2) A method for retrieval of videos based on their relevance to a user defined text query using the semantic word similarity and visual content of videos. For evaluation purposes, we have mainly used the automatic search and the high level feature extraction test set of TRECVIDpsila06 benchmark, and the automatic search test set of TRECVIDpsila07. These two data sets consist of 250 hours of multilingual news video captured from American, Arabic, German and Chinese TV channels. Although our method for retrieving a new concept is an unsupervised method, it outperforms the trained concept detectors (which are supervised) on 7 out of 20 test concepts, and overall it performs very close to the trained detectors. On the other hand, our visual content based semantic retrieval method performs 81% better than the text-based retrieval method. This shows that using visual content alone we can obtain significantly good retrieval results.

We present a novel framework for learning patterns of motion and sizes of objects in static camera surveillance. The proposed method provides a new higher-level layer to the traditional surveillance pipeline for anomalous event detection and scene model feedback. Pixel level probability density functions (pdfs) of appearance have been used for background modelling in the past, but modelling pixel level pdfs of object speed and size from the tracks is novel. Each pdf is modelled as a multivariate Gaussian mixture model (GMM) of the motion (destination location & transition time) and the size (width & height) parameters of the objects at that location. Output of the tracking module is used to perform unsupervised EM-based learning of every GMM. We have successfully used the proposed scene model to detect local as well as global anomalies in object tracks. We also show the use of this scene model to improve object detection through pixel-level parameter feedback of the minimum object size and background learning rate. Most object path modelling approaches first cluster the tracks into major paths in the scene, which can be a source of error. We avoid this by building local pdfs that capture a variety of tracks which are passing through them. Qualitative and quantitative analysis of actual surveillance videos proved the effectiveness of the proposed approach.

This paper presents an approach to reconstruct non-stationary, articulated objects from silhouettes obtained with a monocular video sequence. We introduce the concept of motion blurred scene occupancies, a direct analogy of motion blurred images but in a 3D object scene occupancy space resulting from the motion/deformation of the object. Our approach starts with an image based fusion step that combines color and silhouette information from multiple views. To this end we propose to use a novel construct: the temporal occupancy point (TOP), which is the estimated 3D scene location of a silhouette pixel and contains information about duration of time it is occupied. Instead of explicitly computing the TOP in 3D space we directly obtain itpsilas imaged(projected) locations in each view. This enables us to handle monocular video and arbitrary camera motion in scenarios where complete camera calibration information may not be available. The result is a set of blurred scene occupancy images in the corresponding views, where the values at each pixel correspond to the fraction of total time duration that the pixel observed an occupied scene location. We then use a motion de-blurring approach to de-blur the occupancy images. The de-blurred occupancy images correspond to a silhouettes of the mean/motion compensated object shape and are used to obtain a visual hull reconstruction of the object. We show promising results on challenging monocular datasets of deforming objects where traditional visual hull intersection approaches fail to reconstruct the object correctly.

In this paper, we propose a framework that fuses multiple features for improved action recognition in videos. The fusion of multiple features is important for recognizing actions as often a single feature based representation is not enough to capture the imaging variations (view-point, illumination etc.) and attributes of individuals (size, age, gender etc.). Hence, we use two types of features: i) a quantized vocabulary of local spatio-temporal (ST) volumes (or cuboids), and ii) a quantized vocabulary of spin-images, which aims to capture the shape deformation of the actor by considering actions as 3D objects (x, y, t). To optimally combine these features, we treat different features as nodes in a graph, where weighted edges between the nodes represent the strength of the relationship between entities. The graph is then embedded into a k-dimensional space subject to the criteria that similar nodes have Euclidian coordinates which are closer to each other. This is achieved by converting this constraint into a minimization problem whose solution is the eigenvectors of the graph Laplacian matrix. This procedure is known as Fiedler Embedding. The performance of the proposed framework is tested on publicly available data sets. The results demonstrate that fusion of multiple features helps in achieving improved performance, and allows retrieval of meaningful features and videos from the embedding space.

In this paper, we present a novel approach for automatically learning a compact and yet discriminative appearance-based human action model. A video sequence is represented by a bag of spatiotemporal features called video-words by quantizing the extracted 3D interest points (cuboids) from the videos. Our proposed approach is able to automatically discover the optimal number of video-word clusters by utilizing Maximization of Mutual Information(MMI). Unlike the k-means algorithm, which is typically used to cluster spatiotemporal cuboids into video words based on their appearance similarity, MMI clustering further groups the video-words, which are highly correlated to some group of actions. To capture the structural information of the learnt optimal video-word clusters, we explore the correlation of the compact video-word clusters. We use the modified correlgoram, which is not only translation and rotation invariant, but also somewhat scale invariant. We extensively test our proposed approach on two publicly available challenging datasets: the KTH dataset and IXMAS multiview dataset. To the best of our knowledge, we are the first to try the bag of video-words related approach on the multiview dataset. We have obtained very impressive results on both datasets.

In this paper we introduce a template-based method for recognizing human actions called action MACH. Our approach is based on a maximum average correlation height (MACH) filter. A common limitation of template-based methods is their inability to generate a single template using a collection of examples. MACH is capable of capturing intra-class variability by synthesizing a single Action MACH filter for a given action class. We generalize the traditional MACH filter to video (3D spatiotemporal volume), and vector valued data. By analyzing the response of the filter in the frequency domain, we avoid the high computational cost commonly incurred in template-based approaches. Vector valued data is analyzed using the Clifford Fourier transform, a generalization of the Fourier transform intended for both scalar and vector-valued data. Finally, we perform an extensive set of experiments and compare our method with some of the most recent approaches in the field by using publicly available datasets, and two new annotated human action datasets which include actions performed in classic feature films and sports broadcast television.

In this paper we present a novel approach using a 4D (x,y,z,t) action feature model (4D-AFM) for recognizing actions from arbitrary views. The 4D-AFM elegantly encodes shape and motion of actors observed from multiple views. The modeling process starts with reconstructing 3D visual hulls of actors at each time instant. Spatiotemporal action features are then computed in each view by analyzing the differential geometric properties of spatio-temporal volumes (3D STVs) generated by concatenating the actor’s silhouette over the course of the action (x, y, t). These features are mapped to the sequence of 3D visual hulls over time (4D) to build the initial 4D-AFM. Actions are recognized based on the scores of matching action features from the input videos to the model points of 4D-AFMs by exploiting pairwise interactions of features. Promising recognition results have been demonstrated on the multi-view IXMAS dataset using both single and multi-view input videos.

In this paper, a novel feature for capturing information in a spatio-temporal volume based on regularity flow is presented for action recognition. The regularity flow describes the direction of least intensity change within a spatio-temporal volume. Our feature consists of weighted histograms of the computed regularity flow around selected interest points. We then apply this new feature to recognizing actions with experiments on known benchmark dataset. A more discriminating representation of spatio-temporal volume is obtained by using the feature descriptors with the bag of words model. Action recognition is performed by using this new representation with a trained support vector machine. We show that by utilizing regularity flow based features, recognition can be performed with better performance than the best known features. Additionally, results suggest that our descriptor captures information otherwise not harnessed by existing methods.

We propose a 2D model-based approach for tracking human body parts during articulated motion. A human is modeled as a stick figure with thirteen landmarks, and an action is a sequence of these stick figures. Given the locations of these joints in a model video and only the first frame of a test video, the joint locations are automatically estimated throughout the test video using two geometric constraints. The first constraint is based on the invariance of the ratio of areas under an affine transformation, and provides initial estimates. The second one is based on the fundamental matrix, defined by the corresponding landmarks of the two actors, and refines the initial estimates. Using these estimated locations, the tracking algorithm determines the exact location of each joint in the test video. The novelty of our approach lies in the geometric formulation of human actions and the use of geometric constraints for body joints estimation. The approach is able to handle variations in anthropometry of individuals, viewpoints, execution rate, and style of action execution. Experimental results provide encouraging quantitative and qualitative performance analysis.

Learning dominant motion patterns or activities from a video is an important surveillance problem, especially in crowded environments like markets, subways etc., where tracking of individual objects is hard if not impossible. In this paper, we propose an algorithm that uses instantaneous motion field of the video instead of long-term motion tracks for learning the motion patterns. The motion field is a collection of independent flow vectors detected in each frame of the video where each flow is vector is associated with a spatial location. A motion pattern is then defined as a group of flow vectors that are part of the same physical process or motion pattern. Algorithmically, this is accomplished by first detecting the representative modes (sinks) of the motion patterns, followed by construction of super tracks, which are the collective representation of the discovered motion patterns. We also use the super tracks for event-based video matching. The efficacy of the approach is demonstrated on challenging real-world sequences.

Learning typical motion patterns or activities from videos of crowded scenes is an important visual surveillance problem. To detect typical motion patterns in crowded scenarios, we propose a new method which utilizes the instantaneous motions of a video, i.e, the motion flow field, instead of long-term motion tracks. The motion flow field is a union of independent flow vectors computed in different frames. Detecting motion patterns in this flow field can therefore be formulated as a clustering problem of the motion flow fields, where each motion pattern consists of a group of flow vectors participating in the same process or motion. We first construct a directed neighborhood graph to measure the closeness of flow vectors. A hierarchical agglomerative clustering algorithm is applied to group flow vectors into desired motion patterns.

Retrieving videos using key words requires obtaining the semantic features of the videos. Most work reported in the literature focuses on annotating a video shot with a fixed number of key words, no matter how much information is contained in the video shot. In this paper, we propose a new approach to automatically annotate a video shot with an adaptive number of annotation key words according to the richness of the video content. A Semantic Candidate Set (SCS) with fixed size is discovered using visual features. Then the final annotation set, which has an unfixed number of key words, is obtained from the SCS by using Bayesian Inference, which combines static and dynamic inference to remove the irrelevant candidate key words. We have applied our approach to video retrieval. The experiments demonstrate that video retrieval using our annotation approach outperforms retrieval using a fixed number of annotation words.

In practice, understanding the spatial relationships between the surfaces of an object, can significantly improve the performance of object recognition systems. In this paper we propose a novel framework to recognize objects in pictures taken from arbitrary viewpoints. The idea is to maintain the frontal views of the major faces of objects in a global flatmap. Then an unfolding warping technique is used to change the pose of the query object in the test view so that all visible surfaces of the object can be observed from a frontal viewpoint, improving the handling of serious occlusions and large viewpoint changes. We demonstrate the effectiveness of our approach through analysis of recognition trials of complex objects with comparison to popular methods.

In this paper we present a system that uses only vision to land a UAV on a runway. We describe a method for estimating the relative location of the runway as an image by performing image registration against a stack of images in which the location of the runway is known. An approximation of the camera projection model for a forward-facing view of a runway is derived, allowing the course deviation of the UAV to be estimated from a registered image. The course deviation is used as input to a linear feedback control loop to maintain the correct flight path. Our method is implemented as a real-time multithreaded application, which is used to control an aircraft in Microsoft Flight Simulator. We also show results of applying the vision component of the system to video recorded from an actual UAV.

A camera mounted on an aerial vehicle provides an excellent means for monitoring large areas of a scene. Utilizing several such cameras on different aerial vehicles allows further flexibility, in terms of increased visual scope and in the pursuit of multiple targets. In this paper, we address the problem of associating objects across multiple airborne cameras. Since the cameras are moving and often widely separated, direct appearance-based or proximity-based constraints cannot be used. Instead, we exploit geometric constraints on the relationship between the motion of each object across cameras, to test multiple association hypotheses, without assuming any prior calibration information. Given our scene model, we propose a likelihood function for evaluating a hypothesized association between observations in multiple cameras that is geometrically motivated. Since multiple cameras exist, ensuring coherency in association is an essential requirement, e.g. that transitive closure is maintained between more than two cameras. To ensure such coherency we pose the problem of maximizing the likelihood function as a k-dimensional matching and use an approximation to find the optimal assignment of association. Using the proposed error function, canonical trajectories of each object and optimal estimates of inter-camera transformations (in a maximum likelihood sense) are computed. Finally, we show that as a result of associating objects across the cameras, a concurrent visualization of multiple aerial video streams is possible and that, under special conditions, trajectories interrupted due to occlusion or missing detections can be repaired. Results are shown on a number of real and controlled scenarios with multiple objects observed by multiple cameras, validating our qualitative models, and through simulation quantitative performance is also reported.

Image diffusion can smooth away noise and small-scale structures while retaining important features, thus improving the performances for many image processing algorithms. In this paper, we present a novel diffusion algorithm for which the filtering kernels vary according to the perceptual saliency of boundaries. The effectiveness of the proposed approach is validated by experiments on various medical images.

High-throughput genome-wide RNA interference (RNAi) screening is emerging as an essential tool to assist biologists in understanding complex cellular processes. The large number of images produced in each study make manual analysis intractable; hence, automatic cellular image analysis becomes an urgent need, where segmentation is the first and one of the most important steps. In this paper, a fully automatic method for segmentation of cells from genome-wide RNAi screening images is proposed. Nuclei are first extracted from the DNA channel by using a modified watershed algorithm. Cells are then extracted by modeling the interaction between them as well as combining both gradient and region information in the Actin and Rac channels. A new energy functional is formulated based on a novel interaction model for segmenting tightly clustered cells with significant intensity variance and specific phenotypes. The energy functional is minimized by using a multiphase level set method, which leads to a highly effective cell segmentation method. Promising experimental results demonstrate that automatic segmentation of high-throughput genome-wide multichannel screening can be achieved by using the proposed method, which may also be extended to other multichannel image segmentation problems.

In this paper, we use motion and appearance contexts for persistent tracking of objects in aerial imagery. The motion context in a given environment is a collection of trajectories of objects which are representative of the motion of the occluded or unobserved object. It is learned using a clustering scheme based on the Lyapunov characteristic exponent (LCE) which measures the mean exponential rate of divergence of the nearby trajectories. The learned motion context is then used in a regression framework to predict the location of the unobserved object. The appearance context of an occluded (target) object consists of appearance information of objects which are currently occluded or unobserved. It is incorporated by learning a distribution of interclass variation for each target-unobservable object pair. In addition, intra-class variation distribution is constructed for each occluded object using all of its previous observations. Qualitative and quantitative results are reported on challenging aerial sequences.

This paper proposes a framework in which Lagrangian particle dynamics is used for the segmentation of high density crowd flows and detection of flow instabilities. For this purpose, a flow field generated by a moving crowd is treated as an aperiodic dynamical system. A grid of particles is overlaid on the flow field, and is advected using a numerical integration scheme. The evolution of particles through the flow is tracked using a flow map, whose spatial gradients are subsequently used to setup a Cauchy Green deformation tensor for quantifying the amount by which the neighboring particles have diverged over the length of the integration. The maximum eigenvalue of the tensor is used to construct a finite time Lyapunov exponent (FTLE) field, which reveals the Lagrangian coherent structures (LCS) present in the underlying flow. The LCS divide flow into regions of qualitatively different dynamics and are used to locate boundaries of the flow segments in a normalized cuts framework. Any change in the number of flow segments over time is regarded as an instability, which is detected by establishing correspondences between flow segments over time. The experiments are conducted on a challenging set of videos taken from Google Video and a National Geographic documentary.

In this paper, a projection model is presented for cameras moving at constant velocity (which we refer to as Galilean cameras). To that end, we introduce the concept of spacetime projection and show that perspective imaging and linear pushbroom imaging are specializations of the proposed model. The epipolar geometry between two such cameras is developed and we derive the Galilean fundamental matrix. We show how six different "fundamental" matrices can be directly recovered from the Galilean fundamental matrix including the classic fundamental matrix, the linear pushbroom (LP) fundamental matrix and a fundamental matrix relating epipolar plane images (EPIs). To estimate the parameters of this fundamental matrix and the mapping between videos in the case of planar scenes we describe linear algorithms and report experimental performance of these algorithms.

The fundamental matrix is a central construct in the analysis of images captured from a pair of cameras and many feature-based methods have been proposed for its computation. In this paper, we propose a direct method for estimating the fundamental matrix where the motion between the frames is small (e.g. between successive frames of a video). To achieve this, a warping function is presented for the fundamental matrix by using the brightness constancy constraint in conjunction with geometric constraints. Using this warping function, an iterative hierarchical algorithm is described to recover accurate estimates of the fundamental matrix. We present results of experimentation to evaluate the performance of the proposed approach and demonstrate improved accuracy in the computation of the fundamental matrix.

The ability to associate objects across multiple views allows co-operative use of an ensemble cameras for scene understanding. In this paper, we present a principled solution to object association where both the scene and the object motion are modeled. By making the motion model of each object with respect to time explicit, we are able to solve the trajectory association problem in a unified framework for overlapping or non-overlapping cameras. We recover the assignment of associations while simultaneously computing the maximum likelihood estimates of the inter-camera homographies and the trajectory parameters using the expectation maximization algorithm. Quantitative results on simulations are reported along with several results on real data.

The paper introduces an action recognition framework that uses concepts from the theory of chaotic systems to model and analyze nonlinear dynamics of human actions. Trajectories of reference joints are used as the representation of the non-linear dynamical system that is generating the action. Each trajectory is then used to reconstruct a phase space of appropriate dimension by employing a delay-embedding scheme. The properties of the reconstructed phase space are captured in terms of dynamical and metric invariants that include Lyapunov exponent, correlation integral and correlation dimension. Finally, the action is represented by a feature vector which is a combination of these invariants over all the reference trajectories. Our contributions in this paper include :1) investigation of the appropriateness of theory of chaotic systems for human action modelling and recognition, 2) a new set of features to characterize nonlinear dynamics of human actions, 3) experimental validation of the feasibility and potential merits of carrying out action recognition using methods from theory of chaotic systems.

This paper presents a purely image-based approach to fusing foreground silhouette information from multiple arbitrary views. Our approach does not require 3D constructs like camera calibration to carve out 3D voxels or project visual cones in 3D space. Using planar homographies and foreground likelihood information from a set of arbitrary views, we show that visual hull intersection can be performed in the image plane without requiring to go in 3D space. This process delivers a 2D grid of object occupancy likelihoods representing a cross-sectional slice of the object. Subsequent slices of the object are obtained by extending the process to planes parallel to a reference plane in a direction along the body of the object. We show that homographies of these new planes between views can be computed in the framework of plane to plane homologies using the homography induced by a reference plane and the vanishing point of the reference direction. Occupancy grids are stacked on top of each other, creating a three dimensional data structure that encapsulates the object shape and location. Object structure is finally segmented out by minimizing an energy functional over the surface of the object in a level sets formulation. We show the application of our method on complicated object shapes as well as cluttered environments containing multiple objects.

In this paper, we propose a novel approach for scene modeling. The proposed method is able to automatically discover the intermediate semantic concepts. We utilize Maximization of Mutual Information (MMI) co-clustering approach to discover clusters of semantic concepts, which we call intermediate concepts. Each intermediate concept corresponds to a cluster of visterms in the bag of Vis- terms (BOV) paradigm for scene classification. MMI co- clustering results in fewer but meaningful clusters. Unlike k-means which is used to cluster image patches based on their appearances in BOV, MMI co-clustering can group the visterms which are highly correlated to some concept. Unlike probabilistic latent semantic analysis (pLSA), which can be considered as one-sided soft clustering, MMI co- clustering simultaneously clusters visterms and images, so it is able to boost both clustering. In addition, the MMI co- clustering is an unsupervised method. We have extensively tested our proposed approach on two challenging datasets: the fifteen scene categories and the LSCOM dataset, and promising results are obtained.

In this paper, a novel object class detection method based on 3D object modeling is presented. Instead of using a complicated mechanism for relating multiple 2D training views, the proposed method establishes spatial connections between these views by mapping them directly to the surface of 3D model. The 3D shape of an object is reconstructed by using a homographic framework from a set of model views around the object and is represented by a volume consisting of binary slices. Features are computed in each 2D model view and mapped to the 3D shape model using the same homographic framework. To generalize the model for object class detection, features from supplemental views are also considered. A codebook is constructed from all of these features and then a 3D feature model is built. Given a 2D test image, correspondences between the 3D feature model and the testing view are identified by matching the detected features. Based on the 3D locations of the corresponding features, several hypotheses of viewing planes can be made. The one with the highest confidence is then used to detect the object using feature location matching. Performance of the proposed method has been evaluated by using the PASCAL VOC challenge dataset and promising results are demonstrated.

In this paper we introduce a novel contextual fusion method to improve the detection scores of semantic concepts in images and videos. Our method consists of three phases. For each individual concept, the prior probability of the concept is incorporated with detection score of an individual SVM detector. Then probabilistic estimates of the target concept are computed using all of the individual SVM detectors. Finally, these estimates are linearly combined using weights learned from the training set. This procedure is applied to each target concept individually. We show significant improvements to our detection scores on the TRECVID 2005 development set and LSCOM-Lite annotation set. We achieved on average +3.9% improvements in 29 out of 39 concepts.

We propose a new method for performing accurate background subtraction in scenes with a door, like a building entrance or a hallway. This kind of scene is common in surveillance applications, yet the sporadic motion of a door causes problems for existing systems that falsely report the door as foreground. Our method models the scene's appearance by storing a set of Gaussian pixel distributions corresponding to a discrete sample of the door's range of motion. All of the pixels in the image are dependent on the position of the door, so we use the joint probability for all of them to estimate the maximum-likelihood position of the door. We then perform background subtraction using the specific appearance model indexed by our estimated position. We show that our algorithm accurately segments the foreground region in several actual indoor and outdoor surveillance settings.

A common trait of background subtraction algorithms is that they have learning rates, thresholds, and initial values that are hand-tuned for a scenario in order to produce the desired subtraction result; however, the need to tune these parameters makes it difficult to use state-of-the-art methods, fuse multiple methods, and choose an algorithm based on the current application as it requires the end-user to become proficient in tuning a new parameter set. The proposed solution is to automate this task by using a Particle Swarm Optimization (PSO) algorithm to maximize a fitness function compared to provided ground-truth images. The fitness function used is the F-measure, which is the harmonic mean of recall and precision. This method reduces the total pixel error of the Mixture of Gaussians background subtraction algorithm by more than 50% on the diverse Wallflower data-set.

We describe an approach for detecting and segmenting humans with extensive posture articulations in crowded video sequences. In our method we learn a set of mean posture clusters, and a codebook of local shape distributions for humans in various postures. Detection proceeds in two stages: first instances of the codebook entries cast votes for locations of humans in the video and their respective postures. Subsequently, consistent hypotheses are found as maxima within a voting space. The segmentation of humans in the scene is initialized by the corresponding posture clusters and contours are evolved to obtain precise and consistent segmentations.

In this paper we introduce a 3-dimensional (3D) SIFT descriptor for video or 3D imagery such as MRI data. We also show how this new descriptor is able to better represent the 3D nature of video data in the application of action recognition. This paper will show how 3D SIFT is able to outperform previously used description methods in an elegant and efficient manner. We use a bag of words approach to represent videos, and present a method to discover relationships between spatio-temporal words in order to better describe the video data.

Feature selection and extraction is a key operation in video analysis for achieving a higher level of abstraction. In this paper, we introduce a general framework to extract a new spatio-temporal feature that represents the directions in which a video is regular, i.e., the pixel appearances change the least. We propose to model the directions of regular variations with a 3-D vector field, which is referred to as spatio-temporal regularity flow (SPREF). SPREF vectors are designed to have three cross-sectional parallel components Fx, Fy, and Ftfor convenient use in different applications. They are estimated using all the frames simultaneously by minimizing an energy functional formulated according to its definition. In this paper, we first introduce translational SPREF (T-SPREF) and then extend our framework to affine SPREF (A-SPREF). The successful use of SPREF in a few applications, including object removal, video inpainting, and video compression, is also demonstrated

We propose a method for recovering the affine geometry of a dynamically textured plane from a video sequence taken by an uncalibrated, fixed, perspective camera. Some instances of approximately planar surfaces that are coated with a dynamic texture include large water bodies, (such as lakes and oceans), heavy traffic, dense crowds, escalators, and foliage in the wind. Under the assumption of translational dynamic textures, we propose a direct algorithm for the estimation of the inter-frame elation that does not require explicit identification of texels or movetons. In addition, we develop a general algorithm for recovering the affine geometry of homogeneous dynamic textures by identifying a constraint on the expected values of motion magnitudes. We report experimental results on several real videos of dynamic texture found in the world.

We propose a novel approach for tracking of human joints based on anthropometric constraints. A human is modeled as a pictorial structure consisting of body landmarks (joints) and corresponding links between them. Anthropometric constraints relate the landmarks of two persons if they are in the same posture. Given a test video, where an actor performs the same action as in a model video, and joint locations in the model video, anthropometric constraints are used to determine the epipolar lines, where the potential joint locations are searched in the test video. The edge templates around joints and related links are used to locate joints in the test video. The performance of this method is demonstrated on several different human actions

Finding the location where a picture was taken is an important problem for a variety of applications including surveying, interactive traveling and homeland security among others. The task becomes intractable though when the area under investigation reaches city/town size. The amount of data (pictures/videos) required to visually map a city, comprehensively, can be exhaustive for most search algorithms. In this paper we propose a novel method to effectively tackle this problem. The area is visually mapped as route panoramas that provide a compact yet comprehensive representation of the buildings and landmarks in the area. Given a query image taken at an arbitrary location in the area, we show that we can accurately recover the location of the camera by finding it's epipole in the route panorama of the scene. To this end we show that there exists a fundamental matrix between a route panorama and a perspective image of the same scene. The fundamental matrix is calculated using feature matches as correspondences between the query image and the route panorama

In this paper, we present an integrated system for news video retrieval. The proposed system incorporates both speech and visual information in the search mechanisms. The initial search is based on the automatic speech recognition (ASR) transcript of video. Based on the relevant shots selected from the initial search round, keyword histograms are automatically generated for the refinement of the search query, such that the reformulated query fits better to the target topic. We have also developed an image-based refinement module, which uses the region analysis of the video key-frames. SR-tree like indexing structure is constructed for the region features, and the image-to-image similarity is computed using the Earth mover distance. By performing a series of relevance feedback processes, the set of the true relevant shots is expanded significantly. The proposed system has been applied to a large open-benchmark news video dataset, and very satisfactory improvements have been obtained by applying the proposed automatic query expansion and the region-based refinement

This paper presents a novel approach to object recognition involving a sparse 2D model and matching using video. The model is generated on the basis of geometry and image measurables only. We first identify the underlying topological structure of an image dataset containing different views of the objects and represent it as a neighborhood graph. The graph is then refined by identifying redundant images and removing them using morphing. This gives a smaller dataset leading to reduced space requirements and faster matching. Finally we exploit motion continuity in video and extend our algorithm to perform matching based on video input and demonstrate that the results obtained using a video sequence are much robust than using a single image. Our approach is novel in that we do not require any knowledge of camera calibration or viewpoint while generating the model. We also do not assume any constraint on motion of object in test video other than following a smooth trajectory.

Human vision system actively seeks interesting regions in images to reduce the search effort in tasks, such as object detection and recognition. Similarly, prominent actions in video sequences are more likely to attract our first sight than their surrounding neighbors. In this paper, we propose a spatiotemporal video attention detection technique for detecting the attended regions that correspond to both interesting objects and actions in video sequences. Both spatial and temporal saliency maps are constructed and further fused in a dynamic fashion to produce the overall spatiotemporal attention model. In the temporal attention model, motion contrast is computed based on the planar motions (homography) between images, which is estimated by applying RANSAC on point correspondences in the scene. To compensate the non-uniformity of spatial distribution of interest-points, spanning areas of motion segments are incorporated in the motion contrast computation. In the spatial attention model, a fast method for computing pixel-level saliency maps has been developed using color histograms of images. A hierarchical spatial attention representation is established to reveal the interesting points in images as well as the interesting regions. Finally, a dynamic fusion technique is applied to combine both the temporal and spatial saliency maps, where temporal attention is dominant over the spatial model when large motion contrast exists, and vice versa. The proposed spatiotemporal attention framework has been applied on over 20 testing video sequences, and attended regions are detected to highlight interesting objects and motions present in the sequences with very high user satisfaction rate.

We propose a new framework in wavelet video coding to improve the compression rate by exploiting the spatiotemporal regularity of the data. A sequence of images creates a spatiotemporal volume. This volume is said to be regular along the directions in which the pixels vary the least, hence the entropy is the lowest. The wavelet decomposition of regularized data results in a fewer number of significant coefficients, thus yielding a higher compression rate. The directions of regularity of an image sequence depend on both its motion content and spatial structure. We propose the representation of these directions by a 3-D vector field, which we refer to as the spatiotemporal regularity flow (SPREF). SPREF uses splines to approximate the directions of regularity. The compactness of the spline representation results in a low storage overhead for SPREF, which is a desired property in compression applications. Once SPREF directions are known, they can be converted into actual paths along which the data is regular. Directional decomposition of the data along these paths can be further improved by using a special class of wavelet basis called the 3-D orthonormal bandelet basis. SPREF -based video compression not only removes the temporal redundancy, but it also compensates for the spatial redundancy. Our experiments on several standard video sequences demonstrate that the proposed method results in higher compression rates as compared to the standard wavelet based compression.

In most image processing and computer vision applications, real-world scale can only be determined when calibration information is available. Dynamic scenes further complicate most situations. However, some types of dynamic scenes provide useful information that can be used to recover real-world scale. In this paper, we focus on ocean scenes and propose a method for finding sizes in real-world units and the sea state from an uncalibrated camera. Fourier transforms in the space and time dimensions yield spatial and temporal frequency spectra. For water waves, the dispersion relation defines a square relationship between the wavelength and period of a wave. Our method applies this dispersion relation to recover the real-world scale of an ocean sequence. The sea state--including the peak wavelength and period, the wind speed that generated the waves, and the wave heights--is also determined from the frequency spectrum of the sequence combined with stochastic oceanography models. The process is demonstrated on synthetic and real sequences, validating the results with known scene geometry. This has wide applications in port monitoring and coastal surveillance.

Videos are composed of many shots that are caused by different camera operations, e.g., on/off operations and switching between cameras. One important goal in video analysis is to group the shots into temporal scenes, such that all the shots in a single scene are related to the same subject, which could be a particular physical setting, an ongoing action or a theme. In this paper, we present a general framework for temporal scene segmentation in various video domains. The proposed method is formulated in a statistical fashion and uses the Markov chain Monte Carlo (MCMC) technique to determine the boundaries between video scenes. In this approach, a set of arbitrary scene boundaries are initialized at random locations and are automatically updated using two types of updates: diffusion and jumps. Diffusion is the process of updating the boundaries between adjacent scenes. Jumps consist of two reversible operations: the merging of two scenes and the splitting of an existing scene. The posterior probability of the target distribution of the number of scenes and their corresponding boundary locations is computed based on the model priors and the data likelihood. The updates of the model parameters are controlled by the hypothesis ratio test in the MCMC process, and the samples are collected to generate the final scene boundaries. The major advantage of the proposed framework is two-fold: 1) it is able to find the weak boundaries as well as the strong boundaries, i.e., it does not rely on the fixed threshold; 2) it can be applied to different video domains. We have tested the proposed method on two video domains: home videos and feature films, and accurate results have been obtained.

In this paper, we describe how camera parameters and light source orientation can be recovered from two perspective views of a scene given only two vertical lines and their cast shadows. Compared to the traditional calibration methods that involve images of some precisely machined calibration pattern, our method uses new calibration objects: the vertical objects and their parallel shadow lines, which are common in natural environments. In addition to the benefit of increasing accessibility of the calibration objects, the proposed method is also especially useful in cases where only limited information is available. To demonstrate the accuracy and the applications of the proposed algorithm, we present results on both synthetic and real images.

Boosting based detection methods have successfully been used for robust detection of faces and pedestrians. However, a very large amount of labeled examples are required for training such a classifier. Moreover, once trained, the boosted classifier cannot adjust to the particular scenario in which it is employed. In this paper, we propose a co-training based approach to continuously label incoming data and use it for online update of the boosted classifier that was initially trained from a small labeled example set. The main contribution of our approach is that it is an online procedure in which separate views (features) of the data are used for co-training, while the combined view (all features) is used to make classification decisions in a single boosted framework. The features used for classification are derived from principal component analysis of the appearance templates of the training examples. In order to speed up the classification, background modeling is used to prune away stationary regions in an image. Our experiments indicate that starting from a classifier trained on a small training set, significant performance gains can be made through online updation from the unlabeled data.

When viewed from a system of multiple cameras with non-overlapping fields of view, the appearance of an object in one camera view is usually very different from its appearance in another camera view due to the differences in illumination, pose and camera parameters. In order to handle the change in observed colors of an object as it moves from one camera to another, we show that all brightness transfer functions from a given camera to another camera lie in a low dimensional subspace and demonstrate that this subspace can be used to compute appearance similarity. In the proposed approach, the system learns the subspace of inter-camera brightness transfer functions in a training phase during which object correspondences are assumed to be known. Once the training is complete, correspondences are assigned using the maximum a posteriori (MAP) estimation framework using both location and appearance cues. We evaluate the proposed method under several real world scenarios obtaining encouraging results.

Detecting moving objects using stationary cameras is an important precursor to many activity recognition, object recognition and tracking algorithms. In this paper, three innovations are presented over existing approaches. Firstly, the model of the intensities of image pixels as independently distributed random variables is challenged and it is asserted that useful correlation exists in the intensities of spatially proximal pixels. This correlation is exploited to sustain high levels of detection accuracy in the presence of nominal camera motion and dynamic textures. By using a non-parametric density estimation method over a joint domain-range representation of image pixels, multi-modal spatial uncertainties and complex dependencies between the domain (location) and range (color) are directly modeled. Secondly, temporal persistence is proposed as a detection criteria. Unlike previous approaches to object detection which detect objects by building adaptive models of the only background, the foreground is also modeled to augment the detection of objects (without explicit tracking) since objects detected in a preceding frame contain substantial evidence for detection in a current frame. Third, the background and foreground models are used competitively in a MAP-MRF decision framework, stressing spatial context as a condition of pixel-wise labeling and the posterior function is maximized efficiently using graph cuts. Experimental validation of the proposed method is presented on a diverse set of dynamic scenes.

Given a video sequence, obtaining accurate layer segmentation and alpha matting is very important for various applications. However, when a non-textured or smooth area is present in the scene, the segmentation based on only single motion cue usually cannot provide satisfactory results. Conversely, the most matting approaches require a smooth assumption on foreground and background to obtain a good result. In this paper, we combine the merits of motion segmentation and alpha matting technique together to simultaneously achieve high-quality layer segmentation and alpha mattes. First, we explore a general occlusion constraint and design a novel graph cuts framework to solve the layer-based motion segmentation problem for the textured regions using multiple frames. Then, an alpha matting technique is further used to refine the segmentation and resolve the non-textured ambiguities by determining proper alpha values for the foreground and background respectively.

In this paper, we propose to model an action based on both the shape and the motion of the performing object. When the object performs an action in 3D, the points on the outer boundary of the object are projected as 2D (x, y) contour in the image plane. A sequence of such 2D contours with respect to time generates a spatiotemporal volume (STV) in (x, y, t), which can be treated as 3D object in the (x, y, t) space. We analyze STV by using the differential geometric surface properties to identify action descriptors capturing both spatial and temporal properties. A set of action descriptors is called an action sketch. The first step in our approach is to generate STV by solving the point correspondence problem between consecutive frames. The correspondences are determined using a two-step graph theoretical approach. After the STV is generated, actions descriptors are computed by analyzing the differential geometric properties of STV. Finally, using these descriptors, we perform action recognition, which is also formulated as graph theoretical problem. Several experimental results are presented to demonstrate our approach.

In recent years kernel principal component analysis (kernel PCA) has gained much attention because of its ability to capture nonlinear image features, which are particularly important for encoding image structure. Boosting has been established as a powerful learning algorithm that can be used for feature selection. In this paper we present a novel framework for object class detection that combines the feature reduction and feature selection abilities of kernel PCA and AdaBoost respectively. The classifier obtained in this way is able to handle change in object appearance, illumination conditions, and surrounding clutter. A nonlinear subspace is learned for positive and negative object classes using Kernel PCA. Features are derived by projecting example images onto the learned subspaces. Base learners are modeled using Bayes classifier. AdaBoost is then employed to discover the features that are most relevant for the object detection task at hand. The proposed method has been successfully tested on wide range of object classes (cars, air-planes, pedestrians, motorcycles, etc) using standard data sets and has shown remarkable performance. Using a small training set, a classifier learned in this way was able to generalize the intra-class variation while still maintaining high detection rate. In most object categories we achieved detection rates of above 95% with minimal false alarm rates. We demonstrate the effectiveness of our approach in terms of absolute performance parameters and comparative performance against current state of the art approaches.

A camera mounted on an aerial vehicle provides an excellent means for monitoring large areas of a scene. Utilizing several such cameras on different aerial vehicles allows further flexibility, in terms of increased visual scope and in the pursuit of multiple targets. In this paper, we address the problem of tracking objects across multiple moving airborne cameras. Since the cameras are moving and often widely separated, direct appearance-based or proximity-based constraints cannot be used. Instead, we exploit geometric constraints on the relationship between the motions of each object across cameras, to test multiple correspondence hypotheses, without assuming any prior calibration information. We propose a statistically and geometrically meaningful means of evaluating a hypothesized correspondence between two observations in different cameras. Second, since multiple cameras exist, ensuring coherency in correspondence, i.e. transitive closure is maintained between more than two cameras, is an essential requirement. To ensure such coherency we pose the problem of object tracking across cameras as a k-dimensional matching and use an approximation to find the maximum likelihood assignment of correspondence. Third, we show that as a result of tracking objects across the cameras, a concurrent visualization of multiple aerial video streams is possible. Results are shown on a number of real and controlled scenarios with multiple objects observed by multiple cameras, validating our qualitative models.

One of the fundamental challenges of recognizing actions is accounting for the variability that arises when arbitrary cameras capture humans performing actions. In this paper, we explicitly identify three important sources of variability: (1) viewpoint, (2) execution rate, and (3) anthropometry of actors, and propose a model of human actions that allows us to investigate all three. Our hypothesis is that the variability associated with the execution of an action can be closely approximated by a linear combination of action bases in joint spatio-temporal space. We demonstrate that such a model bounds the rank of a matrix of image measurements and that this bound can be used to achieve recognition of actions based only on imaged data. A test employing principal angles between subspaces that is robust to statistical fluctuations in measurement data is presented to find the membership of an instance of an action. The algorithm is applied to recognize several actions, and promising results have been obtained.

This paper contributes a new boosting paradigm to achieve detection of events in video. Previous boosting paradigms in vision focus on single frame detection and do not scale to video events. Thus new concepts need to be introduced to address questions such as determining if an event has occurred, localizing the event, handling same action performed at different speeds, incorporating previous classifier responses into current decision, using temporal consistency of data to aid detection and recognition. The proposed method has the capability to improve weak classifiers by allowing them to use previous history in evaluating the current frame. A learning mechanism built into the boosting paradigm is also given which allows event level decisions to be made. This is contrasted with previous work in boosting which uses limited higher level temporal reasoning and essentially makes object detection decisions at the frame level. Our approach makes extensive use of temporal continuity of video at the classifier and detector levels. We also introduce a relevant set of activity features. Features are evaluated at multiple zoom levels to improve detection. We show results for a system that is able to recognize 11 actions.

Most work in action recognition deals with sequences acquired by stationary cameras with fixed viewpoints. Due to the camera motion, the trajectories of the body parts contain not only the motion of the performing actor but also the motion of the camera. In addition to the camera motion, different viewpoints of the same action in different environments result in different trajectories, which can not be matched using standard approaches. In order to handle these problems, we propose to use the multi-view geometry between two actions. However, well known epipolar geometry of the static scenes where the cameras are stationary is not suitable for our task. Thus, we propose to extend the standard epipolar geometry to the geometry of dynamic scenes where the cameras are moving. We demonstrate the versatility of the proposed geometric approach for recognition of actions in a number of challenging sequences.

Videos are composed of many shots caused by different camera operations, e.g., on/off operations and switching between cameras. One important goal in video analysis is to group the shots into temporal scenes, such that all the shots in a single scene are related to a particular physical setting, an on-going action or a theme. In this paper, we present a general framework for temporal scene segmentation for various video types. The proposed method is formulated in a statistical fashion and uses the Markov chain Monte Carlo (MCMC) technique to determine the boundaries between video scenes. In this approach, an arbitrary number of scene boundaries are randomly initialized and automatically updated using two types of updates: diffuse and jumps. The posterior probability on the number of scenes and their boundary locations is computed based on the model priors and the data likelihood. The updates of the model parameters are controlled by the hypothesis ratio test in the MCMC process. The proposed framework has been experimented on two types of videos, home videos and feature films, and accurate results have been obtained.

This paper proposes a new technique in wavelet video compression that exploits the spatiotemporal regularity of the video. A sequence of frames is said to be regular along the directions in which the pixels vary the least. The directions of regularity of a sequence depend on both its motion content and its spatial structure. We model these directions by a 3D vector field, referred as the structural flow. This flow determines the paths of regularity along which the entropy of the data is smaller. We use these paths to construct a special class of wavelet basis, i.e., the 3D orthonormal bandelet basis for the directional decomposition of the sequence. Our experiments on several standard video sequences demonstrate the significant improvement in compression compared to the standard wavelet video coding.

In this paper, we present a novel framework for generic object class detection by integrating Kernel PCA with AdaBoost. The classifier obtained in this way is invariant to changes in appearance, illumination conditions and surrounding clutter. A nonlinear shape subspace is learned for positive and negative object classes using kernel PCA. Features are derived by projecting example images onto the learned sub-spaces. Base learners are modeled using Bayes classifier. AdaBoost is then employed to discover the features that are most relevant for the object detection task at hand. Proposed method has been successfully tested on wide range of object classes (cars, airplanes, pedestrians, motorcycles etc) using standard data sets and has shown good performance. Using a small training set, the classifier learned in this way was able to generalize the intra-class variation while still maintaining high detection rate. In most object categories, we achieved detection rates of above 95% with minimal false alarm rates. We demonstrate the comparative performance of our method against current state of the art approaches.

Temporal video segmentation is one of the fundamental and essential tasks in video processing, understanding and management. In this paper, we present an automatic method for segmenting the home videos into temporal logical units. We have developed a statistical framework using Markov chain Monte Carlo (MCMC) technique. The temporal scene boundaries are detected by maximizing the posterior probability of the model parameters. The model parameters contain the number of the scenes and the boundary locations of the scenes. The proposed method has been demonstrated on several home videos, and high accuracy has been obtained.

This paper presents a novel object-based video coding framework for videos obtained from a static camera. As opposed to most existing methods, the proposed method does not require explicit 2D or 3D models of objects and hence is general enough to cater for varying types of objects in the scene. The proposed system detects and tracks objects in the scene and learns the appearance model of each object online using incremental principal component analysis (IPCA). Each object is then coded using the coefficients of the most significant principal components of its learned appearance space. Due to smooth transitions between limited number of poses of an object, usually a limited number of significant principal components contribute to most of the variance in the object's appearance space and therefore only a small number of coefficients are required to code the object. The rigid component of the object's motion is coded in terms of its affine parameters. The framework is applied to compressing videos in surveillance and video phone domains. The proposed method is evaluated on videos containing a variety of scenarios such as multiple objects undergoing occlusion, splitting, merging, entering and exiting, as well as a changing background. Results on standard MPEG-7 videos are also presented. For all the videos, the proposed method displays higher Peak Signal to Noise Ratio (PSNR) compared to MPEG-2 and MPEG-4 methods, and provides comparable or better compression.

Most work in human activity recognition is limited to relatively simple behaviors like sitting down, standing up or other dramatic posture changes. Very little has been achieved in detecting more complicated behaviors especially those characterized by the collective participation of several individuals. In this work we present a novel approach to recognizing the class of activities characterized by their rigidity in formation for example people parades, airplane flight formations or herds of animals. The central idea is to model the entire group as a collective rather than focusing on each individual separately. We model the formation as a 3D polygon with each corner representing a participating entity. Tracks from the entities are treated as tracks of feature points on the 3D polygon. Based on the rank of the track matrix we can determine if the 3D polygon under consideration behaves rigidly or undergoes non-rigid deformation. Our method is invariant to camera motion and does not require an a priori model or a training phase.

Information linkage is becoming more and more important in this digital age. In this paper, we propose a concept tracking method, which links news stories on the same topic across multiple sources. The semantic linkage between the news stories is reflected in combination of both of their visual content and their spoken language content. Visually, each news story is represented by a set of key-frames with or without detected faces. The facial key-frames are linked based on the analysis of the extended facial regions, and the non-facial key-frames are correlated using the global Affine matching. The language similarity is expressed in terms of the normalized text similarity between the stories' keywords. The output results of the story linking are further used in a story ranking task, which indicate the interesting level of the stories. The proposed semantic linking framework and the story ranking method have been tested on a set of 60 hours open-benchmark TRECVID video data, and very satisfactory results for both tasks have been obtained. Copyright © 2005 ACM.

In this paper, we present a scene detection framework on continuously recorded videos. Conventional temporal scene segmentation methods work for the videos composed of discrete shots, where shot boundaries are clearly defined. The proposed method detects scene segments by the spectral clustering technique and fuzzy analysis. The detected scenes are represented by the corresponding representative feature values of the feature clusters, rather than abrupt temporal boundaries. The feature clusters are generated using the spectral clustering technique. The video units have the fuzzy memberships to the feature clusters, which are generated using the Hyperbolic tangent fuzzy function. The final output is collected from the candidate scenes from all clusters. The proposed method has been tested on several video sequences, and very promising results have been obtained.

This paper presents a framework for finding point correspondences in monocular image sequences over multiple frames. The general problem of multiframe point correspondence is NP-hard for three or more frames. A polynomial time algorithm for a restriction of this problem is presented and is used as the basis of the proposed greedy algorithm for the general problem. The greedy nature of the proposed algorithm allows it to be used in real-time systems for tracking and surveillance, etc. In addition, the proposed algorithm deals with the problems of occlusion, missed detections, and false positives by using a single noniterative greedy optimization scheme and, hence, reduces the complexity of the overall algorithm as compared to most existing approaches where multiple heuristics are used for the same purpose. While most greedy algorithms for point tracking do not allow for entry and exit of the points from the scene, this is not a limitation for the proposed algorithm. Experiments with real and synthetic data over a wide range of scenarios and system parameters are presented to validate the claims about the performance of the proposed algorithm.

Accurate detection of moving objects is an important precursor to stable tracking or recognition. In this paper, we present an object detection scheme that has three innovations over existing approaches. First, the model of the intensities of image pixels as independent random variables is challenged and it is asserted that useful correlation exists in intensities of spatially proximal pixels. This correlation is exploited to sustain high levels of detection accuracy in the presence of dynamic backgrounds. By using a nonparametric density estimation method over a joint domain-range representation of image pixels, multimodal spatial uncertainties and complex dependencies between the domain (location) and range (color) are directly modeled. We propose a model of the background as a single probability density. Second, temporal persistence is proposed as a detection criterion. Unlike previous approaches to object detection which detect objects by building adaptive models of the background, the foreground is modeled to augment the detection of objects (without explicit tracking) since objects detected in the preceding frame contain substantial evidence for detection in the current frame. Finally, the background and foreground models are used competitively in a MAP-MRF decision framework, stressing spatial context as a condition of detecting interesting objects and the posterior function is maximized efficiently by finding the minimum cut of a capacitated graph. Experimental validation of the proposed method is performed and presented on a diverse set of dynamic scenes.

Extracting layers from video is very important for video representation, analysis, compression, and synthesis. Assuming that a scene can be approximately described by multiple planar regions, this paper describes a robust and novel approach to automatically extract a set of affine or projective transformations induced by these regions, detect the occlusion pixels over multiple consecutive frames, and segment the scene into several motion layers. First, after determining a number of seed regions using correspondences in two frames, we expand the seed regions and reject the outliers employing the graph cuts method integrated with level set representation. Next, these initial regions are merged into several initial layers according to the motion similarity. Third, an occlusion order constraint on multiple frames is explored, which enforces that the occlusion area increases with the temporal order in a short period and effectively maintains segmentation consistency over multiple consecutive frames. Then, the correct layer segmentation is obtained by using a graph cuts algorithm and the occlusions between the overlapping layers are explicitly determined. Several experimental results are demonstrated to show that our approach is effective and robust.

This paper presents a method to perform a high-level segmentation of videos into scenes. A scene can be defined as a subdivision of a play in which either the setting is fixed, or when it presents continuous action in one place. We exploit this fact and propose a novel approach for clustering shots into scenes by transforming this task into a graph partitioning problem. This is achieved by constructing a weighted undirected graph called a shot similarity graph (SSG), where each node represents a shot and the edges between the shots are weighted by their similarity based on color and motion information. The SSG is then split into subgraphs by applying the normalized cuts for graph partitioning. The partitions so obtained represent individual scenes in the video. When clustering the shots, we consider the global similarities of shots rather than the individual shot pairs. We also propose a method to describe the content of each scene by selecting one representative image from the video as a scene key-frame. Recently, DVDs have become available with a chapter selection option where each chapter is represented by one image. Our algorithm automates this objective which is useful for applications such as video-on-demand, digital libraries, and the Internet. Experiments are presented with promising results on several Hollywood movies and one sitcom.

The mapping that relates the image irradiance to the image brightness (intensity) is known as the Radiometric Response Function or Camera Response Function. This usually unknown mapping is nonlinear and varies from one color channel to another. In this paper, we present a method to estimate the radiometric response functions (of R, G and B channels) of a color camera directly from the images of an arbitrary scene taken under different illumination conditions (The illumination conditions are not assumed to be known). The response function of a channel is modeled as a gamma curve and is recovered by using a constrained nonlinear minimization approach by exploiting the fact that the material properties of the scene remain constant in all the images. The performance of the proposed method is demonstrated experimentally.

We propose a tracking method which tracks the complete object regions, adapts to changing visual features, and handles occlusions. Tracking is achieved by evolving the contour from frame to frame by minimizing some energy functional evaluated in the contour vicinity defined by a band. Our approach has two major components related to the visual features and the object shape. Visual features (color, texture) are modeled by semiparametric models and are fused using independent opinion polling. Shape priors consist of shape level sets and are used to recover the missing object regions during occlusion. We demonstrate the performance of our method on real sequences with and without object occlusions.

A scene can be defined as one of the subdivisions of a play in which the setting is fixed, or when it presents continuous action in one place. We propose a novel two-pass algorithm for scene boundary detection, which utilizes the motion content, shot length and color properties of shots as the features. In our approach, shots are first clustered by computing Backward Shot Coherence (BSC) - a shot color similarity measure that detects Potential Scene Boundaries (PSBs) in the videos. In the second pass we compute Scene Dynamics (SD), a function of shot length and the motion content in the potential scenes. In this pass, a scene merging criteria has been developed to remove weak PSBs in order to reduce over segmentation. We also propose a method to describe the content of each scene by selecting one representative image. The segmentation of video data into number of scenes facilitates an improved browsing of videos in electronic form, such as video on demand, digital libraries, Internet. The proposed algorithm has been tested on a variety of videos that include five Hollywood movies, one sitcom, and one interview program and promising results have been obtained.

Conventional tracking approaches assume proximity in space, time and appearance of objects in successive observations. However, observations of objects are often widely separated in time and space when viewed from multiple non-overlapping cameras. To address this problem, we present a novel approach for establishing object correspondence across non-overlapping cameras. Our multicamera tracking algorithm exploits the redundance in paths that people and cars tend to follow, e.g. roads, walk-ways or corridors, by using motion trends and appearance of objects, to establish correspondence. Our system does not require any inter-camera calibration, instead the system learns the camera topology and path probabilities of objects using Parzen windows, during a training phase. Once the training is complete, correspondences are assigned using the maximum a posteriori (MAP) estimation framework. The learned parameters are updated with changing trajectory patterns. Experiments with real world videos are reported, which validate the proposed approach.

In this paper, we propose a novel method to establish temporal correspondence between the frames of two videos. 3D epipolar geometry is used to eliminate the distortion generated by the projection from 3D to 2D. Although the fundamental matrix contains the extrinsic property of the projective geometry between views, it is sensitive to noise. Therefore, we propose the use of a rank constraint of corresponding points in two views to measure the similarity between trajectories. This rank constraint shows more robustness and avoids computation of the fundamental matrix. A dynamic programming approach using the similarity measurement is proposed to find the non-linear time-warping function for videos containing human activities. In this way, videos of different individuals taken at different times and from distinct viewpoints can be synchronized. A temporal pyramid of trajectories is applied to improve the accuracy of the view-invariant dynamic time-warping approach. We show various applications of this approach such as video synthesis, human action recognition, and computer aider training. Compared to state-of-the-art techniques, our method shows a great improvement. 1.

We present a framework for finding point correspondences in monocular image sequences over multiple frames. The general problem of multiframe point correspondence is NP hard for three or more frames. A polynomial time algorithm for a restriction of this problem is presented, and is used as the basis of proposed greedy algorithm for the general problem. The greedy nature of the proposed algorithm allows it to be used in real time systems for tracking and surveillance etc. In addition, the proposed algorithm deals with the problems of occlusion, missed detections, and false positives, by using a single noniterative greedy optimization scheme, and hence, reduces the complexity of the overall algorithm as compared to most existing approaches, where multiple heuristics are used for the same purpose. While most greedy algorithms for point tracking do not allow for entry and exit of points from the scene, this is not a limitation for the proposed algorithm. Experiments with real and synthetic data show that the proposed algorithm outperforms the existing techniques and is applicable in more general settings.

We describe a novel approach to automatically recover corresponding feature points and epipolar geometry over two wide baseline frames. Our contributions consist of several aspects: First, the use of an affine invariant feature, edge-corner, is introduced to provide a robust and consistent matching primitives. Second, based on SVD decomposition of affine matrix, the affine matching space between two corners can be approximately divided into two independent spaces by rotation angle and scaling factor. Employing this property, a two-stage affine matching algorithm is designed to obtain robust matches over two frames. Third, using the epipolar geometry estimated by these matches, more corresponding feature points are determined. Based on these robust correspondences, the fundamental matrix is refined, and a series of virtual views of the scene are synthesized. Finally, several experiments are presented to illustrate that a number of robust correspondences can be stably determined for two wide baseline images under significant camera motions with illumination changes, occlusions, and self-similarities. After testing a number of examples and comparing with the existing methods, the experimental results strongly demonstrate that our matching method outperforms the state-of-art algorithms for all of the test cases.

In this paper, we present a wide area surveillance system that detects, tracks and classifies moving objects across multiple cameras. At the single camera level, tracking is performed using a voting based approach that utilizes color and shape cues to establish correspondence. The system uses the single camera tracking results along with the relationship between camera field of view (FOV) boundaries to establish correspondence between views of the same object in multiple cameras. To this end, a novel approach is described to find the relationships between the FOV lines of cameras. The proposed approach combines tracking in cameras with overlapping and/or non-overlapping FOVs in a unified framework, without requiring explicit calibration. The proposed algorithm has been implemented in a real time system. The system uses a client-server architecture and runs at 10 Hz with three cameras.

We discuss and present a computer vision system that monitors a sitting baby of ages of 6 months old to one-year-old. The system tracks a baby's mouth and hands and alarms the guardians if it finds some objects in the baby's hand going towards his or her mouth. We use skin detection, grayscale based sum of square differences (SSD) and template matching for tracking the hands and the head of a baby.

In this paper, we propose an approach that retrieves motion of objects from the videos based on the dynamic time warping of view invariant characteristics. The motion is represented as a sequence of dynamic instants and intervals, which are automatically computed using the spatiotemporal curvature of the trajectory of moving object in the videos. Dynamic Time Warping (DTW) method matches trajectories using a view invariant similarity measure. Our system is able to incrementally learn different actions without any initialization mode, therefore it can work in an unsupervised manner. The retrieval of relevant videos can be easily performed by computing a simple distance metric. This paper makes two fundamental contribution to view invariant video retrieval: (1) Dynamic Instant detection in trajectories of moving objects acquired from video. (2) View-invariant Dynamic Time Warping to measure similarity between two trajectories of actions performed by different persons and from different viewpoints. Although the learning algorithm is relatively simple in our approach, we can achieve high recognition rate because of the view-invariant representation and the similarity measure using DTW.

We address the issue of tracking moving objects in an environment covered by multiple uncalibrated cameras with overlapping fields of view, typical of most surveillance setups. In such a scenario, it is essential to establish correspondence between tracks of the same object, seen in different cameras, to recover complete information about the object. We call this the problem of consistent labeling of objects when seen in multiple cameras. We employ a novel approach of finding the limits of field of view (FOV) of each camera as visible in the other cameras. We show that, if the FOV lines are known, it is possible to disambiguate between multiple possibilities for correspondence. We present a method to automatically recover these lines by observing motion in the environment, Furthermore, once these lines are initialized, the homography between the views can also be recovered. We present results on indoor and outdoor sequences containing persons and vehicles.

This paper presents a system for analyzing human driver visual attention. The system relies on estimation of global motion and color statistics to robustly track a person's head and facial features. The system is fully automatic, it can initialize automatically, and reinitialize when necessary. The system classifies rotation in all viewing directions, detects eye/mouth occlusion, detects eye blinking and eye closure, and recovers the three dimensional gaze of the eyes. In addition, the system is able to track both through occlusion due to eye blinking, and eye closure, large mouth movement, and also through occlusion due to rotation. Even when the face is fully occluded due to rotation, the system does not break down. Further the system is able to track through yawning, which is a large local mouth motion. Finally, results are presented, and future work on how this system can be used for more advanced driver visual attention monitoring is discussed.

In this paper, we propose a shape from shading (SFS) approach to recover both the shape and the reflectance properties of symmetric objects using a single image. The common constraint of constant or piece-wise constant albedo for lambertJan surfaces is relaxed to arbitrary albedo. The proposed method can be categorized as a linear shape from shading method, which linearizes the reflectance function for symmetric objects using the symmetry cues of the shape and the albedo, and iteratively computes the depth values. Estimated depth values are then used to recover pixel-wise surface albedo. To show the usefulness of the proposed method, we present experimental results for both synthetic and real images.

Video segmentation is different from segmentation of a single image. While several correct solutions may exist for segmenting a single image, there needs to be a consistency among segmentations of each frame for video segmentation. Previous approaches of video segmentation concentrate on motion, or combine motion and color information in a batch fashion. We propose a maximum a posteriori probability (MAP) framework that uses multiple cues, like spatial location, color and motion, for segmentation. We assign weights to color and motion terms, which are adjusted at every pixel, based on a confidence measure of each feature. We also discuss the appropriate modeling of PDFs of each feature of a region. The correct modeling of the spatial PDF imposes temporal consistency among segments in consecutive frames. This approach unifies the strengths of both color segmentation and motion segmentation in one framework, and shows good results on videos that are not suited for either of these approaches.

Automatically understanding human actions using motion trajectories derived from video sequences is a very challenging problem. Since an action takes place in 3-D, and is projected on 2-D image, depending on the viewpoint of the camera, the projected 2-D trajectory may vary. Therefore, the same action may have very different trajectories, and trajectories of different actions may look the same. This may create a problem in interpretation of trajectories at the higher level. However, if the representation of actions only captures characteristics, which are view-invariant, then the higher level interpretation can proceed without any ambiguity. In most of the current work on action recognition, the issue of view invariance has been ignored. Therefore, proposed methods do not succeed in more general situations. In this paper, we first present a view-invariant representation of action consisting of dynamic instants and intervals, which is computed using spatiotemporal curvature of a trajectory. Then this representation is used by our system to learn human actions without any training. The system is able to incrementally learn different actions starting with no model. It can discover instances of the same action performed by different people, and in different viewpoints.

In this paper, we present a method to remove commercials from talk and game show videos and to segment these videos into host and guest shots. In our approach, we mainly rely on information contained in shot transitions, rather than analyzing the scene content of individual frames. We utilize the inherent differences in scene structure of commercials and talk shows to differentiate between them. Similarly, we make use of the well-defined structure of talk shows, which can be exploited to classify shots as host or guest shots. The entire show is first segmented into camera shots based on color histogram. Then, we construct a data-structure (shot connectivity graph) which links similar shots over time. Analysis of the shot connectivity graph helps us to automatically separate commercials from program segments. This is done by first detecting stories, and then assigning a weight to each story based on its likelihood of being a commercial. Further analysis on stories is done to distinguish shots of the hosts from shots of the guests. We have tested our approach on several full-length shows (including commercials) and have achieved video segmentation with high accuracy. The whole scheme is fast and works even on low quality video (160/spl times/120 pixel images at 5 Hz).

We present a method for tracking the 3D position of a finger, using a single camera placed several meters away from the user. After skin detection, we use motion to identify the gesticulating arm. The finger point is found by analyzing the arm's outline. To derive a 3D trajectory, we first track 2D positions of the user's elbow and shoulder. Given that a human's upper arm and lower arm have consistent length, we observe that the possible locations of a finger and elbow form two spheres with constant radii. From the previously tracked body points, we can reconstruct these spheres, computing the 3D position of the elbow and finger. These steps are fully automated and do not require human intervention. The system presented can be used as a visualization tool, or as a user input interface, in cases when the user would rather not be constrained by the camera system.

We describe a system for analyzing human driver alertness. It relies on optical flow and color predicates to robustly track a person's head and facial features. Our system classifies rotation in all viewing directions, detects eye/mouth occlusion, detects eye blinking, and recovers the 3D gaze of the eyes. We show results and discuss how this system can be used for monitoring driver alertness.

This paper presents an automatic system for fire detection in video sequences. There are many previous methods to detect fire, however, all except two use spectroscopy or particle sensors. The two that use visual information suffer from the inability to cope with a moving camera or a moving scene. One of these is not able to work on general data, such as movie sequences. The other is too simplistic and unrestrictive in determining what is considered fire, so that it can be used reliably only in aircraft dry bays. Our system uses color and motion information computed from video sequences to locate fire. This is done by first using an approach that is based upon creating a Gaussian-smoothed color histogram to determine the fire-colored pixels, and then using the temporal variation of pixels to determine which of these pixels are actually fire. Unlike the two previous vision-based methods for pre detection, our method is applicable to more areas because of its insensitivity to camera motion. Two specific applications not possible with previous algorithms are the recognition of fire in the presence of global camera motion or scene motion and the recognition of fire in movies for possible use in an automatic rating system. We show that our method works in a variety of conditions, and that it can automatically determine when it has insufficient information.

This paper presents a glove-free method for tracking hand movements using a set of 3-D models. In this approach, the hand is represented by five cylindrical models which are fit to the third phalangeal segments of the fingers. Six 3-D motion parameters for each model are calculated that correspond to the movement of the fingertips in the image plane. Trajectories of the moving models are then established to show the 3-D nature of the hand motion.

Since the first shape-from-shading (SFS) technique was developed by Horn in the early 1970s, many different approaches have emerged. In this paper, six well-known SFS algorithms are implemented and compared. The performance of the algorithms was analyzed on synthetic images using mean and standard deviation of depth (Z) error, mean of surface gradient (p, q) error, and CPU timing. Each algorithm works well for certain images, but performs poorly for others. In general, minimization approaches are more robust, while the other approaches are faster.