Mask Diffusion Models (MDMs) have recently emerged as a promising alternative to auto-regressive models (ARMs) for vision-language tasks, owing to their flexible balance of efficiency and accuracy. In this paper, for the first time, we introduce MDMs into the Scene Text Recognition (STR) task. We show that vanilla MDM lags behind ARMs in terms of accuracy, although it improves recognition efficiency. To bridge this gap, we propose MDiff4STR, a Mask Diffusion model enhanced with two key improvement strategies tailored for STR. Specifically, we identify two key challenges in applying MDMs to STR: noising gap between training and inference, and overconfident predictions during inference. Both significantly hinder the performance of MDMs. To mitigate the first issue, we develop six noising strategies that better align training with inference behavior. For the second, we propose a token-replacement noise mechanism that provides a non-mask noise type, encouraging the model to reconsider and revise overly confident but incorrect predictions. We conduct extensive evaluations of MDiff4STR on both standard and challenging STR benchmarks, covering diverse scenarios including irregular, artistic, occluded, and Chinese text, as well as whether the use of pretraining. Across these settings, MDiff4STR consistently outperforms popular STR models, surpassing state-of-the-art ARMs in accuracy, while maintaining fast inference with only three denoising steps. Code: https://github.com/Topdu/OpenOCR.

Distilling knowledge from human demonstrations is a promising way for robots to learn and act. Existing methods, which often rely on coarsely-aligned video pairs, are typically constrained to learning global or task-level features. As a result, they tend to neglect the fine-grained frame-level dynamics required for complex manipulation and generalization to novel tasks. We posit that this limitation stems from a vicious circle of inadequate datasets and the methods they inspire. To break this cycle, we propose a paradigm shift that treats fine-grained human-robot alignment as a conditional video generation problem. To this end, we first introduce H&R, a novel third-person dataset containing 2,600 episodes of precisely synchronized human and robot motions, collected using a VR teleoperation system. We then present Human2Robot, a framework designed to leverage this data. Human2Robot employs a Video Prediction Model to learn a rich and implicit representation of robot dynamics by generating robot videos from human input, which in turn guides a decoupled action decoder. Our real-world experiments demonstrate that this approach not only achieves high performance on seen tasks but also exhibits significant one-shot generalization to novel positions, objects, instances, and even new task categories.

Recent advances in generative artificial intelligence have enabled the creation of highly realistic image forgeries, raising significant concerns about digital media authenticity. While existing detection methods demonstrate promising results on benchmark datasets, they face critical limitations in real-world applications. First, existing detectors typically fail to detect semantic inconsistencies with the person’s identity, such as implausible behaviors or incompatible environmental contexts in given images. Second, these methods rely heavily on low-level visual cues, making them effective for known forgeries but less reliable against new or unseen manipulation techniques. To address these challenges, we present a novel personalized vision-language model (VLM) that integrates low-level visual artifact analysis and high-level semantic inconsistency detection. Unlike previous VLM-based methods, our approach avoids resource-intensive supervised fine-tuning that often struggles to preserve distinct identity characteristics. Instead, we employ a lightweight method that dynamically encodes identity-specific information into specialized identifier tokens. This design enables the model to learn distinct identity characteristics while maintaining robust generalization capabilities. We further enhance detection capabilities through a lightweight detection adapter that extracts fine-grained information from shallow features of the vision encoder, preserving critical low-level evidence. Comprehensive experiments demonstrate that our approach achieves 94.25% accuracy and 94.08% F1 score, outperforming both traditional forgery detectors and general VLMs while requiring only 10 extra tokens.

Existing reinforcement learning (RL) methods struggle with long-horizon robotic manipulation tasks, particularly those involving sparse rewards. While action chunking is a promising paradigm for robotic manipulation, using RL to directly learn continuous action chunks in a stable and data-efficient manner remains a critical challenge. This paper introduces AC3 (Actor-Critic for Continuous Chunks), a novel RL framework that learns to generate high-dimensional, continuous action sequences. To make this learning process stable and data-efficient, AC3 incorporates targeted stabilization mechanisms for both the actor and the critic. First, to ensure reliable policy improvement, the actor is trained with an asymmetric update rule, learning exclusively from successful trajectories. Second, to enable effective value learning despite sparse rewards, the critic's update is stabilized using intra-chunk n-step returns and further enriched by a self-supervised module providing intrinsic rewards at anchor points aligned with each action chunk. We conducted extensive experiments on 25 tasks from the BiGym and RLBench benchmarks. Results show that by using only a few demonstrations and a simple model architecture, AC3 achieves superior success rates on most tasks, validating its effective design.

Large Multi-modal Models (LMMs) have significantly advanced a variety of vision-language tasks. The scalability and availability of high-quality training data play a pivotal role in the success of LMMs. In the realm of food, while comprehensive food datasets such as Recipe1M offer an abundance of ingredient and recipe information, they often fall short of providing ample data for nutritional analysis. The Recipe1M+ dataset, despite offering a subset for nutritional evaluation, is limited in the scale and accuracy of nutrition information. To bridge this gap, we introduce Uni-Food, a unified food dataset that comprises over 100,000 images with various food labels, including categories, ingredients, recipes, and ingredient-level nutritional information. To mitigate the conflicts arising from multi-task supervision during fine-tuning of LMMs, we introduce a novel Linear Rectification Mixture of Diverse Experts (RoDE) approach. RoDE utilizes a diverse array of experts to address tasks of varying complexity, thereby facilitating the coordination of trainable parameters, i.e., it allocates more parameters for more complex tasks and, conversely, fewer parameters for simpler tasks. RoDE implements linear rectification union to refine the router’s functionality, thereby enhancing the efficiency of sparse task allocation. These design choices endow RoDE with features that ensure GPU memory efficiency and ease of optimization. Extensive experiments validate the effectiveness of our approach in addressing the inherent challenges of food-related multitasking. UniFood Project

End-to-end text spotting aims to jointly optimize text detection and recognition within a unified framework. Despite significant progress, designing an accurate and efficient end-to-end text spotter for arbitrary-shaped text remains challenging. We identify the primary bottleneck as the lack of a reliable and efficient text detection method. To address this, we propose a novel parameterized text shape representation based on low-rank approximation for precise detection and a triple assignment detection head for fast inference. Specifically, unlike current data-irrelevant shape representation methods, we exploit shape correlations among labeled text boundaries to construct a robust low-rank subspace. By minimizing an $\ell _{1}$ℓ1-norm objective, we extract orthogonal vectors that capture the intrinsic text shape from noisy annotations, enabling precise reconstruction via the linear combination of only a few basis vectors. Next, the triple assignment scheme decouples training complexity from inference speed. It utilizes a deep sparse branch to guide an ultra-lightweight inference branch, while a dense branch provides rich parallel supervision. Building upon these advancements, we integrate the enhanced detection module with a lightweight recognition branch to form an end-to-end text spotting framework, termed LRANet++, capable of accurately and efficiently spotting arbitrary-shaped text. Extensive experiments on challenging benchmarks demonstrate the superiority of LRANet++ compared to state-of-the-art methods.

The transferability of adversarial examples across different models has drawn considerable attention recently, particularly in targeted transferability. Prior research has empirically shown that optimizing adversarial perturbations at neighboring points with the highest loss value improves transferability. While effective, such a method requires multiple iterations to reach the local maxima and disregards the local minima of the input loss landscape. In this paper, we theoretically show that enhancing adversarial transferability is attainable by flattening the input loss landscape. This is accomplished through the perturbation optimization at both local maxima and minima. Moreover, we propose the Cost-efficient LandscapE Flattening (CLEF) attack to consider local maxima and minima around current inputs in a cost-efficient way to flatten the loss landscape and improve adversarial transferability. Specifically, we reuse the gradients of the previous attack step to assist current inputs in reaching local maxima, and employ probabilistic modeling to learn the distributional representations of perturbations that assist current inputs in reaching local minima. This probabilistic modeling can be pre-trained on dozens of images from other domains, enabling us to directly sample this type of perturbation from the pre-trained distribution when attacking. Experimental results demonstrate that integrating local maxima and minima into targeted transferable attacks can significantly flatten the loss landscape of the crafted adversarial examples, resulting in improved adversarial transferability.

Vision-Language Models (VLMs) such as CLIP have demonstrated remarkable capabilities in understanding relationships between visual and textual data through joint embedding spaces. Despite their effectiveness, these models remain vulnerable to adversarial attacks, particularly in the image modality, posing significant security concerns. Building upon our previous work on Adversarial Prompt Tuning (AdvPT), which introduced learnable text prompts to enhance adversarial robustness in VLMs without extensive parameter training, we present a significant extension by introducing the Neural Augmentor framework for Multi-modal Adversarial Prompt Tuning (NAP-Tuning). As a significant extension, NAP-Tuning first establishes a comprehensive multi-modal (text and visual) and multi-layer prompting framework. The core of this framework is a targeted structural augmentation for feature-level purification, implemented through our Neural Augmentor approach. This framework implements feature purification by incorporating TokenRefiners-lightweight neural modules that learn to reconstruct purified features via residual connections-to directly address distortions in the feature space. This structural intervention is what enables the multi-modal and multi-layer system to effectively perform modality-specific and layer-specific feature rectification. Comprehensive experiments demonstrate that NAP-Tuning significantly outperforms existing methods across various datasets and attack types. Notably, our approach shows significant improvements over the strongest baselines under the challenging AutoAttack benchmark, outperforming them by 32.3% on ViT-B16 and 31.3% on ViT-B32 architectures while maintaining competitive clean accuracy. This work highlights the efficacy of internal feature-level intervention in prompt tuning for adversarial robustness, moving beyond input-side alignment approaches to create an adaptive defense mechanism that can identify and rectify adversarial perturbations across embedding spaces.

Backdoor attacks have become a significant threat to the pre-training and deployment of deep neural networks (DNNs). Although numerous methods for detecting and mitigating backdoor attacks have been proposed, most rely on identifying and eliminating the “shortcut” created by the backdoor, which links a specific source class to a target class. However, these approaches can be easily circumvented by designing multiple backdoor triggers that create shortcuts everywhere and therefore nowhere specific. In this study, we explore the concept of Multi-Trigger Backdoor Attacks (MTBAs), where multiple adversaries leverage different types of triggers to poison the same dataset. By proposing and investigating three types of multi-trigger attacks includingparallel,sequential, andhybridattacks, we demonstrate that 1) multiple triggers can coexist, overwrite, or cross-activate one another, and 2) MTBAs easily break the prevalent shortcut assumption underlying most existing backdoor detection/removal methods, rendering them ineffective. Given the security risk posed by MTBAs, we have created a multi-trigger backdoor poisoning dataset to facilitate future research on detecting and mitigating these attacks, and we also discuss potential defense strategies against MTBAs. Our code is available athttps://github.com/bboylyg/Multi-Trigger-Backdoor-Attacks.

Existing referring video segmentation methods often treat a video as a single event consisting of multiple images, overlooking the fact that a video typically contains multiple distinct events. Under such a mechanism, the model needs to directly understand all the complex content in the video and text, which can easily lead to confusion and hallucinations. To address this issue, we propose to decompose a video to a set of simple events by learnable Event Query, and understand complex video content in an event-by-event, easy-to-understand manner. This is based on the observation that natural language expressions often divide a video into distinct, text-related segments, each representing a separate event within a compound event. We introduce EVIS, an Event-Aware Video Instructed Segmentation Assistant, which utilizes text-guided Event Queries to partition a video into simple events, extracting event-aware visual-text features to achieve a hierarchical understanding of the video. Additionally, we propose Object-Pixel-Hybrid Learning, which enables the MLLMs to track targets in long-term videos by integrating fine-grained pixel features with prior object queries. Extensive experimental results on 5 public benchmarks demonstrate EVIS's strong performance in addressing the referring video segmentation task. Code and trained models will be publicly released.

In this paper, we propose the LoRA of Change (LoC) framework for image editing with visual instructions, i.e., before-after image pairs. Compared to the ambiguities, insufficient specificity, and diverse interpretations of natural language, visual instructions can accurately reflect users' intent. Building on the success of LoRA in text-based image editing and generation, we dynamically learn an instruction-specific LoRA to encode the "change" in a before-after image pair, enhancing the interpretability and reusability of our model. Furthermore, generalizable models for image editing with visual instructions typically require quad data, i.e., a before-after image pair, along with query and target images. Due to the scarcity of such quad data, existing models are limited to a narrow range of visual instructions. To overcome this limitation, we introduce the LoRA Reverse optimization technique, enabling large-scale training with paired data alone. Extensive qualitative and quantitative experiments demonstrate that our model produces high-quality images that align with user intent and support a broad spectrum of real-world visual instructions.

Video anomaly detection (VAD) has been extensively researched due to its potential for intelligent video systems. However, most existing methods based on CNNs and transformers still suffer from substantial computational burdens and have room for improvement in learning spatial-temporal normality. Recently, Mamba has shown great potential for modeling long-range dependencies with linear complexity, providing an effective solution to the above dilemma. To this end, we propose a lightweight and effective Mamba-based network named STNMamba, which incorporates carefully designed Mamba modules to enhance the learning of spatial-temporal normality. Firstly, we develop a dual-encoder architecture, where the spatial encoder equipped with Multi-Scale Vision Space State Blocks (MS-VSSB) extracts multi-scale appearance features, and the temporal encoder employs Channel-Aware Vision Space State Blocks (CA-VSSB) to capture significant motion patterns. Secondly, a Spatial-Temporal Interaction Module (STIM) is introduced to integrate spatial and temporal information across multiple levels, enabling effective modeling of intrinsic spatial-temporal consistency. Within this module, the Spatial-Temporal Fusion Block (STFB) is proposed to fuse the spatial and temporal features into a unified feature space, and the memory bank is utilized to store spatial-temporal prototypes of normal patterns, restricting the model's ability to represent anomalies. Extensive experiments on three benchmark datasets demonstrate that our STNMamba achieves competitive performance with fewer parameters and lower computational costs than existing methods.

Recently, text-driven video generation has achieved tremendous progress. However, existing methods neglect the contexts of long short-term frames in the video, thereby compromising temporal consistency. They also encounter challenges of heavy memory costs due to the use of the standard temporal attention mechanism and misalignment between training videos and captions. Additionally, previous approaches for long video generation are flawed because they are hard to ensure content diversity and consistency. To alleviate these issues, we propose a novel Long Short-term Temporal Diffusion (LSTD) model to generate videos with superior temporal consistency. We introduce two novel temporal modules,i.e., the Short-term Temporal Convolution and the Long-term Temporal Attention. The former can learn short-term features with a shallow structure, and the latter concentrates on long-term information of complex motion with a new memory-efficient attention mechanism. The combination of the two modules can ensure the temporal consistency of the generated videos. Furthermore, a novel inference method for long video generation is also proposed, which can iteratively generate hundreds of video frames. Experimental results on UCF-101, MSR-VTT, and two long video benchmarks prove that our method achieves superior zero-shot inference performance even when the size of the training data is reduced by 26.5 times.

Enabling models to recognize vast open-world categories has been a longstanding pursuit in object detection. By leveraging the generalization capabilities of vision-language models, current open-world detectors can recognize a broader range of vocabularies, despite being trained on limited categories. However, when the scale of the category vocabularies during training expands to a real-world level, previous classifiers aligned with coarse class names significantly reduce the recognition performance of these detectors. In this paper, we introduce Prova, a multi-modal prototype classifier for vast-vocabulary object detection. Prova extracts comprehensive multi-modal prototypes as initialization of alignment classifiers to tackle the vast-vocabulary object recognition failure problem. On V3Det, this simple method greatly enhances the performance among one-stage, two-stage, and DETR-based detectors with only additional projection layers in both supervised and open-vocabulary settings. In particular, Prova improves Faster R-CNN, FCOS, and DINO by 3.3, 6.2, and 2.9 AP respectively in the supervised setting of V3Det. For the open-vocabulary setting, Prova achieves a new state-of-the-art performance with 32.8 base AP and 11.0 novel AP, which is of 2.6 and 4.3 gain over the previous methods.

Scene Text Recognition (STR) methods have demonstrated robust performance in word-level text recognition. However, in real applications the text image is sometimes long due to detected with multiple horizontal words. It triggers the requirement to build long text recognition models from readily available short (i.e., word-level) text datasets, which has been less studied previously. In this paper, we term this task Out of Length (OOL) text recognition. We establish the first Long Text Benchmark (LTB) to facilitate the assessment of different methods in long text recognition. Meanwhile, we propose a novel method called OOL Text Recognition with sub-String Matching (SMTR). SMTR comprises two cross-attention-based modules: one encodes a sub-string containing multiple characters into next and previous queries, and the other employs the queries to attend to the image features, matching the sub-string and simultaneously recognizing its next and previous character. SMTR can recognize text of arbitrary length by iterating the process above. To avoid being trapped in recognizing highly similar sub-strings, we introduce a regularization training to compel SMTR to effectively discover subtle differences between similar sub-strings for precise matching. In addition, we propose an inference augmentation strategy to alleviate confusion caused by identical sub-strings in the same text and improve the overall recognition efficiency. Extensive experimental results reveal that SMTR, even when trained exclusively on short text, outperforms existing methods in public short text benchmarks and exhibits a clear advantage on LTB.

The exceptional generative capability of text-to-image models has raised substantial safety concerns regarding the generation of Not-Safe-For-Work (NSFW) content and potential copyright infringement. To address these concerns, previous methods safeguard the models by eliminating inappropriate concepts. Nonetheless, these models alter the parameters of the backbone network and exert considerable influences on the structural (low-frequency) components of the image, which undermines the model's ability to retain irrelevant concepts. In this work, we propose our Dual encoder Modulation network (DuMo), which achieves precise erasure of inappropriate target concepts with minimum impairment to non-target concepts. In contrast to previous methods, DuMo employs the Eraser with PRior Knowledge (EPR) module which modifies the skip connection features of the U-NET and primarily achieves concept erasure on details (high-frequency) components of the image. To minimize the demage to non-target concepts during erasure, the parameters of the backbone U-NET are frozen and the prior knowledge from the original skip connection features is introduced to the erasure process. Meanwhile, the phenomenon is observed that distinct erasing preferences for the image structure and details are demonstrated by the EPR at different timesteps and layers. Therefore, we adopt a novel Time-Layer MOdulation process (TLMO) that adjusts the erasure scale of EPR module's outputs across different layers and timesteps, automatically balancing the erasure effects and model's generative ability. Our method achieves state-of-the-art performance on Explicit Content Erasure (detecting only 34 nude parts), Cartoon Concept Removal (with an average LPIPS_da of 0.428, 0.113 higher than SOTA at 0.315), and Artistic Style Erasure (with an average LPIPS_da of 0.387, 0.088 higher than SOTA at 0.299), clearly outperforming alternative methods.

Transferable adversarial examples highlight the vulnerability of deep neural networks (DNNs) to imperceptible perturbations across various real-world applications. While there have been notable advancements in untargeted transferable attacks, targeted transferable attacks remain a significant challenge. In this work, we focus on generative approaches for targeted transferable attacks. Current generative attacks focus on reducing overfitting to surrogate models and the source data domain, but they often overlook the importance of enhancing transferability through additional semantics. To address this issue, we introduce a novel plug-and-play module into the general generator architecture to enhance adversarial transferability. Specifically, we propose a Semantic Injection Module (SIM) that utilizes the semantics contained in an additional guiding image to improve transferability. The guiding image provides a simple yet effective method to incorporate target semantics from the target class to create targeted and highly transferable attacks. Additionally, we propose new loss formulations that can integrate the semantic injection module more effectively for both targeted and untargeted attacks. We conduct comprehensive experiments under both targeted and untargeted attack settings to demonstrate the efficacy of our proposed approach.

Recent advances in diffusion-based generative models have demonstrated superior performance in subject-driven image generation. Identity (ID) preserving image generation, as a subtask of subject-driven image generation, aims to generate customized images for specific human identity and has broad application potential. However, this task remains challenging due to the requirement for high ID fidelity and precise detail preservation. Additionally, generating high-quality context presents another challenge, as existing methods struggle to achieve both high ID fidelity and satisfactory context simultaneously. To address the issues of insufficient ID fidelity, we introduce a simple yet effective test-time fine-tuning approach. Specifically, we propose an attribute-driven training method that establishes global-level and local-level tasks to learn the global face feature and fine-grained attribute features, respectively. Furthermore, we introduce a novel ID-context decoupling framework that decouples image context generation from human ID generation, ensuring the quality of contextual content as well as facilitating the learning of ID information. Through extensive experiments, we demonstrate the effectiveness of the proposed method and showcase its capabilities across various applications.

Diffusion models, as a type of generative model, have achieved impressive results in generating images and videos conditioned on textual conditions. However, the generation process of diffusion models involves denoising dozens of steps to produce photorealistic images/videos, which is computationally expensive. Unlike previous methods that design ``one-size-fits-all'' approaches for speed up, we argue denoising steps should be sample-specific conditioned on the richness of input texts. To this end, we introduce AdaDiff, a lightweight framework designed to learn instance-specific step usage policies, which are then used by the diffusion model for generation. AdaDiff is optimized using a policy gradient method to maximize a carefully designed reward function, balancing inference time and generation quality. We conduct experiments on three image generation and two video generation benchmarks and demonstrate that our approach achieves similar visual quality compared to the baseline using a fixed 50 denoising steps while reducing inference time by at least 33%, going as high as 40%. Furthermore, our method can be used on top of other acceleration methods to provide further speed benefits. Lastly, qualitative analysis shows that AdaDiff allocates more steps to more informative prompts and fewer steps to simpler prompts.

Wei Tang, Yixin Cao, Yang Deng, Jiahao Ying, Bo Wang, Yizhe Yang, Yuyue Zhao, Qi Zhang, Xuanjing Huang, Yu-Gang Jiang, Yong Liao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Zhiheng Xi, Yiwen Ding, Wenxiang Chen, Boyang Hong, Honglin Guo, Junzhe Wang, Xin Guo, Dingwen Yang, Chenyang Liao, Wei He, Songyang Gao, Lu Chen, Rui Zheng, Yicheng Zou, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang, Zuxuan Wu, Yu-Gang Jiang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Rui Wang, Bohao Li, Xiyang Dai, Jianwei Yang, Yi-Ling Chen, Zhen Xing, Yifan Yang, Dongdong Chen, Xipeng Qiu, Zuxuan Wu, Yu-Gang Jiang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Class-incremental learning (CIL) enables models to learn new classes progressively while preserving knowledge of previously learned ones. Recent advances in this field have shifted towards parameter-efficient fine-tuning techniques, with many approaches building upon the framework that maintains a pool of learnable prompts. Although effective, these methods introduce substantial computational overhead, primarily due to prompt pool querying and increased input sequence lengths from prompt concatenation. In this work, we present a novel prompt-based approach that addresses this limitation. Our method trains a single set of shared prompts across all tasks and, rather than concatenating prompts to the input, directly modifies the CLS token's attention computation by adding the prompts to it. This simple and lightweight design not only significantly reduces computational complexity-both in terms of inference costs and the number of trainable parameters-but also eliminates the need to optimize prompt lengths for different downstream tasks, offering a more efficient yet powerful solution for rehearsal-free class-incremental learning. Extensive experiments across a diverse range of CIL benchmarks demonstrate the effectiveness of our approach, highlighting its potential to establish a new prompt-based CIL paradigm. Furthermore, experiments on general recognition benchmarks beyond the CIL setting also show strong performance, positioning our method as a promising candidate for a general parameter-efficient fine-tuning approach.

Connectionist temporal classification (CTC)-based scene text recognition (STR) methods, e.g., SVTR, are widely employed in OCR applications, mainly due to their simple architecture, which only contains a visual model and a CTC-aligned linear classifier, and therefore fast inference. However, they generally exhibit worse accuracy than encoder-decoder-based methods (EDTRs) due to struggling with text irregularity and linguistic missing. To address these challenges, we propose SVTRv2, a CTC model endowed with the ability to handle text irregularities and model linguistic context. First, a multi-size resizing strategy is proposed to resize text instances to appropriate predefined sizes, effectively avoiding severe text distortion. Meanwhile, we introduce a feature rearrangement module to ensure that visual features accommodate the requirement of CTC, thus alleviating the alignment puzzle. Second, we propose a semantic guidance module. It integrates linguistic context into the visual features, allowing CTC model to leverage language information for accuracy improvement. This module can be omitted at the inference stage and would not increase the time cost. We extensively evaluate SVTRv2 in both standard and recent challenging benchmarks, where SVTRv2 is fairly compared to popular STR models across multiple scenarios, including different types of text irregularity, languages, long text, and whether employing pretraining. SVTRv2 surpasses most EDTRs across the scenarios in terms of accuracy and inference speed. Code: https://github.com/Topdu/OpenOCR.

Text-guided video prediction (TVP) involves predicting the motion of future frames from the initial frame according to an instruction, which has wide applications in virtual reality, robotics, and content creation. Previous TVP methods make significant breakthroughs by adapting Stable Diffusion for this task. However, they struggle with frame consistency and temporal stability primarily due to the limited scale of video datasets. We observe that pretrained Image2Video diffusion models possess good priors for video dynamics but they lack textual control. Hence, transferring Image2Video models to leverage their video dynamic priors while injecting instruction control to generate controllable videos is both a meaningful and challenging task. To achieve this, we introduce the Multi-Modal Large Language Model (MLLM) to predict future video states based on initial frames and text instructions. More specifically, we design a dual query transformer (DQFormer) architecture, which integrates the instructions and frames into the conditional embeddings for future frame prediction. Additionally, we develop Long-Short Term Temporal Adapters and Spatial Adapters that can quickly transfer general video diffusion models to specific scenarios with minimal training costs. Experimental results show that our method significantly outperforms state-of-the-art techniques on four datasets: Something Something V2, Epic Kitchen-100, Bridge Data, and UCF-101. Notably, AID achieves 91.2% and 55.5% FVD improvements on Bridge and SSv2 respectively, demonstrating its effectiveness in various domains. More examples can be found at our website https://chenhsing.github.io/AID.

Referring audio-visual segmentation (RAVS) has recently seen significant advancements, yet challenges remain in integrating multimodal information and deeply understanding and reasoning about audiovisual content. To extend the boundaries of RAVS and facilitate future research in this field, we propose Omnimodal Referring Audio-Visual Segmentation (OmniAVS), a new dataset containing 2,104 videos and 61,095 multimodal referring expressions. OmniAVS stands out with three key innovations: (1) 8 types of multimodal expressions that flexibly combine text, speech, sound, and visual cues; (2) an emphasis on understanding audio content beyond just detecting their presence; and (3) the inclusion of complex reasoning and world knowledge in expressions. Furthermore, we introduce Omnimodal Instructed Segmentation Assistant (OISA), to address the challenges of multimodal reasoning and fine-grained understanding of audiovisual content in OmniAVS. OISA uses MLLM to comprehend complex cues and perform reasoning-based segmentation. Extensive experiments show that OISA outperforms existing methods on OmniAVS and achieves competitive results on other related tasks.

Diffusion models have been recognized for their ability to generate images that are not only visually appealing but also of high artistic quality. As a result, Layout-to-Image (L2I) generation has been proposed to leverage region-specific positions and descriptions to enable more precise and controllable generation. However, previous methods primarily focus on UNet-based models (\eg SD1.5 and SDXL), and limited effort has explored Multimodal Diffusion Transformers (MM-DiTs), which have demonstrated powerful image generation capabilities. Enabling MM-DiT for layout-to-image generation seems straightforward but is challenging due to the complexity of how layout is introduced, integrated, and balanced among multiple modalities. To this end, we explore various network variants to efficiently incorporate layout guidance into MM-DiT, and ultimately present SiamLayout. To inherit the advantages of MM-DiT, we use a separate set of network weights to process the layout, treating it as equally important as the image and text modalities. Meanwhile, to alleviate the competition among modalities, we decouple the image-layout interaction into a siamese branch alongside the image-text one and fuse them in the later stage. Moreover, we contribute a large-scale layout dataset, named LayoutSAM, which includes 2.7 million image-text pairs and 10.7 million entities. Each entity is annotated with a bounding box and a detailed description. We further construct the LayoutSAM-Eval benchmark as a comprehensive tool for evaluating the L2I generation quality. Finally, we introduce the Layout Designer, which taps into the potential of large language models in layout planning, transforming them into experts in layout generation and optimization. These components form CreatiLayout -- a systematic solution that integrates the layout model, dataset, and planner for creative layout-to-image generation.

General-purposed embodied agents are designed to understand the users' natural instructions or intentions and act precisely to complete universal tasks. Recently, methods based on foundation models especially Vision-Language-Action models (VLAs) have shown a substantial potential to solve language-conditioned manipulation (LCM) tasks well. However, existing benchmarks do not adequately meet the needs of VLAs and relative algorithms. To better define such general-purpose tasks in the context of LLMs and advance the research in VLAs, we present VLABench, an open-source benchmark for evaluating universal LCM task learning. VLABench provides 100 carefully designed categories of tasks, with strong randomization in each category of task and a total of 2000+ objects. VLABench stands out from previous benchmarks in four key aspects: 1) tasks requiring world knowledge and common sense transfer, 2) natural language instructions with implicit human intentions rather than templates, 3) long-horizon tasks demanding multi-step reasoning, and 4) evaluation of both action policies and language model capabilities. The benchmark assesses multiple competencies including understanding of mesh\&texture, spatial relationship, semantic instruction, physical laws, knowledge transfer and reasoning, etc. To support the downstream finetuning, we provide high-quality training data collected via an automated framework incorporating heuristic skills and prior information. The experimental results indicate that both the current state-of-the-art pretrained VLAs and the workflow based on VLMs face challenges in our tasks.

Continual learning, also known as lifelong learning or incremental learning, refers to the process by which a model learns from a stream of incoming data over time. A common problem in continual learning is the classification layer’s bias towards the most recent task. Traditionally, methods have relied on incorporating data from past tasks during training to mitigate this issue. However, the recent shift in continual learning to memory-free environments has rendered these approaches infeasible. In this study, we propose a solution focused on the testing phase. We first introduce a simple Out-of-Task Detection method, OTD, designed to accurately identify samples from past tasks during testing. Leveraging OTD, we then propose: (1) an Adaptive Retention mechanism for dynamically tuning the classifier layer on past task data; (2) an Adaptive Correction mechanism for revising predictions when the model classifies data from previous tasks into classes from the current task. We name our approach Adaptive Retention & Correction (ARC). While designed for memory-free environments, ARC also proves effective in memorybased settings. Extensive experiments show that our proposed method can be plugged in to virtually any existing continual learning approach without requiring any modifications to its training procedure. Specifically, when integrated with state-of-the-art approaches, ARC achieves an average performance increase of 2.7% and 2.6% on the CIFAR-100 and Imagenet-R datasets, respectively

In this paper, we focus on black-box defense for VLMs against jailbreak attacks. Existing black-box defense methods are either unimodal or bimodal. Unimodal methods enhance either the vision or language module of the VLM, while bimodal methods robustify the model through text-image representation realignment. However, these methods suffer from two limitations: 1) they fail to fully exploit the cross-modal information, or 2) they degrade the model performance on benign inputs. To address these limitations, we propose a novel blue-team method BlueSuffix that defends target VLMs against jailbreak attacks without compromising its performance under black-box setting. BlueSuffix includes three key components: 1) a visual purifier against jailbreak images, 2) a textual purifier against jailbreak texts, and 3) a blue-team suffix generator using reinforcement fine-tuning for enhancing cross-modal robustness. We empirically show on four VLMs (LLaVA, MiniGPT-4, InstructionBLIP, and Gemini) and four safety benchmarks (Harmful Instruction, AdvBench, MM-SafetyBench, and RedTeam-2K) that BlueSuffix outperforms the baseline defenses by a significant margin. Our BlueSuffix opens up a promising direction for defending VLMs against jailbreak attacks. Code is available at https://github.com/Vinsonzyh/BlueSuffix.

This paper addresses the problem of category-level pose estimation for articulated objects in robotic manipulation tasks. Recent works have shown promising results in estimating part pose and size at the category level. However, these approaches primarily follow a complex multi-stage pipeline that first segments part instances in the point cloud and then estimates the Normalized Part Coordinate Space (NPCS) representation for 6D poses. These approaches suffer from high computational costs and low performance in real-time robotic tasks. To address these limitations, we propose YOEO, a single-stage method that simultaneously outputs instance segmentation and NPCS representations in an end-to-end manner. We use a unified network to generate point-wise semantic labels and centroid offsets, allowing points from the same part instance to vote for the same centroid. We further utilize a clustering algorithm to distinguish points based on their estimated centroid distances. Finally, we first separate the NPCS region of each instance. Then, we align the separated regions with the real point cloud to recover the final pose and size. Experimental results on the GAPart dataset demonstrate the pose estimation capabilities of our proposed single-shot method. We also deploy our synthetically-trained model in a real-world setting, providing real-time visual feedback at 200Hz, enabling a physical Kinova robot to interact with unseen articulated objects. This showcases the utility and effectiveness of our proposed method2.

Accurate 6D pose estimation is key for robotic manipulation, enabling precise object localization for tasks like grasping. We present RAG-6DPose, a retrieval-augmented approach that leverages 3D CAD models as a knowledge base by integrating both visual and geometric cues. Our RAG-6DPose roughly contains three stages: 1) Building a Multi-Modal CAD Knowledge Base by extracting 2D visual features from multi-view CAD rendered images and also attaching 3D points; 2) Retrieving relevant CAD features from the knowledge base based on the current query image via our ReSPC module; and 3) Incorporating retrieved CAD information to refine pose predictions via retrieval-augmented decoding. Experimental results on standard benchmarks and real-world robotic tasks demonstrate the effectiveness and robustness of our approach, particularly in handling occlusions and novel viewpoints. Supplementary material is available on our project website: https://sressers.github.io/RAG-6DPose.

The field of video generation has witnessed remarkable advances in recent years, driven by innovations in deep generative models. Nevertheless, the fidelity of AI-generated videos remains far from perfect, with synthesized content frequently exhibiting visual artifacts, such as temporally inconsistent motion, physically implausible trajectories, unnatural object deformations, and local blurring, that undermine realism and user trust. Precise detection and spatial localization of these artifacts are of critical importance: not only are they essential for automatic quality control pipelines that improves user experience, but they also provide actionable diagnostic signals for researchers and practitioners to guide model development and evaluation. Despite its significance, the research community currently lacks a comprehensive benchmark tailored for artifact localization in AI-generated videos. Existing datasets either focus solely on detection at the video or frame level, or lack fine-grained spatial annotations necessary for developing and benchmarking localization methods. To fill this gap, we present BrokenVideos, a benchmark dataset comprising ~3,254 AI-generated videos with carefully-annotated, pixel-level masks indicating regions of visual corruption. Each annotation is the result of careful human inspection, ensuring high-quality ground truth for artifact localization tasks. We demonstrate that training existing video artifact detection models and multi-modal large language models (MLLMs) on BrokenVideos substantially enhances their ability to localize corrupted regions within generated content. Through extensive experiments and cross-model evaluations, we show that BrokenVideos provides a critical foundation for both benchmarking and advancing artifact localization research. We hope our dataset can catalyze further innovation in both video generation and its quality assurance. The dataset is available at: https://broken-video-detection-datetsets.github.io/Broken-Video-Detection-Datasets.github.io/.

Large-scale pre-training frameworks like CLIP have revolutionized multimodal learning, but their reliance on web-scraped datasets, frequently containing private user data, raises serious concerns about misuse. Unlearnable Examples (UEs) have emerged as a promising countermeasure against unauthorized model training, employing carefully crafted unlearnable noise to disrupt the learning of meaningful representations from protected data. Current approaches typically generate UEs by jointly optimizing unlearnable noise for both images and their associated text descriptions (or labels). However, this optimization process is often computationally prohibitive for on-device execution, forcing reliance on external third-party services. This creates a fundamental privacy paradox: users must initially expose their data to these very services to achieve protection, thereby compromising privacy in the process. Such a contradiction has severely hindered the development of practical, scalable data protection solutions. To resolve this paradox, we introduce Text-to-Unlearnable Example (T2UE), a novel framework that enables users to generate UEs using only text descriptions. T2UE circumvents the need for original image data by employing a text-to-image (T2I) model to map text descriptions into the image (noise) space, combined with an error-minimization framework to produce effective unlearnable noise. Extensive experiments show that T2UE-protected data substantially degrades performance in downstream tasks (e.g., cross-modal retrieval) for state-of-the-art models. Notably, the protective effect generalizes across diverse architectures and even to supervised learning settings. Our work demonstrates the feasibility of ''zero-contact data protection'', where personal data can be safeguarded based solely on their textual descriptions, eliminating the need for direct data exposure.

Federated Prompt Tuning (FPT) is an efficient method for cross-client collaborative fine-tuning of large Vision-Language Models (VLMs). However, models tuned using FPT are vulnerable to adversarial attacks, leading to misclassification in downstream tasks. In this work, we introduce Federated Adversarial Prompt Tuning (FedAPT), a novel method designed to enhance the adversarial robustness of FPT. We identify a key issue in FedAPT under non-independent and identically distributed (non-IID) settings: a class information gap between clients and the global model. Clients rely solely on limited local label information to generate adversarial samples for training, while the global model must defend against adversarial attacks from global labels. To address this issue, we propose a class-aware prompt generator that generates visual prompts from text prompts. This generator is guided by a Global Label Embedding (serving as a ''beacon'') which encodes cross-client label information to create more globally-aligned visual prompts. Additionally, we propose a cross-layer generator sharing strategy to enhance prompt coupling across different layers of the model, further boosting adversarial robustness. Extensive experiments on multiple image classification datasets demonstrate the superiority of FedAPT in improving adversarial robustness, outperforming existing methods by a large margin. FedAPT also exhibits exceptional generalization in cross-domain and cross-dataset scenarios, indicating its effectiveness in real-world applications.

Recent advances in image editing systems reveal critical limitations in handling complex real-world scenarios requiring multimodal condition controls. While text instructions enable broad semantic guidance, visual examples provide precise visual reference in specific scenarios, existing unimodal approaches fail to synergize these complementary modalities effectively. We propose EditMaster, a unified framework that integrates text and visual controls through multimodal instruction learning, enabling precise image manipulation with bidirectional consistency. Our framework introduces three core innovations: A multimodal large language model enhanced with extended visual tokens replaces CLIP text encoders, generating pre-edited visual guidance that aligns textual commands with visual examples to guide diffusion model toward high-quality outputs; The Mask-Based Decoupled Residual Exemplar-Attention module preserves unedited regions through spatial masking while integrating visual details via residual pathways; A systematic data construction method converts unimodal editing datasets into a task-specific multimodal dataset, eliminating the need for de novo data construction. Experiments show that our approach outperforms unimodal baselines and excels in complex multimodal instruction editing, setting a new benchmark for this field.

For robotic manipulation, existing robotics datasets and simulation benchmarks predominantly cater to robot-arm platforms. However, for humanoid robots equipped with dual arms and dexterous hands, simulation tasks and high-quality demonstrations are notably lacking. Bimanual dexterous manipulation is inherently more complex, as it requires coordinated arm movements and hand operations, making autonomous data collection challenging. This paper presents HumanoidGen, an automated task creation and demonstration collection framework that leverages atomic dexterous operations and LLM reasoning to generate relational constraints. Specifically, we provide spatial annotations for both assets and dexterous hands based on the atomic operations, and perform an LLM planner to generate a chain of actionable spatial constraints for arm movements based on object affordances and scenes. To further improve planning ability, we employ a variant of Monte Carlo tree search to enhance LLM reasoning for long-horizon tasks and insufficient annotation. In experiments, we create a novel benchmark with augmented scenarios to evaluate the quality of the collected data. The results show that the performance of the 2D and 3D diffusion policies can scale with the generated dataset. Project page is https://openhumanoidgen.github.io.

In daily life, people often move through spaces to find objects that meet their needs, posing a key challenge in embodied AI. Traditional Demand-Driven Navigation (DDN) handles one need at a time but does not reflect the complexity of real-world tasks involving multiple needs and personal choices. To bridge this gap, we introduce Task-Preferenced Multi-Demand-Driven Navigation (TP-MDDN), a new benchmark for long-horizon navigation involving multiple sub-demands with explicit task preferences. To solve TP-MDDN, we propose AWMSystem, an autonomous decision-making system composed of three key modules: BreakLLM (instruction decomposition), LocateLLM (goal selection), and StatusMLLM (task monitoring). For spatial memory, we design MASMap, which combines 3D point cloud accumulation with 2D semantic mapping for accurate and efficient environmental understanding. Our Dual-Tempo action generation framework integrates zero-shot planning with policy-based fine control, and is further supported by an Adaptive Error Corrector that handles failure cases in real time. Experiments demonstrate that our approach outperforms state-of-the-art baselines in both perception accuracy and navigation robustness.

Cross-Domain Few-Shot Object Detection (CD-FSOD) aims to detect novel objects with only a handful of labeled samples from previously unseen domains. While data augmentation and generative methods have shown promise in few-shot learning, their effectiveness for CD-FSOD remains unclear due to the need for both visual realism and domain alignment. Existing strategies, such as copy-paste augmentation and text-to-image generation, often fail to preserve the correct object category or produce backgrounds coherent with the target domain, making them non-trivial to apply directly to CD-FSOD. To address these challenges, we propose Domain-RAG, a training-free, retrieval-guided compositional image generation framework tailored for CD-FSOD. Domain-RAG consists of three stages: domain-aware background retrieval, domain-guided background generation, and foreground-background composition. Specifically, the input image is first decomposed into foreground and background regions. We then retrieve semantically and stylistically similar images to guide a generative model in synthesizing a new background, conditioned on both the original and retrieved contexts. Finally, the preserved foreground is composed with the newly generated domain-aligned background to form the generated image. Without requiring any additional supervision or training, Domain-RAG produces high-quality, domain-consistent samples across diverse tasks, including CD-FSOD, remote sensing FSOD, and camouflaged FSOD. Extensive experiments show consistent improvements over strong baselines and establish new state-of-the-art results. Codes will be released upon acceptance.The source code and instructions are available at https://github.com/LiYu0524/Domain-RAG.

Large Multimodal Models (LMMs) have made significant breakthroughs with the advancement of instruction tuning. However, while existing models can understand images and videos at a holistic level, they still struggle with instance-level understanding that requires a more fine-grained comprehension and alignment. Instance-level understanding is crucial for LMMs, as it focuses on the specific elements that we are most interested in. Excitingly, existing works find that the state-of-the-art LMMs exhibit strong instance understanding capabilities when provided with explicit visual cues. Motivated by this, we proposed Inst-IT, a solution to enhance LMMs in Instance understanding via explicit visual prompt Instruction Tuning for instance guidance. Inst-IT consists of a benchmark to diagnose multimodal instance-level understanding, a large-scale instruction-tuning dataset, and a continuous instruction-tuning training paradigm to effectively enhance spatial-temporal instance understanding capabilities of existing LMMs. Experimental results show that, enhanced by Inst-IT, our models not only achieve outstanding performance on Inst-IT-Bench and other instance understanding benchmarks, but also demonstrate significant improvements across various generic image and video understanding benchmarks. This highlights that our method not only boosts instance-level understanding but also strengthens the overall capabilities of generic image and video comprehension.

Multimodal large language models have unlocked new possibilities for various multimodal tasks. However, their potential in image manipulation detection remains unexplored. When directly applied to the IMD task, M-LLMs often produce reasoning texts that suffer from hallucinations and overthinking. To address this, we propose ForgerySleuth, which leverages M-LLMs to perform comprehensive clue fusion and generate segmentation outputs indicating specific regions that are tampered with. Moreover, we construct the ForgeryAnalysis dataset through the Chain-of-Clues prompt, which includes analysis and reasoning text to upgrade the image manipulation detection task. A data engine is also introduced to build a larger-scale dataset for the pre-training phase. Our extensive experiments demonstrate the effectiveness of ForgeryAnalysis and show that ForgerySleuth significantly outperforms existing methods in generalization, robustness, and explainability.

Achieving fine-grained spatio-temporal understanding in videos remains a major challenge for current Video Large Multimodal Models (Video LMMs). Addressing this challenge requires mastering two core capabilities: video referring understanding, which captures the semantics of video regions, and video grounding, which segments object regions based on natural language descriptions. However, most existing approaches tackle these tasks in isolation, limiting progress toward unified, referentially grounded video interaction. We identify a key bottleneck in the lack of high-quality, unified video instruction data and a comprehensive benchmark for evaluating referentially grounded video chat. To address these challenges, we contribute in three core aspects: dataset, model, and benchmark. First, we introduce SAMA-239K, a large-scale dataset comprising 15K videos specifically curated to enable joint learning of video referring understanding, grounding, and multi-turn video chat. Second, we propose the SAMA model, which incorporates a versatile spatio-temporal context aggregator and a Segment Anything Model to jointly enhance fine-grained video comprehension and precise grounding capabilities. Finally, we establish SAMA-Bench, a meticulously designed benchmark consisting of 5,067 questions from 522 videos, to comprehensively evaluate the integrated capabilities of Video LMMs in multi-turn, spatio-temporal referring understanding and grounded dialogue. Extensive experiments and benchmarking results show that SAMA not only achieves strong performance on SAMA-Bench but also sets a new state-of-the-art on general grounding benchmarks, while maintaining highly competitive performance on standard visual understanding benchmarks.

As Video Large Multimodal Models (VLMMs) rapidly advance, their inherent complexity introduces significant safety challenges, particularly the issue of mismatched generalization where static safety alignments fail to transfer to dynamic video contexts. We introduce SafeVid, a framework designed to instill video-specific safety principles in VLMMs. SafeVid uniquely transfers robust textual safety alignment capabilities to the video domain by employing detailed textual video descriptions as an interpretive bridge, facilitating LLM-based rule-driven safety reasoning. This is achieved through a closed-loop system comprising: 1) generation of SafeVid-350K, a novel 350,000-pair video-specific safety preference dataset; 2) targeted alignment of VLMMs using Direct Preference Optimization (DPO); and 3) comprehensive evaluation via our new SafeVidBench benchmark. Alignment with SafeVid-350K significantly enhances VLMM safety, with models like LLaVA-NeXT-Video demonstrating substantial improvements (e.g., up to 42.39%) on SafeVidBench. SafeVid provides critical resources and a structured approach, demonstrating that leveraging textual descriptions as a conduit for safety reasoning markedly improves the safety alignment of VLMMs in complex multimodal scenarios.

Autoregressive (AR) models have demonstrated strong potential in visual generation, offering competitive performance with simple architectures and optimization objectives. However, existing methods are typically limited to single-modality conditions, \eg, text or category labels, restricting their applicability in real-world scenarios that demand image synthesis from diverse forms of controls. In this work, we present \system, the first unified autoregressive framework for Any-to-Image generation. By discretizing various visual conditions through a shared visual tokenizer and text prompts with a text tokenizer, \system supports a broad spectrum of conditional inputs within a single model, including text (text-to-image generation), spatial signals (segmentation-to-image and depth-to-image), and visual context (image editing, frame prediction, and text-to-video generation). To mitigate the risk of information leakage from condition tokens to content tokens, we introduce Disentangled Causal Attention (DCA), which separates the full-sequence causal mask into condition causal attention and content causal attention. It serves as a training-time regularizer without affecting the standard next-token prediction during inference. With this design, \system achieves new state-of-the-art results across a range of benchmark, \eg, 0.63 on GenEval and 80.02 on VBench, demonstrating its effectiveness in flexible and high-fidelity visual generation.

Scalable Vector Graphics (SVG) is an important image format widely adopted in graphic design because of their resolution independence and editability. The study of generating high-quality SVG has continuously drawn attention from both designers and researchers in the AIGC community. However, existing methods either produces unstructured outputs with huge computational cost or is limited to generating monochrome icons of over-simplified structures. To produce high-quality and complex SVG, we propose OmniSVG, a unified framework that leverages pre-trained Vision-Language Models (VLMs) for end-to-end multimodal SVG generation. By parameterizing SVG commands and coordinates into discrete tokens, OmniSVG decouples structural logic from low-level geometry for efficient training while maintaining the expressiveness of complex SVG structure. To further advance the development of SVG synthesis, we introduce MMSVG-2M, a multimodal dataset with two million richly annotated SVG assets, along with a standardized evaluation protocol for conditional SVG generation tasks. Extensive experiments show that OmniSVG outperforms existing methods and demonstrates its potential for integration into professional SVG design workflows.

The growing interest in generating recipes from food images has drawn substantial research attention in recent years. Existing works for recipe generation primarily utilize a two-stage training method—first predicting ingredients from a food image and then generating instructions from both the image and ingredients. Large Multi-modal Models (LMMs), which have achieved notable success across a variety of vision and language tasks, shed light on generating both ingredients and instructions directly from images. Nevertheless, LMMs still face the common issue of hallu-cinations during recipe generation, leading to suboptimal performance. To tackle this issue, we propose a retrieval augmented large multimodal model for recipe generation. We first introduce Stochastic Diversified Retrieval Augmentation (SDRA) to retrieve recipes semantically related to the image from an existing datastore as a supplement, integrating them into the prompt to add diverse and rich context to the input image. Additionally, Self-Consistency Ensemble Voting mechanism is proposed to determine the most confident prediction recipes as the final output. It calculates the consistency among generated recipe candidates, which use different retrieval recipes as context for generation. Extensive experiments validate the effectiveness of our proposed method, which demonstrates state-of-the-art (SOTA) performance in recipe generation on the Recipe1M dataset.

This paper proposes a large-scale multi-modal dataset for referring motion expression video segmentation, focusing on segmenting and tracking target objects in videos based on language description of objects' motions. Existing referring video segmentation datasets often focus on salient objects and use language expressions rich in static attributes, potentially allowing the target object to be identified in a single frame. Such datasets underemphasize the role of motion in both videos and languages. To explore the feasibility of using motion expressions and motion reasoning clues for pixel-level video understanding, we introduce MeViS, a dataset containing 33,072 human-annotated motion expressions in both text and audio, covering 8,171 objects in 2,006 videos of complex scenarios. We benchmark 15 existing methods across 4 tasks supported by MeViS, including 6 referring video object segmentation (RVOS) methods, 3 audio-guided video object segmentation (AVOS) methods, 2 referring multi-object tracking (RMOT) methods, and 4 video captioning methods for the newly introduced referring motion expression generation (RMEG) task. The results demonstrate weaknesses and limitations of existing methods in addressing motion expression-guided video understanding. We further analyze the challenges and propose an approach LMPM++ for RVOS/AVOS/RMOT that achieves new state-of-the-art results. Our dataset provides a platform that facilitates the development of motion expression-guided video understanding algorithms in complex video scenes.

Scene text recognition (STR) methods have struggled to attain high accuracy and fast inference speed. Auto-Regressive (AR)-based models implement the recognition in a character-by-character manner, showing superiority in accuracy but with slow inference speed. Alternatively, Parallel Decoding (PD)-based models infer all characters in a single decoding pass, offering faster inference speed but generally worse accuracy. To realize the dual goals of "AR-level accuracy and PD-level speed", we propose a Context Perception Parallel Decoder (CPPD) to perceive the related context and predict the character sequence in a PD pass. CPPD devises a character counting module to infer the occurrence count of each character, and a character ordering module to deduce the content-free reading order and positions. Meanwhile, the character prediction task associates the positions with characters. They together build a comprehensive recognition context, which benefits the decoder to focus accurately on characters with the attention mechanism, thereby improving the recognition accuracy. We construct a series of CPPD models and also plug the proposed modules into existing STR decoders. Experiments on both English and Chinese benchmarks demonstrate that the CPPD models achieve highly competitive accuracy while running much faster than existing leading models. Moreover, the plugged models achieve significant accuracy improvements.

Multi-modal models have shown appealing performance in visual recognition tasks, as free-form text-guided training evokes the ability to understand fine-grained visual content. However, current models cannot be trivially applied to scene text recognition (STR) due to the compositional difference between natural and text images. We propose a novel instruction-guided scene text recognition (IGTR) paradigm that formulates STR as an instruction learning problem and understands text images by predicting character attributes, e.g., character frequency, position, etc. IGTR first devises instruction triplets, providing rich and diverse descriptions of character attributes. To effectively learn these attributes through question-answering, IGTR develops a lightweight instruction encoder, a cross-modal feature fusion module and a multi-task answer head, which guides nuanced text image understanding. Furthermore, IGTR realizes different recognition pipelines simply by using different instructions, enabling a character-understanding-based text reasoning paradigm that differs from current methods considerably. Experiments on English and Chinese benchmarks show that IGTR outperforms existing models by significant margins, while maintaining a small model size and fast inference speed. Moreover, by adjusting the sampling of instructions, IGTR offers an elegant way to tackle the recognition of rarely appearing and morphologically similar characters, which were previous challenges.

Deep learning models have emerged as strong and efficient tools that can be applied to a broad spectrum of complex learning problems and many real-world applications. However, more and more works show that deep models are vulnerable to adversarial examples. Compared to vanilla attack settings, this paper advocates a more practical setting of data-free black-box attack, for which the attackers can completely not access the structures and parameters of the target model, as well as the intermediate features and any training data associated with the model. To tackle this task, previous methods generate transferable adversarial examples from a transparent substitute model to the target model. However, we found that these works have the limitations of taking static substitute model structure for different targets, only using hard synthesized examples once, and still relying on data statistics of the target model. This may potentially harm the performance of attacking the target model. To this end, we propose a novel Dynamic Routing and Knowledge Re-Learning framework (DraKe) to effectively learn a dynamic substitute model from the target model. Specifically, given synthesized training samples, a dynamic substitute structure learning strategy is proposed to adaptively generate optimal substitute model structure via a policy network according to different target models and tasks. To facilitate the substitute training, we present a graph-based structure information learning to capture the structural knowledge learned from the target model. For the inherent limitation that online data generation can only be learned once, a dynamic knowledge re-learning strategy is proposed to adjust the weights of optimization objectives and re-learn hard samples. Extensive experiments on four public image classification datasets and one face recognition benchmark are conducted to evaluate the efficacy of our Drake. We can obtain significant improvement compared with state-of-the-art competitors. More importantly, our DraKe consistently achieves attack superiority for different target models (e.g., residual networks, and vision transformers), showing great potential for complex real-world applications.

Visual Object Tracking (VOT) aims to estimate the positions of target objects in a video sequence, which is an important vision task with various real-world applications. Depending on whether the initial states of target objects are specified by provided annotations in the first frame or the categories, VOT could be classified as instance tracking (e.g., SOT and VOS) and category tracking (e.g., MOT, MOTS, and VIS) tasks. Different definitions have led to divergent solutions for these two types of tasks, resulting in redundant training expenses and parameter overhead. In this paper, combing the advantages of the best practices developed in both communities, we propose a novel tracking-with-detection paradigm, where tracking supplements appearance priors for detection and detection provides tracking with candidate bounding boxes for the association. Equipped with such a design, a unified tracking model, OmniTracker, is further presented to resolve all the tracking tasks with a fully shared network architecture, model weights, and inference pipeline, eliminating the need for task-specific architectures and reducing redundancy in model parameters. We conduct extensive experimentation on seven prominent tracking datasets of different tracking tasks, including LaSOT, TrackingNet, DAVIS16-17, MOT17, MOTS20, and YTVIS19, and demonstrate that OmniTracker achieves on-par or even better results than both task-specific and unified tracking models.

Anomaly detection has garnered extensive applications in real industrial manufacturing due to its remarkable effectiveness and efficiency. However, previous generative-based models have been limited by suboptimal reconstruction quality, hampering their overall performance. We introduce DiffusionAD, a novel anomaly detection pipeline comprising a reconstruction sub-network and a segmentation sub-network. A fundamental enhancement lies in our reformulation of the reconstruction process using a diffusion model into a noise-to-norm paradigm. Here, the anomalous region loses its distinctive features after being disturbed by Gaussian noise and is subsequently reconstructed into an anomaly-free one. Afterward, the segmentation sub-network predicts pixel-level anomaly scores based on the similarities and discrepancies between the input image and its anomaly-free reconstruction. Additionally, given the substantial decrease in inference speed due to the iterative denoising nature of diffusion models, we revisit the denoising process and introduce a rapid one-step denoising paradigm. This paradigm achieves hundreds of times acceleration while preserving comparable reconstruction quality. Furthermore, considering the diversity in the manifestation of anomalies, we propose a norm-guided paradigm to integrate the benefits of multiple noise scales, enhancing the fidelity of reconstructions. Comprehensive evaluations on four standard and challenging benchmarks reveal that DiffusionAD outperforms current state-of-the-art approaches and achieves comparable inference speed, demonstrating the effectiveness and broad applicability of the proposed pipeline.

Online media data, in the form of images and videos, are becoming mainstream communication channels. However, recent advances in deep learning (DL), particularly deep generative models, open the doors for producing perceptually convincing images and videos at a low cost, which not only poses a serious threat to the trustworthiness of digital information but also has severe societal implications. This motivates a growing interest in research in media tampering detection (TD), i.e., using DL techniques to examine whether media data have been maliciously manipulated. Depending on the content of the targeted images, media forgery could be divided into image tampering and Deepfake techniques. The former typically moves or erases the visual elements in ordinary images, while the latter manipulates the expressions and even the identity of human faces. Accordingly, the means of defense include image TD and Deepfake detection (DFD), which share a wide variety of properties. In this article, we provide a comprehensive review of the current media TD approaches and discuss the challenges and trends in this field for future research.

Incremental Learning (IL) aims to learn deep models on sequential tasks continually, where each new task includes a batch of new classes and deep models have no access to task ID information at the inference time. Recent vast pre-trained models (PTMs) have achieved outstanding performance by prompt technique in practical IL without the old samples (rehearsal-free) and with a memory constraint (memory-constrained): Prompt-extending and Prompt-fixed methods. However, prompt-extending methods need a large memory buffer to maintain an ever-expanding prompt pool and meet an extra challenging prompt selection problem. Prompt-fixed methods only learn a single set of prompts on one of the incremental tasks and can not handle all the incremental tasks effectively. To achieve a good balance between the memory cost and the performance on all the tasks, we propose a Parameter-Efficient Cross-Task Prompt (PECTP) framework with Prompt Retention Module (PRM) and classifier Head Retention Module (HRM). To make the final learned prompts effective on all incremental tasks, PRM constrains the evolution of cross-task prompts’ parameters from Outer Prompt Granularity and Inner Prompt Granularity. Besides, we employ HRM to inherit old knowledge in the previously learned classifier heads to facilitate the cross-task prompts’ generalization ability. Extensive experiments show the effectiveness of our method. The source codes are available at https://github.com/RAIAN08/PECTP.

Open-set single-source domain generalization aims to use a single-source domain to learn a robust model that can be generalized to unknown target domains with both domain shifts and label shifts. The scarcity of the source domain and the unknown data distribution of the target domain pose a great challenge for domain-invariant feature learning and unknown class recognition. In this article, we propose a novel learning approach based on domain expansion and boundary growth to expand the scarce source samples and enlarge the boundaries across the known classes that indirectly broaden the boundary between the known and unknown classes. Specifically, we achieve domain expansion by employing both background suppression and style augmentation on the source data to synthesize new samples. Then we force the model to distill consistent knowledge from the synthesized samples so that the model can learn domain-invariant information. Furthermore, we realize boundary growth across classes by using edge maps as an additional modality of samples when training multi-binary classifiers. In this way, it enlarges the boundary between the inliers and outliers, and consequently improves the unknown class recognition during open-set generalization. Extensive experiments show that our approach can achieve significant improvements and reach state-of-the-art performance on several cross-domain image classification datasets.

The precise recognition of food categories plays a pivotal role for intelligent health management, attracting significant research attention in recent years. Prominent benchmarks, such as Food-101 and VIREO Food-172, provide abundant food image resources that catalyze the prosperity of research in this field. Nevertheless, these datasets are well-curated from canteen scenarios and thus deviate from food appearances in daily life. This discrepancy poses great challenges in effectively transferring classifiers trained on these canteen datasets to broader daily-life scenarios encountered by humans. Toward this end, we present two new benchmarks, namely DailyFood-172 and DailyFood-16, specifically designed to curate food images from everyday meals. These two datasets are used to evaluate the transferability of approaches from the well-curated food image domain to the everyday-life food image domain. In addition, we also propose a simple yet effective baseline method named Multi-Cluster Reference Learning (MCRL) to tackle the aforementioned domain gap. MCRL is motivated by the observation that food images in daily-life scenarios exhibit greater intra-class appearance variance compared with those in well-curated benchmarks. Notably, MCRL can be seamlessly coupled with existing approaches, yielding non-trivial performance enhancements. We hope our new benchmarks can inspire the community to explore the transferability of food recognition models trained on well-curated datasets toward practical real-life applications.

Exploring a substantial amount of unlabeled data, semi-supervised learning boosts the recognition performance when only a limited number of labels are provided. However, conventional methods assume a class-balanced data distribution, which is difficult to realize in practice due to the long-tailed nature of real-world data. While addressing the data imbalance is a well-explored area in supervised learning paradigms, directly transferring existing approaches to SSL is nontrivial, as prior knowledge about unlabeled data distribution remains unknown in SSL. In light of this, we introduce the Balanced Memory Bank (BMB), a framework for long-tailed semi-supervised learning. The core of BMB is an online-updated memory bank that caches historical features alongside their corresponding pseudo-labels, and the memory is also carefully maintained to ensure the data therein are class-rebalanced. Furthermore, an adaptive weighting module is incorporated to work jointly with the memory bank to further re-calibrate the biased training process. Experimental results across various datasets demonstrate the superior performance of BMB compared with state-of-the-art approaches. For instance, an improvement of 8.2% on the 1% labeled subset of ImageNet127 and 4.3% on the 50% labeled subset of ImageNet-LT.

Vision transformer (ViT) variants have made rapid advances on a variety of computer vision tasks. However, their performance on corrupted inputs, which are inevitable in realistic use cases due to variations in lighting and weather, has not been explored comprehensively. In this paper, we probe the robustness gap among ViT variants and ask how these modern architectural developments affect performance under common types of corruption. Through extensive and rigorous benchmarking, we demonstrate that simple architectural designs such as overlapping patch embedding and convolutional feed-forward networks can promote the robustness of ViTs. Moreover, since the de facto training of ViTs relies heavily on data augmentation, exactly which augmentation strategies make ViTs more robust is worth investigating. We survey the efficacy of previous methods and verify that adversarial noise training is powerful. In addition, we introduce a novel conditional method for generating dynamic augmentation parameters conditioned on input images, which offers state-of-the-art robustness to common corruptions.

Large Multi-modal Models (LMMs) have made impressive progress in many vision-language tasks. Nevertheless, the performance of general LMMs in specific domains is still far from satisfactory. This paper proposes FoodLMM, a versatile food assistant based on LMMs with various capabilities, including food recognition, ingredient recognition, recipe generation, nutrition estimation, food segmentation, and multi-round conversation. To facilitate FoodLMM in dealing with tasks beyond pure text output, we introduce a series of novel task-specific tokens and heads, enabling the model to predict food nutritional values and multiple segmentation masks. We adopt a two-stage training strategy. In the first stage, we utilize multiple public food benchmarks for multi-task learning by leveraging the instruct-following paradigm. In the second stage, we construct a multi-round conversation dataset and a reasoning segmentation dataset to fine-tune the model, enabling it to conduct professional dialogues and generate segmentation masks based on complex reasoning in the food domain. Our fine-tuned FoodLMM achieves state-of-the-art results across several food benchmarks. Our code, models, and datasets are available athttps://github.com/YuehaoYin/FoodLMM.

Camera-based bird-eye-view (BEV) perception paradigm has made significant progress in the autonomous driving field. Under such a paradigm, accurate BEV representation construction relies on reliable depth estimation for multi-camera images. However, existing approaches exhaustively predict depths for every pixel without prioritizing objects, which are precisely the entities requiring detection in the 3D space. To this end, we propose IA-BEV, which integrates image-plane instance awareness into the depth estimation process within a BEV-based detector. First, a category-specific structural priors mining approach is proposed for enhancing the efficacy of monocular depth generation. Besides, a self-boosting learning strategy is further proposed to encourage the model to place more emphasis on challenging objects in computation-expensive temporal stereo matching. Together they provide advanced depth estimation results for high-quality BEV features construction, benefiting the ultimate 3D detection. The proposed method achieves state-of-the-art performances on the challenging nuScenes benchmark, and extensive experimental results demonstrate the effectiveness of our designs.

We introduce a novel visual question answering (VQA) task in the context of autonomous driving, aiming to answer natural language questions based on street-view clues. Compared to traditional VQA tasks, VQA in autonomous driving scenario presents more challenges. Firstly, the raw visual data are multi-modal, including images and point clouds captured by camera and LiDAR, respectively. Secondly, the data are multi-frame due to the continuous, real-time acquisition. Thirdly, the outdoor scenes exhibit both moving foreground and static background. Existing VQA benchmarks fail to adequately address these complexities. To bridge this gap, we propose NuScenes-QA, the first benchmark for VQA in the autonomous driving scenario, encompassing 34K visual scenes and 460K question-answer pairs. Specifically, we leverage existing 3D detection annotations to generate scene graphs and design question templates manually. Subsequently, the question-answer pairs are generated programmatically based on these templates. Comprehensive statistics prove that our NuScenes-QA is a balanced large-scale benchmark with diverse question formats. Built upon it, we develop a series of baselines that employ advanced 3D detection and VQA techniques. Our extensive experiments highlight the challenges posed by this new task. Codes and dataset are available at https://github.com/qiantianwen/NuScenes-QA.

Recently, regression-based methods, which predict parameterized text shapes for text localization, have gained popularity in scene text detection. However, the existing parameterized text shape methods still have limitations in modeling arbitrary-shaped texts due to ignoring the utilization of text-specific shape information. Moreover, the time consumption of the entire pipeline has been largely overlooked, leading to a suboptimal overall inference speed. To address these issues, we first propose a novel parameterized text shape method based on low-rank approximation. Unlike other shape representation methods that employ data-irrelevant parameterization, our approach utilizes singular value decomposition and reconstructs the text shape using a few eigenvectors learned from labeled text contours. By exploring the shape correlation among different text contours, our method achieves consistency, compactness, simplicity, and robustness in shape representation. Next, we propose a dual assignment scheme for speed acceleration. It adopts a sparse assignment branch to accelerate the inference speed, and meanwhile, provides ample supervised signals for training through a dense assignment branch. Building upon these designs, we implement an accurate and efficient arbitrary-shaped text detector named LRANet. Extensive experiments are conducted on several challenging benchmarks, demonstrating the superior accuracy and efficiency of LRANet compared to state-of-the-art methods. Code is available at: https://github.com/ychensu/LRANet.git

Jun Zhan, Junqi Dai, Jiasheng Ye, Yunhua Zhou, Dong Zhang, Zhigeng Liu, Xin Zhang, Ruibin Yuan, Ge Zhang, Linyang Li, Hang Yan, Jie Fu, Tao Gui, Tianxiang Sun, Yu-Gang Jiang, Xipeng Qiu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Images suffer from heavy spatial redundancy because pixels in neighboring regions are spatially correlated. Existing approaches strive to overcome this limitation by reducing less meaningful image regions. However, current leading methods rely on supervisory signals. They may compel models to preserve content that aligns with labeled categories and discard content belonging to unlabeled categories. This categorical inductive bias makes these methods less effective in real-world scenarios. To address this issue, we propose a self-supervised framework for image redundancy reduction called Learning to Rank Patches (LTRP). We observe that image reconstruction of masked image modeling models is sensitive to the removal of visible patches when the masking ratio is high (e.g., 90%). Building upon it, we implement LTRP via two steps: inferring the semantic density score of each patch by quantifying variation between reconstructions with and without this patch, and learning to rank the patches with the pseudo score. The entire process is self-supervised, thus getting out of the dilemma of categorical inductive bias. We design extensive experiments on different datasets and tasks. The results demonstrate that LTRP outperforms both supervised and other self-supervised methods due to the fair assessment of image content. Code is available at https://github.com/irsLu/1trp.

We study text-based image editing (TBIE) of a single image by counterfactual inference because it is an elegant formulation to precisely address the requirement: the edited image should retain the fidelity of the original one. Through the lens of the formulation, we find that the crux of TBIE is that existing techniques hardly achieve a good tradeoff between editability and fidelity, mainly due to the overfitting of the single-image fine-tuning. To this end, we propose a Doubly Abductive Counterfactual inference framework (DAC). We first parameterize an exogenous variable as a UNet LoRA, whose abduction can encode all the image details. Second, we abduct another exogenous variable parameterized by a text encoder LoRA, which recovers the lost editability caused by the overfitted first abduction. Thanks to the second abduction, which exclusively encodes the visual transition from post-edit to pre-edit, its inversion-subtracting the LoRA-effectively reverts pre-edit back to post-edit, thereby accomplishing the edit. Through extensive experiments, our DAC achieves a good tradeoff between editability and fidelity. Thus, we can support a wide spectrum of user editing intents, including addition, removal, manipulation, replacement, style transfer, and facial change, which are extensively validated in both qualitative and quantitative evaluations. Codes are in https://github.com/xuesong39/DAC.

Existing diffusion-based video editing models have made gorgeous advances for editing attributes of a source video over time but struggle to manipulate the motion information while preserving the original protagonist's appearance and background. To address this, we propose MotionEditor, the first diffusion model for video motion editing. MotionEditor incorporates a novel content-aware motion adapter into ControlNet to capture temporal motion correspondence. While ControlNet enables direct generation based on skeleton poses, it encounters challenges when modifying the source motion in the inverted noise due to contradictory signals between the noise (source) and the condition (reference). Our adapter complements Control-Net by involving source content to transfer adapted control signals seamlessly. Further, we build up a two-branch ar-chitecture (a reconstruction branch and an editing branch) with a high-fidelity attention injection mechanism facilitating branch interaction. This mechanism enables the editing branch to query the key and value from the reconstruction branch in a decoupled manner, making the editing branch retain the original background and protagonist appearance. We also propose a skeleton alignment algorithm to address the discrepancies in pose size and position. Experiments demonstrate the promising motion editing ability of MotionEditor, both qualitatively and quantitatively. To the best of our knowledge, MotionEditor is the first to use diffusion models specifically for video motion editing, considering the origin dynamic background and camera movement.

The core of video understanding tasks, such as recognition, captioning, and tracking, is to automatically de-tect objects or actions in a video and analyze their temporal evolution. Despite sharing a common goal, different tasks often rely on distinct model architectures and annotation formats. In contrast, natural language processing benefits from a unified output space, i.e., text sequences, which simplifies the training of powerful foundational language models, such as GPT-3, with extensive training cor-pora. Inspired by this, we seek to unify the output space of video understanding tasks by using languages as labels and additionally introducing time and box tokens. In this way, a variety of video tasks could be formulated as video-grounded token generation. This enables us to address var-ious types of video tasks, including classification (such as action recognition), captioning (covering clip captioning, video question answering, and dense video captioning), and localization tasks (such as visual object tracking) within a fully shared encoder-decoder architecture, following a generative framework. Through comprehensive experiments, we demonstrate such a simple and straightforward idea is quite effective and can achieve state-of-the-art or compet-itive results on seven video benchmarks, providing a novel perspective for more universal video understanding. Code is available at https://github.com/wangjk666/OmniVid.

The recent wave of AI-generated content has witnessed the great development and success of Text-to-Image (T2I) technologies. By contrast, Text-to- Video (T2V) still falls short of expectations though attracting increasing interest. Existing works either train from scratch or adapt large T2I model to videos, both of which are computation and re-source expensive. In this work, we propose a Simple Dif-fusion Adapter (SimDA) that fine-tunes only 24M out of I.IB parameters of a strong T2I model, adapting it to video generation in a parameter-efficient way. In particular, we turn the T2I model for T2V by designing light-weight spatial and temporal adapters for transfer learning. Besides, we change the original spatial attention to the proposed Latent-Shift Attention (LSA) for temporal consistency. With a similar model architecture, we further train a video super-resolution model to generate high-definition (1024 x 1024) videos. In addition to T2V generation in the wild, SimDA could also be utilized in one-shot video editing with only 2 minutes tuning. Doing so, our method could minimize the training effort with extremely few tunable parameters for model adaptation.

Recent diffusion-based unrestricted attacks generate imperceptible adversarial examples with high transferability compared to previous unrestricted attacks and restricted attacks. However, existing works on diffusion-based unrestricted attacks are mostly focused on images yet are seldom explored in videos. In this paper, we propose the Recursive Token Merging for Video Diffusion-based Unrestricted Adversarial Attack (ReToMe-VA), which is the first framework to generate imperceptible adversarial video clips with higher transferability. Specifically, to achieve spatial imperceptibility, ReToMe-VA adopts a Timestep-wise Adversarial Latent Optimization (TALO) strategy that optimizes perturbations in diffusion models' latent space at each denoising step. TALO offers iterative and accurate updates to generate more powerful adversarial frames. TALO can further reduce memory consumption in gradient computation. Moreover, to achieve temporal imperceptibility, ReToMe-VA introduces a Recursive Token Merging (ReToMe) mechanism by matching and merging tokens across video frames in the self-attention module, resulting in temporally consistent adversarial videos. ReToMe concurrently facilitates inter-frame interactions into the attack process, inducing more diverse and robust gradients, thus leading to better adversarial transferability. Extensive experiments demonstrate the efficacy of ReToMe-VA, particularly in surpassing state-of-the-art attacks in adversarial transferability by more than 14.16% on average.

Current research in food analysis primarily concentrates on tasks such as food recognition, recipe retrieval and nutrition estimation from a single image. Nevertheless, there is a significant gap in exploring the impact of food intake on physiological indicators (e.g., weight) over time. This paper addresses this gap by introducing the DietDiary dataset, which encompasses daily dietary diaries and corresponding weight measurements of real users. Furthermore, we propose a novel task of weight prediction with a dietary diary that aims to leverage historical food intake and weight to predict future weights. To tackle this task, we propose a model-agnostic time series forecasting framework. Specifically, we introduce a Unified Meal Representation Learning (UMRL) module to extract representations for each meal. Additionally, we design a diet-aware loss function to associate food intake with weight variations. By conducting experiments on the DietDiary dataset with two state-of-the-art time series forecasting models, NLinear and iTransformer, we demonstrate that our proposed framework achieves superior performance compared to the original models. We make our dataset, code, and models publicly available at: https://yxg1005.github.io/weight-prediction.

The emergence of text-to-image generation models has led to the recognition that image enhancement, performed as post-processing, would significantly improve the visual quality of the generated images. Exploring diffusion models to enhance the generated images nevertheless is not trivial and necessitates to delicately enrich plentiful details while preserving the visual appearance of key content in the original image. In this paper, we propose a novel framework, namely FreeEnhance, for content-consistent image enhancement using the off-the-shelf image diffusion models. Technically, FreeEnhance is a two-stage process that firstly adds random noise to the input image and then capitalizes on a pre-trained image diffusion model (i.e., Latent Diffusion Models) to denoise and enhance the image details. In the noising stage, FreeEnhance is devised to add lighter noise to the region with higher frequency to preserve the high-frequent patterns (e.g., edge, corner) in the original image. In the denoising stage, we present three target properties as constraints to regularize the predicted noise, enhancing images with high acutance and high visual quality. Extensive experiments conducted on the HPDv2 dataset demonstrate that our FreeEnhance outperforms the state-of-the-art image enhancement models in terms of quantitative metrics and human preference. More remarkably, FreeEnhance also shows higher human preference compared to the commercial image enhancement solution of Magnific AI.

Recent advancements in Large Vision-Language Models (VLMs) have underscored their superiority in various multimodal tasks. However, the adversarial robustness of VLMs has not been fully explored. Existing methods mainly assess robustness through unimodal adversarial attacks that perturb images, while assuming inherent resilience against text-based attacks. Different from existing attacks, in this work we propose a more comprehensive strategy that jointly attacks both text and image modalities to exploit a broader spectrum of vulnerability within VLMs. Specifically, we propose a dual optimization objective aimed at guiding the model to generate highly toxic affirmative responses. Our attack method begins by optimizing an adversarial image prefix from random noise to generate diverse harmful responses in the absence of text input, thus imbuing the image with toxic semantics. Subsequently, an adversarial text suffix is integrated and co-optimized with the adversarial image prefix to maximize the probability of eliciting affirmative responses to various harmful instructions. The discovered adversarial image prefix and text suffix are collectively denoted as a Universal Master Key (UMK). When integrated into various malicious queries, UMK can circumvent the alignment defenses of VLMs and lead to the generation of objectionable content, known as jailbreaks. The experimental results demonstrate that our universal attack strategy can effectively jailbreak MiniGPT-4 with a 96% success rate, highlighting the fragility of VLMs and the exigency for new alignment strategies. Codes are available at https://github.com/roywang021/UMK. Disclaimer: This paper contains potentially disturbing and offensive content.

Pre-trained Vision-Language Models (VLMs) have shown great ability in various Vision-Language tasks. However, these VLMs exhibit inherent vulnerabilities to transferable adversarial examples, which could potentially undermine their performance and reliability in real-world applications. Cross-modal interactions have been demonstrated to be the key point to boosting adversarial transferability, but the utilization of them is limited in existing multimodal adversarial attacks. Stable Diffusion, which contains multiple cross-attention modules, possesses great potential in facilitating adversarial transferability by leveraging abundant cross-modal interactions. Therefore, We propose a Multimodal Diffusion-based Attack (MDA), which conducts adversarial attacks against VLMs using Stable Diffusion. Specifically, MDA initially generates adversarial text, which is subsequently utilized to optimize the adversarial image during the diffusion process. Besides leveraging adversarial text in calculating downstream loss, MDA also takes it as the guiding prompt in adversarial image generation during the denoising process, which enriches the ways of cross-modal interactions, thus strengthening the adversarial transferability. Compared with pixel-based attacks, MDA introduces perturbations in the latent space rather than pixel space to manipulate high-level semantics, which is also beneficial to improving adversarial transferability. Experimental results demonstrate that the adversarial examples generated by MDA are highly transferable across different VLMs on different downstream tasks, surpassing state-of-the-art methods by a large margin.

Recent advancements in "deepfake" techniques have paved the way for generating various media forgeries. In response to the potential hazards of these media forgeries, many researchers engage in exploring detection methods, increasing the demand for high-quality media forgery datasets. Despite this, existing datasets have certain limitations. Firstly, most datasets focus on manipulating visual modality and usually lack diversity, as only a few forgery approaches are considered. Secondly, the quality of media is often inadequate in clarity and naturalness. Meanwhile, the size of the dataset is also limited. Thirdly, it is commonly observed that real-world forgeries are motivated by identity, yet the identity information of the individuals portrayed in these forgeries within existing datasets remains under-explored. For detection, identity information could be an essential clue to boost performance. Moreover, official media concerning relevant identities on the Internet can serve as prior knowledge, aiding both the audience and forgery detectors in determining the true identity. Therefore, we propose an identity-driven multimedia forgery dataset, IDForge, which contains 249,138 video shots sourced from 324 wild videos of 54 celebrities collected from the Internet. The fake video shots involve 9 types of manipulation across visual, audio, and textual modalities. Additionally, IDForge provides extra 214,438 real video shots as a reference set for the 54 celebrities. Correspondingly, we propose the Reference-assisted Multimodal Forgery Detection Network (R-MFDN), aiming at the detection of deepfake videos. Through extensive experiments on the proposed dataset, we demonstrate the effectiveness of R-MFDN on the multimedia detection task. The dataset is available at: https://github.com/xyyandxyy/IDForge.

Scene text recognition (STR) pre-training methods have achieved remarkable progress, primarily relying on synthetic datasets. However, the domain gap between synthetic and real images poses a challenge in acquiring feature representations that align well with images on real scenes, thereby limiting the performance of these methods. We note that vision-language models like CLIP, pre-trained on extensive real image-text pairs, effectively align images and text in a unified embedding space, suggesting the potential to derive the representations of real images from text alone. Building upon this premise, we introduce a novel method named Decoder Pre-training with only text for STR (DPTR). DPTR treats text embeddings produced by the CLIP text encoder as pseudo visual embeddings and uses them to pre-train the decoder. An Offline Randomized Perturbation (ORP) strategy is introduced. It enriches the diversity of text embeddings by incorporating natural image embeddings extracted from the CLIP image encoder, effectively directing the decoder to acquire the potential representations of real images. In addition, we introduce a Feature Merge Unit (FMU) that guides the extracted visual embeddings focusing on the character foreground within the text image, thereby enabling the pre-trained decoder to work more efficiently and accurately. Extensive experiments across various STR decoders and language recognition tasks underscore the broad applicability and remarkable performance of DPTR, providing a novel insight for STR pre-training. Code is available at https://github.com/Topdu/OpenOCR.

Yixu Wang, Yan Teng, Kexin Huang, Chengqi Lyu, Songyang Zhang, Wenwei Zhang, Xingjun Ma, Yu-Gang Jiang, Yu Qiao, Yingchun Wang. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Large Multimodal Model (LMM) is a hot research topic in the computer vision area and has also demonstrated remarkable potential across multiple disciplinary fields. A recent trend is to further extend and enhance the perception capabilities of LMMs. The current methods follow the paradigm of adapting the visual task outputs to the format of the language model, which is the main component of a LMM. This adaptation leads to convenient development of such LMMs with minimal modifications, however, it overlooks the intrinsic characteristics of diverse visual tasks and hinders the learning of perception capabilities. To address this issue, we propose a novel LMM architecture named Lumen, a Large multimodal model with versatile vision-centric capability enhancement. We decouple the LMM's learning of perception capabilities into task-agnostic and task-specific stages. Lumen first promotes fine-grained vision-language concept alignment, which is the fundamental capability for various visual tasks. Thus the output of the task-agnostic stage is a shared representation for all the tasks we address in this paper. Then the task-specific decoding is carried out by flexibly routing the shared representation to lightweight task decoders with negligible training efforts. Comprehensive experimental results on a series of vision-centric and VQA benchmarks indicate that our Lumen model not only achieves or surpasses the performance of existing LMM-based approaches in a range of vision-centric tasks while maintaining general visual understanding and instruction following capabilities.

Understanding documents with rich layouts and multi-modal components is a long-standing and practical task. Recent Large Vision-Language Models (LVLMs) have made remarkable strides in various tasks, particularly in single-page document understanding (DU). However, their abilities on long-context DU remain an open problem. This work presents MMLONGBENCH-DOC, a long-context, multi- modal benchmark comprising 1,082 expert-annotated questions. Distinct from previous datasets, it is constructed upon 135 lengthy PDF-formatted documents with an average of 47.5 pages and 21,214 textual tokens. Towards comprehensive evaluation, answers to these questions rely on pieces of evidence from (1) different sources (text, image, chart, table, and layout structure) and (2) various locations (i.e., page number). Moreover, 33.7\% of the questions are cross-page questions requiring evidence across multiple pages. 20.6\% of the questions are designed to be unanswerable for detecting potential hallucinations. Experiments on 14 LVLMs demonstrate that long-context DU greatly challenges current models. Notably, the best-performing model, GPT-4o, achieves an F1 score of only 44.9\%, while the second-best, GPT-4V, scores 30.5\%. Furthermore, 12 LVLMs (all except GPT-4o and GPT-4V) even present worse performance than their LLM counterparts which are fed with lossy-parsed OCR documents. These results validate the necessity of future research toward more capable long-context LVLMs.

Most large multimodal models (LMMs) are implemented by feeding visual tokens as a sequence into the first layer of a large language model (LLM). The resulting architecture is simple but significantly increases computation and memory costs, as it has to handle a large number of additional tokens in its input layer. This paper presents a new architecture *DeepStack* for LMMs. Considering $N$ layers in the language and vision transformer of LMMs, we stack the visual tokens into $N$ groups and feed each group to its aligned transformer layer from bottom to top. Surprisingly, this simple method greatly enhances the power of LMMs to model interactions among visual tokens across layers but with minimal additional cost. We apply *DeepStack* to both language and vision transformer in LMMs, and validate the effectiveness of *DeepStack* LMMs with extensive empirical results. Using the same context length, our DeepStack 7B and 13B parameters surpass their counterparts by 2.7 and 2.9 on average across 9 benchmarks, respectively. Using only one-fifth of the context length, DeepStack rivals closely to the counterparts that use the full context length. These gains are particularly pronounced on high-resolution tasks, *e.g.*, 4.2, 11.0, and 4.0 improvements on TextVQA, DocVQA, and InfoVQA compared to LLaVA-1.5-7B, respectively. We further apply *DeepStack* to vision transformer layers, which brings us a similar amount of improvements, 3.8 on average compared with LLaVA-1.5-7B.

Image segmentation is a crucial vision task that groups pixels within an image into semantically meaningful segments, which is pivotal in obtaining a fine-grained understanding of real-world scenes. However, an increasing privacy concern exists regarding training large-scale image segmentation models on unauthorized private data. In this work, we exploit the concept of unlearnable examples to make images unusable to model training by generating and adding unlearnable noise into the original images. Particularly, we propose a novel Unlearnable Segmentation (UnSeg) framework to train a universal unlearnable noise generator that is capable of transforming any downstream images into their unlearnable version. The unlearnable noise generator is finetuned from the Segment Anything Model (SAM) via bilevel optimization on an interactive segmentation dataset towards minimizing the training error of a surrogate model that shares the same architecture with SAM (but trains from scratch). We empirically verify the effectiveness of UnSeg across 6 mainstream image segmentation tasks, 10 widely used datasets, and 7 different network architectures, and show that the unlearnable images can reduce the segmentation performance by a large margin. Our work provides useful insights into how to leverage foundation models in a data-efficient and computationally affordable manner to protect images against image segmentation models.

Tokenizer, serving as a translator to map the intricate visual data into a compact latent space, lies at the core of visual generative models. Based on the finding that existing tokenizers are tailored to either image or video inputs, this paper presents OmniTokenizer, a transformer-based tokenizer for joint image and video tokenization. OmniTokenizer is designed with a spatial-temporal decoupled architecture, which integrates window attention and causal attention for spatial and temporal modeling, respectively. To exploit the complementary nature of image and video data, we further propose a progressive training strategy, where OmniTokenizer is first trained on image data on a fixed resolution to develop the spatial encoding capacity and then jointly trained on image and video data on multiple resolutions to learn the temporal dynamics. OmniTokenizer, for the first time, handles both image and video inputs within a unified framework and proves the possibility of realizing their synergy. Extensive experiments demonstrate that OmniTokenizer achieves state-of-the-art (SOTA) reconstruction performance on various image and video datasets, e.g., 1.11 reconstruction FID on ImageNet and 42 reconstruction FVD on UCF-101, beating the previous SOTA methods by 13% and 26%, respectively. Additionally, we also show that when integrated with OmniTokenizer, both language model-based approaches and diffusion models can realize advanced visual synthesis performance, underscoring the superiority and versatility of our method.

Video diffusion models are able to generate high-quality videos by learning strong spatial-temporal priors on large-scale datasets. In this paper, we aim to investigate whether such priors derived from a generative process are suitable for video recognition, and eventually joint optimization of generation and recognition. Building upon Stable Video Diffusion, we introduce GenRec, the first unified framework trained with a random-frame conditioning process so as to learn generalized spatial-temporal representations. The resulting framework can naturally supports generation and recognition, and more importantly is robust even when visual inputs contain limited information. Extensive experiments demonstrate the efficacy of GenRec for both recognition and generation. In particular, GenRec achieves competitive recognition performance, offering 75.8% and 87.2% accuracy on SSV2 and K400, respectively. GenRec also performs the best on class-conditioned image-to-video generation, achieving 46.5 and 49.3 FVD scores on SSV2 and EK-100 datasets. Furthermore, GenRec demonstrates extraordinary robustness in scenarios that only limited frames can be observed. Code will be available at https://github.com/wengzejia1/GenRec.

The cross-model transferability of adversarial examples makes black-box attacks to be practical. However, it typically requires access to the input of the same modality as black-box models to attain reliable transferability. Unfortunately, the collection of datasets may be difficult in security-critical scenarios. Hence, developing cross-modal attacks for fooling models with different modalities of inputs would highly threaten real-world DNNs applications. The above considerations motivate us to investigate cross-modal transferability of adversarial examples. In particular, we aim to generate video adversarial examples from white-box image models to attack video CNN and ViT models. We introduce the Image To Video (I2V) attack based on the observation that image and video models share similar low-level features. For each video frame, I2V optimizes perturbations by reducing the similarity of intermediate features between benign and adversarial frames on image models. Then I2V combines adversarial frames together to generate video adversarial examples. I2V can be easily extended to simultaneously perturb multi-layer features extracted from an ensemble of image models. To efficiently integrate various features, we introduce an adaptive approach to re-weight the contributions of each layer based on its cosine similarity values of the previous attack step. Experimental results demonstrate the effectiveness of the proposed method.

Despite significant results achieved by Contrastive Language-Image Pretraining (CLIP) in zero-shot image recognition, limited effort has been made exploring its potential for zero-shot video recognition. This paper presents Open-VCLIP++, a simple yet effective framework that adapts CLIP to a strong zero-shot video classifier, capable of identifying novel actions and events during testing. Open-VCLIP++ minimally modifies CLIP to capture spatial-temporal relationships in videos, thereby creating a specialized video classifier while striving for generalization. We formally demonstrate that training Open-VCLIP++ is tantamount to continual learning with zero historical data. To address this problem, we introduce Interpolated Weight Optimization, a technique that leverages the advantages of weight interpolation during both training and testing. Furthermore, we build upon large language models to produce fine-grained video descriptions. These detailed descriptions are further aligned with video features, facilitating a better transfer of CLIP to the video domain. Our approach is evaluated on three widely used action recognition datasets, following a variety of zero-shot evaluation protocols. The results demonstrate that our method surpasses existing state-of-the-art techniques by significant margins. Specifically, we achieve zero-shot accuracy scores of 88.1%, 58.7%, and 81.2% on UCF, HMDB, and Kinetics-600 datasets respectively, outpacing the best-performing alternative methods by 8.5%, 8.2%, and 12.3%. We also evaluate our approach on the MSR-VTT video-text retrieval dataset, where it delivers competitive video-to-text and text-to-video retrieval performance, while utilizing substantially less fine-tuning data compared to other methods.

Video question answering (VideoQA) is an essential task in vision-language understanding, which has attracted numerous research attention recently. Nevertheless, existing works mostly achieve promising performances on short videos of duration within 15 seconds. For VideoQA on minute-level long-term videos, those methods are likely to fail because of lacking the ability to deal with noise and redundancy caused by scene changes and multiple actions in the video. Considering the fact that the question often remains concentrated in a short temporal range, we propose to first locate the question to a segment in the video and then infer the answer using the located segment only. Under this scheme, we propose “Locate before Answering” (LocAns), a novel approach that integrates a question localization module and an answer prediction module into an end-to-end model. During the training phase, the available answer label not only serves as the supervision signal of the answer prediction module, but also is used to generate pseudo temporal labels for the question localization module. Moreover, we design a decoupled alternative training strategy to update the two modules separately. In the experiments, LocAns achieves state-of-the-art performance on three modern long-term VideoQA datasets, NExT-QA, ActivityNet-QA, and AGQA. Its qualitative examples show the reliable performance of the question localization.

Videos are multimodal in nature. Conventional video recognition pipelines typically fuse multimodal features for improved performance. However, this is not only computationally expensive but also neglects the fact that different videos rely on different modalities for predictions. This article introduces Hierarchical and Conditional Modality Selection (HCMS), a simple yet efficient multimodal learning framework for efficient video recognition. HCMS operates on a low-cost modality, i.e., audio clues, by default, and dynamically decides on-the-fly whether to use computationally expensive modalities, including appearance and motion clues, on a per-input basis. This is achieved by the collaboration of three LSTMs that are organized in a hierarchical manner. In particular, LSTMs that operate on high-cost modalities contain a gating module, which takes as inputs lower-level features and historical information to adaptively determine whether to activate its corresponding modality; otherwise, it simply reuses historical information. We conduct extensive experiments on two large-scale video benchmarks, FCVID and ActivityNet, and the results demonstrate the proposed approach can effectively explore multimodal information for improved classification performance while requiring much less computation.

3D object detection in autonomous driving aims to reason “what” and “where” the objects of interest present in a 3D world. Following the conventional wisdom of previous 2D object detection, existing methods often adopt the canonical Cartesian coordinate system with perpendicular axis. However, we conjugate that this does not fit the nature of the ego car’s perspective, as each onboard camera perceives the world in shape of wedge intrinsic to the imaging geometry with radical (non perpendicular) axis. Hence, in this paper we advocate the exploitation of the Polar coordinate system and propose a new Polar Transformer (PolarFormer) for more accurate 3D object detection in the bird’s-eye-view (BEV) taking as input only multi-camera 2D images. Specifically, we design a cross-attention based Polar detection head without restriction to the shape of input structure to deal with irregular Polar grids. For tackling the unconstrained object scale variations along Polar’s distance dimension, we further introduce a multi-scale Polar representation learning strategy. As a result, our model can make best use of the Polar representation rasterized via attending to the corresponding image observation in a sequence-to-sequence fashion subject to the geometric constraints. Thorough experiments on the nuScenes dataset demonstrate that our PolarFormer outperforms significantly state-of-the-art 3D object detection alternatives.

There is a growing interest in developing unlearnable examples (UEs) against visual privacy leaks on the Internet. UEs are training samples added with invisible but unlearnable noise, which have been found can prevent unauthorized training of machine learning models. UEs typically are generated via a bilevel optimization framework with a surrogate model to remove (minimize) errors from the original samples, and then applied to protect the data against unknown target models. However, existing UE generation methods all rely on an ideal assumption called label-consistency, where the hackers and protectors are assumed to hold the same label for a given sample. In this work, we propose and promote a more practical label-agnostic setting, where the hackers may exploit the protected data quite differently from the protectors. E.g., amclass unlearnable dataset held by the protector may be exploited by the hacker as a n-class dataset. Existing UE generation methods are rendered ineffective in this challenging setting. To tackle this challenge, we present a novel technique called Unlearnable Clusters (UCs) to generate label-agnostic unlearnable examples with cluster-wise perturbations. Furthermore, we propose to leverage Vision-and-Language Pre-trained Models (VLPMs) like CLIP as the surrogate model to improve the transferability of the crafted UCs to diverse domains. We empirically verify the effectiveness of our proposed approach under a variety of settings with different datasets, target models, and even commercial platforms Microsoft Azure and Baidu PaddlePaddle. Code is available at https://github.com/jiamingzhang94/Unlearnable-Clusters.

Anomaly detection and localization are widely used in industrial manufacturing for its efficiency and effectiveness. Anomalies are rare and hard to collect and supervised models easily over-fit to these seen anomalies with a handful of abnormal samples, producing unsatisfactory performance. On the other hand, anomalies are typically subtle, hard to discern, and of various appearance, making it difficult to detect anomalies and let alone locate anomalous regions. To address these issues, we propose a framework called Prototypical Residual Network (PRN), which learns feature residuals of varying scales and sizes between anomalous and normal patterns to accurately reconstruct the segmentation maps of anomalous regions. PRN mainly consists of two parts: multi-scale prototypes that explicitly represent the residual features of anomalies to normal patterns; a multisize self-attention mechanism that enables variable-sized anomalous feature learning. Besides, we present a variety of anomaly generation strategies that consider both seen and unseen appearance variance to enlarge and diversify anomalies. Extensive experiments on the challenging and widely used MVTec AD benchmark show that PRN outperforms current state-of-the-art unsupervised and supervised methods. We further report SOTA results on three additional datasets to demonstrate the effectiveness and generalizability of PRN.

Cross-Domain Few-Shot Learning (CD-FSL) is a recently emerging task that tackles few-shot learning across different domains. It aims at transferring prior knowledge learned on the source dataset to novel target datasets. The CD-FSL task is especially challenged by the huge domain gap between different datasets. Critically, such a domain gap actually comes from the changes of visual styles, and wave-SAN [10] empirically shows that spanning the style distribution of the source data helps alleviate this issue. However, wave-SAN simply swaps styles of two images. Such a vanilla operation makes the generated styles “real” and “easy”, which still fall into the original set of the source styles. Thus, inspired by vanilla adversarial learning, a novel model-agnostic meta Style Adversarial training (StyleAdv) method together with a novel style adversarial attack method is proposed for CD-FSL. Particularly, our style attack method synthesizes both “virtual” and “hard” adversarial styles for model training. This is achieved by perturbing the original style with the signed style gradients. By continually attacking styles and forcing the model to recognize these challenging adversarial styles, our model is gradually robust to the visual styles, thus boosting the generalization ability for novel target datasets. Besides the typical CNN-based backbone, we also employ our StyleAdv method on large-scale pre-trained vision transformer. Extensive experiments conducted on eight various target datasets show the effectiveness of our method. Whether built upon ResNet or ViT, we achieve the new state of the art for CD-FSL. Code is available at https://github.com/lovelyqian/StyleAdv-CDFSL.

Fusing LiDAR and camera information is essential for accurate and reliable 3D object detection in autonomous driving systems. This is challenging due to the difficulty of combining multi-granularity geometric and semantic features from two drastically different modalities. Recent approaches aim at exploring the semantic densities of camera features through lifting points in 2D camera images (referred to as “seeds”) into 3D space, and then incorporate 2D semantics via cross-modal interaction or fusion techniques. However, depth information is under-investigated in these approaches when lifting points into 3D space, thus 2D semantics can not be reliably fused with 3D points. Moreover, their multi-modal fusion strategy, which is implemented as concatenation or attention, either can not effectively fuse 2D and 3D information or is unable to perform fine-grained interactions in the voxel space. To this end, we propose a novel framework with better utilization of the depth information and fine-grained cross-modal interaction between LiDAR and camera, which consists of two important components. First, a Multi-Depth Unprojection (MDU) method is used to enhance the depth quality of the lifted points at each interaction level. Second, a Gated Modality-Aware Convolution (GMA-Conv) block is applied to modulate voxels involved with the camera modality in a fine-grained manner and then aggregate multi-modal features into a unified space. Together they provide the detection head with more comprehensive features from LiDAR and camera. On the nuScenes test benchmark, our proposed method, abbreviated as MSMD-Fusion, achieves state-of-the-art results on both 3D object detection and tracking tasks without using test-time-augmentation and ensemble techniques. The code is available at https://github.com/SxJyJay/MSMDFusion.

Combining multiple datasets enables performance boost on many computer vision tasks. But similar trend has not been witnessed in object detection when combining multiple datasets due to two inconsistencies among detection datasets: taxonomy difference and domain gap. In this paper, we address these challenges by a new design (named Detection Hub) that is dataset-aware and category-aligned. It not only mitigates the dataset inconsistency but also provides coherent guidance for the detector to learn across multiple datasets. In particular, the dataset-aware design is achieved by learning a dataset embedding that is used to adapt object queries as well as convolutional kernels in detection heads. The categories across datasets are semantically aligned into a unified space by replacing one-hot category representations with word embedding and leveraging the semantic coherence of language embedding. Detection Hub fulfills the benefits of large data on object detection. Experiments demonstrate that joint training on multiple datasets achieves significant performance gains over training on each dataset alone. Detection Hub further achieves SoTA performance on UODB benchmark with wide variety of datasets.

The cost of pathological examination makes virtual restaining of pathological images meaningful. However, due to the ultra-high resolution of pathological images, traditional virtual restaining methods have to divide a WSI image into patches for model training and inference. Such a limitation leads to the lack of global information, resulting in observable differences in color, brightness and contrast when the re-stained patches are merged to generate an image of larger size. We summarize this issue as the square effect. Some existing methods try to solve this issue through overlapping between patches or simple postprocessing. But the former one is not that effective, while the latter one requires carefully tuning. In order to eliminate the square effect, we design a bidirectional feature fusion generative adversarial network (BFF-GAN) with a global branch and a local branch. It learns the inter-patch connections through the fusion of global and local features plus patch-wise attention. We perform experiments on both the private dataset RCC and the public dataset ANHIR. The results show that our model achieves competitive performance and is able to generate extremely real images that are deceptive even for experienced pathologists, which means it is of great clinical significance.

Vision Transformers (ViTs) have achieved overwhelming success, yet they suffer from vulnerable resolution scalability, i.e., the performance drops drastically when presented with input resolutions that are unseen during training. We introduce, ResFormer, a framework that is built upon the seminal idea of multi-resolution training for improved performance on a wide spectrum of, mostly unseen, testing resolutions. In particular, ResFormer operates on replicated images of different resolutions and enforces a scale consistency loss to engage interactive information across different scales. More importantly, to alternate among varying resolutions effectively, especially novel ones in testing, we propose a global-local positional embedding strategy that changes smoothly conditioned on input sizes. We conduct extensive experiments for image classification on ImageNet. The results provide strong quantitative evidence that ResFormer has promising scaling abilities towards a wide range of resolutions. For instance, ResFormer-B-MR achieves a Top-1 accuracy of 75.86% and 81.72% when evaluated on relatively low and high resolutions respectively (i.e., 96 and 640), which are 48% and 7.49% better than DeiT-B. We also demonstrate, moreover, ResFormer is flexible and can be easily extended to semantic segmentation, object detection and video action recognition.

Benefiting from masked visual modeling, self-supervised video representation learning has achieved remarkable progress. However, existing methods focus on learning representations from scratch through reconstructing low-level features like raw pixel values. In this paper, we propose masked video distillation (MVD), a simple yet effective two-stage masked feature modeling framework for video representation learning: firstly we pretrain an image (or video) model by recovering low-level features of masked patches, then we use the resulting features as targets for masked feature modeling. For the choice of teacher models, we observe that students taught by video teachers perform better on temporally-heavy video tasks, while image teachers transfer stronger spatial representations for spatially-heavy video tasks. Visualization analysis also indicates different teachers produce different learned patterns for students. To leverage the advantage of different teachers, we design a spatial-temporal co-teaching method for MVD. Specifically, we distill student models from both video teachers and image teachers by masked feature modeling. Extensive experimental results demonstrate that video transformers pre-trained with spatial-temporal co-teaching outperform models distilled with a single teacher on a multitude of video datasets. Our MVD with vanilla ViT achieves state-of-the-art performance compared with previous methods on several challenging video downstream tasks. For example, with the ViT-Large model, our MVD achieves 86.4% and 76.7% Top-1 accuracy on Kinetics-400 and Something-Something-v2, outperforming VideoMAE by 1.2% and 2.4% respectively. When a larger ViT-Huge model is adopted, MVD achieves the state-of-the-art performance with 77.3% Top-1 accuracy on Something-Something-v2. Code will be available at https://github.com/ruiwang2021/mvd.

Exploring dense matching between the current frame and past frames for long-range context modeling, memory-based methods have demonstrated impressive results in video object segmentation (VOS) recently. Nevertheless, due to the lack of instance understanding ability, the above approaches are oftentimes brittle to large appearance variations or viewpoint changes resulted from the movement of objects and cameras. In this paper, we argue that instance understanding matters in VOS, and integrating it with memory-based matching can enjoy the synergy, which is intuitively sensible from the definition of VOS task, i.e., identifying and segmenting object instances within the video. Towards this goal, we present a two-branch network for VOS, where the query-based instance segmentation (IS) branch delves into the instance details of the current frame and the VOS branch performs spatial-temporal matching with the memory bank. We employ the well-learned object queries from IS branch to inject instance-specific information into the query key, with which the instance-augmented matching is further performed. In addition, we introduce a multi-path fusion block to effectively combine the memory readout with multi-scale features from the instance segmentation decoder, which incorporates high-resolution instance-aware features to produce final segmentation results. Our method achieves state-of-the-art performance on DAVIS 2016/2017 val (92.6% and 87.1%), DAVIS 2017 test-dev (82.8%), and YouTube-VOS 2018/2019 val (86.3% and 86.3%), outperforming alternative methods by clear margins.

In this paper, we propose a novel transfer-based targeted attack method that optimizes the adversarial perturbations without any extra training efforts for auxiliary networks on training data. Our new attack method is proposed based on the observation that highly universal adversarial perturbations tend to be more transferable for targeted attacks. Therefore, we propose to make the perturbation to be agnostic to different local regions within one image, which we called as self-universality. Instead of optimizing the perturbations on different images, optimizing on different regions to achieve self-universality can get rid of using extra data. Specifically, we introduce a feature similarity loss that encourages the learned perturbations to be universal by maximizing the feature similarity between adversarial perturbed global images and randomly cropped local regions. With the feature similarity loss, our method makes the features from adversarial perturbations to be more dominant than that of benign images, hence improving targeted transferability. We name the proposed attack method as Self-Universality (SU) attack. Extensive experiments demonstrate that SU can achieve high success rates for transfer-based targeted attacks. On ImageNet-compatible dataset, SU yields an improvement of 12% compared with existing state-of-the-art methods. Code is available at https://github.com/zhipengwei/Self-Universality.

Semi-supervised action recognition is a challenging but critical task due to the high cost of video annotations. Existing approaches mainly use convolutional neural networks, yet current revolutionary vision transformer models have been less explored. In this paper, we investigate the use of transformer models under the SSL setting for action recognition. To this end, we introduce SVFormer, which adopts a steady pseudo-labeling framework (i.e., EMA-Teacher) to cope with unlabeled video samples. While a wide range of data augmentations have been shown effective for semi-supervised image classification, they generally produce limited results for video recognition. We therefore introduce a novel augmentation strategy, Tube Token-Mix, tailored for video data where video clips are mixed via a mask with consistent masked tokens over the temporal axis. In addition, we propose a temporal warping augmentation to cover the complex temporal variation in videos, which stretches selected frames to various temporal durations in the clip. Extensive experiments on three datasets Kinetics-400, UCF-101, and HMDB-51 verify the advantage of SVFormer. In particular, SVFormer outperforms the state-of-the-art by 31.5% with fewer training epochs under the 1% labeling rate of Kinetics-400. Our method can hopefully serve as a strong benchmark and encourage future search on semi-supervised action recognition with Transformer networks. Code is released at https://github.com/ChenHsing/SVFormer.

Contrastive language-image pretraining (CLIP) has demonstrated remarkable success in various image tasks. However, how to extend CLIP with effective temporal modeling is still an open and crucial problem. Existing factorized or joint spatial-temporal modeling trades off between the efficiency and performance. While modeling temporal information within straight through tube is widely adopted in literature, we find that simple frame alignment already provides enough essence without temporal attention. To this end, in this paper, we proposed a novel Implicit Learnable Alignment (ILA) method, which minimizes the temporal modeling effort while achieving incredibly high performance. Specifically, for a frame pair, an interactive point is predicted in each frame, serving as a mutual information rich region. By enhancing the features around the interactive point, two frames are implicitly aligned. The aligned features are then pooled into a single token, which is leveraged in the subsequent spatial self-attention. Our method allows eliminating the costly or insufficient temporal self-attention in video. Extensive experiments on benchmarks demonstrate the superiority and generality of our module. Particularly, the proposed ILA achieves a top-1 accuracy of 88.7% on Kinetics-400 with much fewer FLOPs compared with Swin-L and ViViT-H. Code is released at https://github.com/Francis-Rings/ILA.

Multilingual text recognition (MLTR) systems typically focus on a fixed set of languages, which makes it difficult to handle newly added languages or adapt to ever-changing data distribution. In this paper, we propose the Incremental MLTR (IMLTR) task in the context of incremental learning (IL), where different languages are introduced in batches. IMLTR is particularly challenging due to rehearsal-imbalance, which refers to the uneven distribution of sample characters in the rehearsal set, used to retain a small amount of old data as past memories. To address this issue, we propose a Multiplexed Routing Network (MRN). MRN trains a recognizer for each language that is currently seen. Subsequently, a language domain predictor is learned based on the rehearsal set to weigh the recognizers. Since the recognizers are derived from the original data, MRN effectively reduces the reliance on older data and better fights against catastrophic forgetting, the core issue in IL. We extensively evaluate MRN on MLT17 and MLT19 datasets. It outperforms existing general-purpose IL methods by large margins, with average accuracy improvements ranging from 10.3% to 35.8% under different settings. Code is available at https://github.com/simplify23/MRN.

Vision-language pre-trained models (e.g., CLIP) trained on large-scale datasets via self-supervised learning, are drawing increasing research attention since they can achieve superior performances on multi-modal downstream tasks. Nevertheless, we find that the adversarial perturbations crafted on vision-language pre-trained models can be used to attack different corresponding downstream task models. Specifically, to investigate such adversarial transferability, we introduce a task-agnostic method named Global and Local Augmentation (GLA) attack to generate highly transferable adversarial examples on CLIP, to attack black-box downstream task models. GLA adopts random crop and resize at both global and local patch levels, to create more diversity and make adversarial noises robust. Then GLA generates the adversarial perturbations by minimizing the cosine similarity between intermediate features from augmented adversarial and benign examples. Extensive experiments on three CLIP image encoders with different backbones and three different downstream tasks demonstrate the superiority of our method compared with other strong baselines. The code is available at https://github.com/yqlvcoding/GLAattack.

Neural networks for visual content understanding have recently evolved from convolutional ones to transformers. The prior (CNN) relies on small-windowed kernels to capture the regional clues, demonstrating solid local expressiveness. On the contrary, the latter (transformer) establishes long-range global connections between localities for holistic learning. Inspired by this complementary nature, there is a growing interest in designing hybrid models which utilize both techniques. Current hybrids merely replace convolutions as simple approximations of linear projection or juxtapose a convolution branch with attention without considering the importance of local/global modeling. To tackle this, we propose a new hybrid named Adaptive Split-Fusion Transformer (ASF-former) that treats convolutional and attention branches differently with adaptive weights. Specifically, an ASF-former encoder equally splits feature channels into half to fit dual-path inputs. Then, the outputs of the dual-path are fused with weights calculated from visual cues. We also design a compact convolutional path from a concern of efficiency. Extensive experiments on standard benchmarks show that our ASF-former outperforms its CNN, transformer, and hybrid counterparts in terms of accuracy (83.9% on ImageNet-1K), under similar conditions (12.9G MACs / 56.7M Params, without large-scale pre-training). The code is available at: https://github.com/szx503045266/ASF-former.

Deep neural networks (DNNs) have been found to be vulnerable to backdoor attacks, raising security concerns about their deployment in mission-critical applications. While existing defense methods have demonstrated promising results, it is still not clear how to effectively remove backdoor-associated neurons in backdoored DNNs. In this paper, we propose a novel defense called *Reconstructive Neuron Pruning* (RNP) to expose and prune backdoor neurons via an unlearning and then recovering process. Specifically, RNP first unlearns the neurons by maximizing the model's error on a small subset of clean samples and then recovers the neurons by minimizing the model's error on the same data. In RNP, unlearning is operated at the neuron level while recovering is operated at the filter level, forming an asymmetric reconstructive learning procedure. We show that such an asymmetric process on only a few clean samples can effectively expose and prune the backdoor neurons implanted by a wide range of attacks, achieving a new state-of-the-art defense performance. Moreover, the unlearned model at the intermediate step of our RNP can be directly used to improve other backdoor defense tasks including backdoor removal, trigger recovery, backdoor label detection, and backdoor sample detection. Code is available at https://github.com/bboylyg/RNP.

Contrastive Language-Image Pretraining (CLIP) has demonstrated impressive zero-shot learning abilities for image understanding, yet limited effort has been made to investigate CLIP for zero-shot video recognition. We introduce Open-VCLIP, a simple yet effective approach that transforms CLIP into a strong zero-shot video classifier that can recognize unseen actions and events at test time. Our framework extends CLIP with minimal modifications to model spatial-temporal relationships in videos, making it a specialized video classifier, while striving for generalization. We formally show that training an Open-VCLIP is equivalent to continual learning with zero historical data. To address this problem, we propose Interpolated Weight Optimization, which utilizes the benefit of weight interpolation in both training and test time. We evaluate our method on three popular and challenging action recognition datasets following various zero-shot evaluation protocols and we demonstrate our approach outperforms state-of-the-art methods by clear margins. In particular, we achieve 87.9%, 58.3%, 81.1% zero-shot accuracy on UCF, HMDB and Kinetics-600 respectively, outperforming state-of-the-art methods by 8.3%, 7.8% and 12.2%. Code is released at https://github.com/wengzejia1/Open-VCLIP.

Text irregularities pose significant challenges to scene text recognizers. Thin-Plate Spline (TPS)-based rectification is widely regarded as an effective means to deal with them. Currently, the calculation of TPS transformation parameters purely depends on the quality of regressed text borders. It ignores the text content and often leads to unsatisfactory rectified results for severely distorted text. In this work, we introduce TPS++, an attention-enhanced TPS transformation that incorporates the attention mechanism to text rectification for the first time. TPS++ formulates the parameter calculation as a joint process of foreground control point regression and content-based attention score estimation, which is computed by a dedicated designed gated-attention block. TPS++ builds a more flexible content-aware rectifier, generating a natural text correction that is easier to read by the subsequent recognizer. Moreover, TPS++ shares the feature backbone with the recognizer in part and implements the rectification at feature-level rather than image-level, incurring only a small overhead in terms of parameters and inference time. Experiments on public benchmarks show that TPS++ consistently improves the recognition and achieves state-of-the-art accuracy. Meanwhile, it generalizes well on different backbones and recognizers. Code is at https://github.com/simplify23/TPS_PP.

Existing cross-domain transferable attacks mostly focus on exploring the adversarial transferability across homomodal domains, while the adversarial transferability across heteromodal domains, e.g., image domains to video domains, has received less attention. This paper investigates cross-modal transferable attacks from image domains to video domains with the generator-oriented approach, i.e., crafting adversarial perturbations for each frame of video clips with the perturbation generator trained in the ImageNet domain to attack target video models. To this end, we propose an effective Generative Cross-Modal Attacks (GCMA) framework to enhance adversarial transferability from image domains to video domains. To narrow the domain gap between image and video data, we first propose a random motion module that warps images with synthetic random optical flows. We then integrate the random motion module into the feature disruption loss to incorporate additional temporal cues in the training phase. Specifically, feature disruption loss minimizes the cosine similarity between intermediate features of warped benign and adversarial images. Furthermore, motivated by the positive correlation between transferability and temporal consistency of adversarial video clips, we also introduce a temporal consistency loss that maximizes the cosine similarity between intermediate features of warped adversarial images and adversarial counterparts of warped benign images. Finally, GCMA trains the perturbation generator by simultaneously optimizing feature disruption loss and temporal consistency loss. Extensive experiments demonstrate the effectiveness of our proposed method, achieving state-of-the-art performance on Kinetics-400 and UCF-101. Our code is available at https://github.com/kay-ck/GCMA.

Recently, one-stage visual grounders attract high attention due to their comparable accuracy but significantly higher efficiency than two-stage grounders. However, inter-object relation modeling has not been well studied for one-stage grounders. Inter-object relationship modeling, though important, is not necessarily performed among all objects, as only part of them are related to the text query and may confuse the model. We call these objects "suspected objects". However, exploring their relationships in the one-stage paradigm is non-trivial because: (1) no object proposals are available as the basis on which to select suspected objects and perform relationship modeling; (2) suspected objects are more confusing than others, as they may share similar semantics, be entangled with certain relationships, etc, and thereby more easily mislead the model's prediction. Toward this end, we propose a Suspected Object Transformation mechanism (SOT), which can be seamlessly integrated into existing CNN and Transformer-based one-stage visual grounders to encourage the target object selection among the suspected ones. Suspected objects are dynamically discovered from a learned activation map adapted to the model's current discrimination ability during training. Afterward, on top of suspected objects, a Keyword-Aware Discrimination module (KAD) and an Exploration by Random Connection strategy (ERC) are concurrently proposed to help the model rethink its initial prediction. On the one hand, KAD leverages keywords contributing high to suspected object discrimination. On the other hand, ERC allows the model to seek the correct object instead of being trapped in a situation that always exploits the current false prediction. Extensive experiments demonstrate the effectiveness of our proposed method.

Cross-modal text-to-video retrieval aims to find semantically related videos for a text query. Since video and text are distinct modalities, the major challenge comes from building the correspondence between two modalities, thus relevant samples could be matched. Inherently, the text contains multiple relatively complete semantic units and each one is composed of three primary components, i.e., subject, predicate and object (SVO triplet). Therefore, it requires similar modeling of video content -- objects and their relations, to correctly retrieve videos for texts. To model fine-grained visual relations, this paper proposes a Multi-Granularity Matching (MGM) framework that considers both fine-grained relation triplet matching and coarse-grained global semantic matching for text-to-video retrieval. Specifically, in the proposed framework, we represent videos as SVO triplet tracklets by extracting frame-level relation triplets followed by temporal relation association across frames. Moreover, we design a transformer-based Bi-directional Fusion Block (BFB) to express each SVO triplet with a highly unified representation. The constructed SVO triplet tracklets provide a reasonable way to model fine-grained video contents, fulfilling a better alignment between videos and texts. Extensive experiments conducted on three benchmark datasets, i.e., MSR-VTT, LSMDC and MSVD, demonstrate the effectiveness of our proposed method.

Current face forgery detection methods have made significant progress in achieving high intra-dataset accuracy by building a deterministic binary detector. However, deterministic networks cannot effectively capture noise and distribution shifts in the input, which makes them less robust and prone to poor generalization in real-world scenarios. To address this problem, in this paper, we propose an Uncertainty-Aware Learning (UAL) method for face forgery detection. Specifically, we extend the Transformer model in a probabilistic manner by modeling dependencies between patches as Gaussian random variables. Additionally, we introduce a Patch Selection Module that can efficiently and accurately identify discriminative regions with high-uncertainty information, which are further utilized for final classification. Furthermore, with the quantified uncertainty of the entire image, we design a novel Uncertainty-Aware One-Center Loss that enhances intra-class compactness for genuine faces only, thereby improving the inter-class separability in the embedding space. We conduct extensive experiments to demonstrate the effectiveness of the proposed method, and the results verify that, our Uncertainty-Aware Learning method enjoys better robustness and generalization ability comparing against other state-of-the-art methods.

Deep learning has achieved great success in video recognition, yet still struggles to recognize novel actions when faced with only a few examples. To tackle this challenge, few-shot action recognition methods have been proposed to transfer knowledge from a source dataset to a novel target dataset with only one or a few labeled videos. However, existing methods mainly focus on modeling the temporal relations between the query and support videos while ignoring the spatial relations. In this paper, we find that the spatial misalignment between objects also occurs in videos, notably more common than the temporal inconsistency. We are thus motivated to investigate the importance of spatial relations and propose a more accurate few-shot action recognition method that leverages both spatial and temporal information. Particularly, a novel Spatial Alignment Cross Transformer (SA-CT) which learns to re-adjust the spatial relations and incorporates the temporal information is contributed. Experiments reveal that, even without using any temporal information, the performance of SA-CT is comparable to temporal based methods on 3/4 benchmarks. To further incorporate the temporal information, we propose a simple yet effective Temporal Mixer module. The Temporal Mixer enhances the video representation and improves the performance of the full SA-CT model, achieving very competitive results. In this work, we also exploit large-scale pretrained models for few-shot action recognition, providing useful insights for this research direction.

Most existing methods for unsupervised domain adaptation (UDA) rely on a shared network to extract domain-invariant features. However, when facing multiple source domains, optimizing such a network involves updating the parameters of the entire network, making it both computationally expensive and challenging, particularly when coupled with min-max objectives. Inspired by recent advances in prompt learning that adapts high-capacity models for downstream tasks in a computationally economic way, we introduce Multi-Prompt Alignment (MPA), a simple yet efficient framework for multi-source UDA. Given a source and target domain pair, MPA first trains an individual prompt to minimize the domain gap through a contrastive loss. Then, MPA denoises the learned prompts through an auto-encoding process and aligns them by maximizing the agreement of all the reconstructed prompts. Moreover, we show that the resulting subspace acquired from the auto-encoding process can easily generalize to a streamlined set of target domains, making our method more efficient for practical usage. Extensive experiments show that MPA achieves state-of-the-art results on three popular datasets with an impressive average accuracy of 54.1% on DomainNet.

Long-tailed object detection (LTOD) aims to handle the extreme data imbalance in real-world datasets, where many tail classes have scarce instances. One popular strategy is to explore extra data with image-level labels, yet it produces limited results due to (1) semantic ambiguity---an image-level label only captures a salient part of the image, ignoring the remaining rich semantics within the image; and (2) location sensitivity---the label highly depends on the locations and crops of the original image, which may change after data transformations like random cropping. To remedy this, we propose RichSem, a simple but effective method, which is robust to learn rich semantics from coarse locations without the need of accurate bounding boxes. RichSem leverages rich semantics from images, which are then served as additional ``soft supervision'' for training detectors. Specifically, we add a semantic branch to our detector to learn these soft semantics and enhance feature representations for long-tailed object detection. The semantic branch is only used for training and is removed during inference. RichSem achieves consistent improvements on both overall and rare-category of LVIS under different backbones and detectors. Our method achieves state-of-the-art performance without requiring complex training and testing procedures. Moreover, we show the effectiveness of our method on other long-tailed datasets with additional experiments.

The transferability of adversarial examples across different convolutional neural networks (CNNs) makes it feasible to perform black-box attacks, resulting in security threats for CNNs. However, fewer endeavors have been made to investigate transferable attacks for vision transformers (ViTs), which achieve superior performance on various computer vision tasks. Unlike CNNs, ViTs establish relationships between patches extracted from inputs by the self-attention module. Thus, adversarial examples crafted on CNNs might hardly attack ViTs. To assess the security of ViTs comprehensively, we investigate the transferability across different ViTs in both untargetd and targeted scenarios. More specifically, we propose a Pay No Attention (PNA) attack, which ignores attention gradients during backpropagation to improve the linearity of backpropagation. Additionally, we introduce a PatchOut/CubeOut attack for image/video ViTs. They optimize perturbations within a randomly selected subset of patches/cubes during each iteration, preventing over-fitting to the white-box surrogate ViT model. Furthermore, we maximize the L2 norm of perturbations, ensuring that the generated adversarial examples deviate significantly from the benign ones. These strategies are designed to be harmoniously compatible. Combining them can enhance transferability by jointly considering patch-based inputs and the self-attention of ViTs. Moreover, the proposed combined attack seamlessly integrates with existing transferable attacks, providing an additional boost to transferability. We conduct experiments on ImageNet and Kinetics-400 for image and video ViTs, respectively. Experimental results demonstrate the effectiveness of the proposed method.

Recognizing the ingredients composition for given food images facilitates the estimation of nutrition facts, which is crucial to various health relevant applications. Nevertheless, ingredient recognition is a multi-label long-tailed classification problem, where each image may contain multiple labels and the class distributions are highly imbalanced. Most existing approaches leverage off-the-shelf Convolutional Neural Networks (CNN) for multi-label ingredient recognition, overlooking the long-tailed issue, which results in low accuracy for tail ingredient categories. To address this problem, this paper proposes a dynamic Mixup (D-Mixup) approach, aiming to dynamically augment minority ingredients, in order to boost the recognition performance for tail ingredient categories. Specifically, our D-Mixup approach dynamically selects two training images based on the predictions of the previous training epoch, and generates a new synthetic image to train the recognition network. In this way, the training samples of tailed classes can be dynamically enlarged and better discriminative representations can be learnt for rare classes. Extensive experiments on both VIREO Food-172 dataset and UEC Food-100 dataset demonstrate the effectiveness of the proposed D-Mixup method.

Given a text description, Temporal Language Grounding (TLG) aims to localize temporal boundaries of the segments that contain the specified semantics in an untrimmed video. TLG is inherently a challenging task, as it requires comprehensive understanding of both sentence semantics and video contents. Previous works either tackle this task in a fully-supervised setting that requires a large amount of temporal annotations or in a weakly-supervised setting that usually cannot achieve satisfactory performance. Since manual annotations are expensive, to cope with limited annotations, we tackle TLG in a semi-supervised way by incorporating self-supervised learning, and proposeSelf-SupervisedSemi-SupervisedTemporalLanguageGrounding (S$^{4}$TLG). S$^{4}$TLG consists of two parts: (1) A pseudo label generation module that adaptively produces instant pseudo labels for unlabeled samples based on predictions from a teacher model; (2) A self-supervised feature learning module with inter-modal and intra-modal contrastive losses to learn video feature representations under the constraints of video content consistency and video-text alignment. We conduct extensive experiments on the ActivityNet-CD-OOD and Charades-CD-OOD datasets. The results demonstrate that our proposed S$^{4}$TLG can achieve competitive performance compared to fully-supervised state-of-the-art methods while only requiring a small portion of temporal annotations.

Visual Question Answering aims to answer the free-form natural language question based on the visual clues in a given image. It is a difficult problem as it requires understanding the fine-grained structured information of both language and image for compositional reasoning. To establish the compositional reasoning, recent works attempt to introduce the scene graph in VQA. However, as the generated scene graphs are usually quite noisy, it greatly limits the performance of question answering. Therefore, this paper proposes to refine the scene graphs for improving the effectiveness. Specifically, we present a novelSceneGraphRefinement network (SGR), which introduces a transformer-based refinement network to enhance the object and relation features for better classification. Moreover, as the question provides valuable clues for distinguishing whether the$\left\langle \mathit{subject, predicate, object} \right\rangle$triplets are helpful or not, the SGR network exploits the semantic information presented in the questions to select the most relevant relations for question answering. Extensive experiments are conducted on the GQA benchmark demonstrate the effectiveness of our method.

Blind face inpainting refers to the task of reconstructing visual contents without explicitly indicating the corrupted regions in a face image. Inherently, this task faces two challenges: (1) how to detect various mask patterns of different shapes and contents; (2) how to restore visually plausible and pleasing contents in the masked regions. In this paper, we propose a novel two-stage blind face inpainting method named Frequency-guided Transformer and Top-Down Refinement Network (FT-TDR) to tackle these challenges. Specifically, we first use a transformer-based network to detect the corrupted regions to be inpainted as masks by modeling the relation among different patches. For improved detection results, we also exploit the frequency modality as complementary information and capture the local contextual incoherence to enhance boundary consistency. Then a top-down refinement network is proposed to hierarchically restore features at different levels and generate contents that are semantically consistent with the unmasked face regions. Extensive experiments demonstrate that our method outperforms current state-of-the-art blind and non-blind face inpainting methods qualitatively and quantitatively.

Unsupervised domain adaptation (UDA) aims to transfer knowledge learned from a fully-labeled source domain to a different unlabeled target domain. Most existing UDA methods learn domain-invariant feature representations by minimizing feature distances across domains. In this work, we build upon contrastive self-supervised learning to align features so as to reduce the domain discrepancy between training and testing sets. Exploring the same set of categories shared by both domains, we introduce a simple yet effective framework CDCL, for domain alignment. In particular, given an anchor image from one domain, we minimize its distances to cross-domain samples from the same class relative to those from different categories. Since target labels are unavailable, we use a clustering-based approach with carefully initialized centers to produce pseudo labels. In addition, we demonstrate that CDCL is a general framework and can be adapted to the data-free setting, where the source data are unavailable during training, with minimal modification. We conduct experiments on two widely used domain adaptation benchmarks, i.e., Office-31 and VisDA-2017, for image classification tasks, and demonstrate that CDCL achieves state-of-the-art performance on both datasets.

Recent research has demonstrated that Deep Neural Networks (DNNs) are vulnerable to adversarial patches which introduce perceptible but localized changes to the input. Nevertheless, existing approaches have focused on generating adversarial patches on images, their counterparts in videos have been less explored. Compared with images, attacking videos is much more challenging as it needs to consider not only spatial cues but also temporal cues. To close this gap, we introduce a novel adversarial attack in this paper, the bullet-screen comment (BSC) attack, which attacks video recognition models with BSCs. Specifically, adversarial BSCs are generated with a Reinforcement Learning (RL) framework, where the environment is set as the target model and the agent plays the role of selecting the position and transparency of each BSC. By continuously querying the target models and receiving feedback, the agent gradually adjusts its selection strategies in order to achieve a high fooling rate with non-overlapping BSCs. As BSCs can be regarded as a kind of meaningful patch, adding it to a clean video will not affect people’s understanding of the video content, nor will arouse people’s suspicion. We conduct extensive experiments to verify the effectiveness of the proposed method. On both UCF-101 and HMDB-51 datasets, our BSC attack method can achieve about 90% fooling rate when attacking three mainstream video recognition models, while only occluding < 8% areas in the video. Our code is available at https://github.com/kay-ck/BSC-attack.

Vision transformers (ViTs) have demonstrated impressive performance on a series of computer vision tasks, yet they still suffer from adversarial examples. In this paper, we posit that adversarial attacks on transformers should be specially tailored for their architecture, jointly considering both patches and self-attention, in order to achieve high transferability. More specifically, we introduce a dual attack framework, which contains a Pay No Attention (PNA) attack and a PatchOut attack, to improve the transferability of adversarial samples across different ViTs. We show that skipping the gradients of attention during backpropagation can generate adversarial examples with high transferability. In addition, adversarial perturbations generated by optimizing randomly sampled subsets of patches at each iteration achieve higher attack success rates than attacks using all patches. We evaluate the transferability of attacks on state-of-the-art ViTs, CNNs and robustly trained CNNs. The results of these experiments demonstrate that the proposed dual attack can greatly boost transferability between ViTs and from ViTs to CNNs. In addition, the proposed method can easily be combined with existing transfer methods to boost performance.

Although deep-learning based video recognition models have achieved remarkable success, they are vulnerable to adversarial examples that are generated by adding human-imperceptible perturbations on clean video samples. As indicated in recent studies, adversarial examples are transferable, which makes it feasible for black-box attacks in real-world applications. Nevertheless, most existing adversarial attack methods have poor transferability when attacking other video models and transfer-based attacks on video models are still unexplored. To this end, we propose to boost the transferability of video adversarial examples for black-box attacks on video recognition models. Through extensive analysis, we discover that different video recognition models rely on different discriminative temporal patterns, leading to the poor transferability of video adversarial examples. This motivates us to introduce a temporal translation attack method, which optimizes the adversarial perturbations over a set of temporal translated video clips. By generating adversarial examples over translated videos, the resulting adversarial examples are less sensitive to temporal patterns existed in the white-box model being attacked and thus can be better transferred. Extensive experiments on the Kinetics-400 dataset and the UCF-101 dataset demonstrate that our method can significantly boost the transferability of video adversarial examples. For transfer-based attack against video recognition models, it achieves a 61.56% average attack success rate on the Kinetics-400 and 48.60% on the UCF-101.

Built on top of self-attention mechanisms, vision transformers have demonstrated remarkable performance on a variety of tasks recently. While achieving excellent performance, they still require relatively intensive computational cost that scales up drastically as the numbers of patches, self-attention heads and transformer blocks increase. In this paper, we argue that due to the large variations among images, their need for modeling long-range dependencies between patches differ. To this end, we introduce AdaViT, an adaptive computation framework that learns to derive usage policies on which patches, self-attention heads and transformer blocks to use throughout the backbone on a per-input basis, aiming to improve inference efficiency of vision transformers with a minimal drop of accuracy for image recognition. Optimized jointly with a transformer backbone in an end-to-end manner, a light-weight decision network is attached to the backbone to produce decisions on-the-fly. Extensive experiments on ImageNet demonstrate that our method obtains more than 2 × improvement on efficiency compared to state-of-the-art vision transformers with only 0.8% drop of accuracy, achieving good efficiency/accuracy trade-offs conditioned on different computational budgets. We further conduct quantitative and qualitative analysis on learned usage polices and provide more insights on the redundancy in vision transformers. Code is available at ht tps: / / gi thub. com/MengLcool/AdaVi T.

This paper studies the BERT pretraining of video transformers. It is a straightforward but worth-studying extension given the recent success from BERT pretraining of image transformers. We introduce BEVT which decouples video representation learning into spatial representation learning and temporal dynamics learning. In particular, BEVT first performs masked image modeling on image data, and then conducts masked image modeling jointly with masked video modeling on video data. This design is motivated by two observations: 1) transformers learned on image datasets provide decent spatial priors that can ease the learning of video transformers, which are often times computationally-intensive if trained from scratch; 2) discriminative clues, i.e., spatial and temporal information, needed to make correct predictions vary among different videos due to large intra-class and inter-class variations. We conduct extensive experiments on three challenging video benchmarks where BEVT achieves very promising results. On Kinetics 400, for which recognition mostly relies on discriminative spatial representations, BEVT achieves comparable results to strong supervised baselines. On Something-Something-V2 and Diving 48, which contain videos relying on temporal dynamics, BEVT outperforms by clear margins all alternative baselines and achieves state-of-the-art performance with a 71.4% and 87.2% Top-1 accuracy respectively. Code is available at https://github.com/xyzforever/BEVT.

Recent advances in image editing techniques have posed serious challenges to the trustworthiness of multimedia data, which drives the research of image tampering detection. In this paper, we propose ObjectFormer to detect and localize image manipulations. To capture subtle manipulation traces that are no longer visible in the RGB domain, we extract high-frequency features of the images and combine them with RGB features as multimodal patch embeddings. Additionally, we use a set of learnable object prototypes as mid-level representations to model the object-level consistencies among different regions, which are further used to refine patch embeddings to capture the patch-level consistencies. We conduct extensive experiments on various datasets and the results verify the effectiveness of the proposed method, outperforming state-of-the-art tampering detection and localization methods.

Recent studies have shown that adversarial examples handcrafted on one white-box model can be used to at-tack other black-box models. Such cross-model transferability makes it feasible to perform black-box attacks, which has raised security concerns for real-world DNNs applications. Nevertheless, existing works mostly focus on investigating the adversarial transferability across different deep models that share the same modality of input data. The cross-modal transferability of adversarial perturbation has never been explored. This paper investigates the transferability of adversarial perturbation across different modalities, i.e., leveraging adversarial perturbation generated on white-box image models to attack black-box video models. Specifically, motivated by the observation that the low-level feature space between images and video frames are similar, we propose a simple yet effective cross-modal attack method, named as Image To Video (I2V) attack. I2V generates adversarial frames by minimizing the cosine similarity between features of pretrained image models from adversarial and benign examples, then combines the generated adversarial frames to perform black-box attacks on video recognition models. Extensive experiments demonstrate that I2V can achieve high attack success rates on different black-box video recognition models. On Kinetics-400 and UCF-101, I2V achieves an average attack success rate of 77.88% and 65.68%, respectively, which sheds light on the feasibility of cross-modal adversarial attacks.

Real-world data typically follow a long-tailed distribution, where a few majority categories occupy most of the data while most minority categories contain a limited number of samples. Classification models minimizing crossentropy struggle to represent and classify the tail classes. Although the problem of learning unbiased classifiers has been well studied, methods for representing imbalanced data are under-explored. In this paper, we focus on representation learning for imbalanced data. Recently, supervised contrastive learning has shown promising performance on balanced data recently. However, through our theoretical analysis, we find that for long-tailed data, it fails to form a regular simplex which is an ideal geometric configuration for representation learning. To correct the optimization behavior of SCL and further improve the performance of long-tailed visual recognition, we propose a novel loss for balanced contrastive learning (BCL). Compared with SCL, we have two improvements in BCL: classaveraging, which balances the gradient contribution of negative classes; class-complement, which allows all classes to appear in every mini-batch. The proposed balanced contrastive learning (BCL) method satisfies the condition of forming a regular simplex and assists the optimization of cross-entropy. Equipped with BCL, the proposed two-branch framework can obtain a stronger feature representation and achieve competitive performance on long-tailed benchmark datasets such as CIFAR-10-LT, CIFAR-100-LT, ImageNet-LT, and iNaturalist2018.

Real-world data is often unbalanced and exhibits long-tailed distribution over classes. Vanilla classification models trained on imbalanced datasets inherently exhibit bias towards dominant classes. Existing debiasing methods mostly balance the data or the loss during training. Nevertheless, these data-acquiring methods are not suitable for situations where training data are unavailable. In this paper, we appeal to solutions without access to training data and propose a datafree debiasing (Free-D) method that serves as a plug-and-play module for any standard classification model. Specifically, our method adjusts both the feature representation via feature representation shifting and the classifier weight via class prior compensation in a data-free manner. We evaluate and compare our methods on four long-tailed visual recognition datasets, i.e., long-tailed CIFAR-10/-100, ImageNet-LT, and Places-LT. Extensive experiments demonstrate that the proposed data-free method achieves comparable results of other data-acquired methods.

Dominant scene text recognition models commonly contain two building blocks, a visual model for feature extraction and a sequence model for text transcription. This hybrid architecture, although accurate, is complex and less efficient. In this study, we propose a Single Visual model for Scene Text recognition within the patch-wise image tokenization framework, which dispenses with the sequential modeling entirely. The method, termed SVTR, firstly decomposes an image text into small patches named character components. Afterward, hierarchical stages are recurrently carried out by component-level mixing, merging and/or combining. Global and local mixing blocks are devised to perceive the inter-character and intra-character patterns, leading to a multi-grained character component perception. Thus, characters are recognized by a simple linear prediction. Experimental results on both English and Chinese scene text recognition tasks demonstrate the effectiveness of SVTR. SVTR-L (Large) achieves highly competitive accuracy in English and outperforms existing methods by a large margin in Chinese, while running faster. In addition, SVTR-T (Tiny) is an effective and much smaller model, which shows appealing speed at inference. The code is publicly available at https://github.com/PaddlePaddle/PaddleOCR.

The widespread dissemination of Deepfakes demands effective approaches that can detect perceptually convincing forged images. In this paper, we aim to capture the subtle manipulation artifacts at different scales using transformer models. In particular, we introduce a Multi-modal Multi-scale TRansformer (M2TR), which operates on patches of different sizes to detect local inconsistencies in images at different spatial levels. M2TR further learns to detect forgery artifacts in the frequency domain to complement RGB information through a carefully designed cross modality fusion block. In addition, to stimulate Deepfake detection research, we introduce a high-quality Deepfake dataset, SR-DF, which consists of 4,000 DeepFake videos generated by state-of-the-art face swapping and facial reenactment methods. We conduct extensive experiments to verify the effectiveness of the proposed method, which outperforms state-of-the-art Deepfake detection methods by clear margins.

Deep-learning based video models, which have remarkable performance on action recognition tasks, are recently proved to be vulnerable to adversarial samples, even those generated in the black-box setting. However, these black-box attack methods are insufficient to attack videos models in real-world applications due to the requirement of lots of queries. To this end, we propose to boost the efficiency of black-box attacks on video recognition models. Although videos carry rich temporal information, they include redundant spatial information from adjacent frames. This motivates us to introduce the adaptive temporal grouping (ATG) method, which groups video frames by the similarity of their features extracted from the ImageNet-pretrained image model. By selecting one key-frame from each group, ATG helps any black-box attack methods to optimize the adversarial perturbations over key-frames instead of all frames, where the estimated gradient of key-frame is shared with other frames in each group. To balance the efficiency and precision of estimated gradients, ATG adaptively adjusts the group number by the magnitude of the current perturbation and the current query number. Through extensive experiments on the HMDB-51 dataset and the UCF-101 dataset, we demonstrate that ATG can significantly reduce the number of queries by more than 10% for the targeted attack.

Cooking video captioning aims to generate the text instructions that describes the cooking procedures presented in the video. Current approaches tend to use large neural models or use more robust feature extractors to increase the expressive ability of features, ignoring the strong correlation between consecutive cooking steps in the video. However, it is intuitive that previous cooking steps can provide clues for the next cooking step. Specially, consecutive cooking steps tend to share the same ingredients. Therefore, accurate ingredients recognition can help to introduce more fine-grained information in captioning. To improve the performance of video procedural caption in cooking video, this paper proposes a framework that introduces ingredient recognition module which uses the copy mechanism to fuse the predicted ingredient information into the generated sentence. Moreover, we integrate the visual information of the previous step into the generation of the current step, and the visual information of the two steps together assist in the generation process. Extensive experiments verify the effectiveness of our propose framework and it achieves the promising performances on both YouCookII and Cooking-COIN datasets.

Recently, Cross-Domain Few-Shot Learning (CD-FSL) which aims at addressing the Few-Shot Learning (FSL) problem across different domains has attracted rising attention. The core challenge of CD-FSL lies in the domain gap between the source and novel target datasets. Though many attempts have been made for CD-FSL without any target data during model training, the huge domain gap makes it still hard for existing CD-FSL methods to achieve very satisfactory results. Alternatively, learning CD-FSL models with few labeled target domain data which is more realistic and promising is advocated in previous work. Thus, in this paper, we stick to this setting and technically contribute a novel Multi-Expert Domain Decompositional Network (ME-D2N). Concretely, to solve the data imbalance problem between the source data with sufficient examples and the auxiliary target data with limited examples, we build our model under the umbrella of multi-expert learning. Two teacher models which can be considered to be experts in their corresponding domain are first trained on the source and the auxiliary target sets, respectively. Then, the knowledge distillation technique is introduced to transfer the knowledge from two teachers to a unified student model. Taking a step further, to help our student model learn knowledge from different domain teachers simultaneously, we further present a novel domain decomposition module that learns to decompose the student model into two domain-related sub-parts. This is achieved by a novel domain-specific gate that learns to assign each filter to only one specific domain in a learnable way. Extensive experiments demonstrate the effectiveness of our method. Codes and models are available at https://github.com/lovelyqian/ME-D2N\_for\_CDFSL.

Video domain adaptation is non-trivial due to video is inherently involved with multi-dimensional and multi-modal information. Existing works mainly adopt adversarial learning and self-supervised tasks to align features. Nevertheless, the explicit interaction between source and target in the temporal dimension, as well as the adaptation between modalities, are unexploited. In this paper, we propose Mix-Domain-Adversarial Neural Network and Dynamic-Modal-Distillation (MD-DMD), a novel multi-modal adversarial learning framework for unsupervised video domain adaptation. Our approach incorporates the temporal information between source and target domains, as well as the diversity of adaptability between modalities. On the one hand, for every single modality, we mix the frames from source and target domains to form mix-samples, then let the adversarial-discriminator predict the mix ratio of a mix-sample to further enhance the ability of the model to capture domain-invariant feature representations. On the other hand, we dynamically estimate the adaptability for different modalities during training, then pick the most adaptable modality as a teacher to guide other modalities by knowledge distillation. As a result, modalities are capable of learning transferable knowledge from each other, which leads to more effective adaptation. Experiments on two video domain adaptation benchmarks demonstrate the superiority of our proposed MD-DMD over state-of-the-art methods.

Given sufficient training data on the source domain, cross-domain few-shot learning (CD-FSL) aims at recognizing new classes with a small number of labeled examples on the target domain. The key to addressing CD-FSL is to narrow the domain gap and transferring knowledge of a network trained on the source domain to the target domain. To help knowledge transfer, this paper introduces an intermediate domain generated by mixing images in the source and the target domain. Specifically, to generate the optimal intermediate domain for different target data, we propose a novel target guided dynamic mixup (TGDM) framework that leverages the target data to guide the generation of mixed images via dynamic mixup. The proposed TGDM framework contains a Mixup-3T network for learning classifiers and a dynamic ratio generation network (DRGN) for learning the optimal mix ratio. To better transfer the knowledge, the proposed Mixup-3T network contains three branches with shared parameters for classifying classes in the source domain, target domain, and intermediate domain. To generate the optimal intermediate domain, the DRGN learns to generate an optimal mix ratio according to the performance on auxiliary target data. Then, the whole TGDM framework is trained via bi-level meta-learning so that TGDM can rectify itself to achieve optimal performance on target data. Extensive experimental results on several benchmark datasets verify the effectiveness of our method.

This paper presents OmniVL, a new foundation model to support both image-language and video-language tasks using one universal architecture. It adopts a unified transformer-based visual encoder for both image and video inputs, and thus can perform joint image-language and video-language pretraining. We demonstrate, for the first time, such a paradigm benefits both image and video tasks, as opposed to the conventional one-directional transfer (e.g., use image-language to help video-language). To this end, we propose a \emph{decoupled} joint pretraining of image-language and video-language to effectively decompose the vision-language modeling into spatial and temporal dimensions and obtain performance boost on both image and video tasks. Moreover, we introduce a novel unified vision-language contrastive (UniVLC) loss to leverage image-text, video-text, image-label (e.g., image classification), video-label (e.g., video action recognition) data together, so that both supervised and noisily supervised pretraining data are utilized as much as possible. Without incurring extra task-specific adaptors, OmniVL can simultaneously support visual only tasks (e.g., image classification, video action recognition), cross-modal alignment tasks (e.g., image/video-text retrieval), and multi-modal understanding and generation tasks (e.g., image/video question answering, captioning). We evaluate OmniVL on a wide range of downstream tasks and achieve state-of-the-art or competitive results with similar model size and data scale.

Video moment retrieval aims at finding the start and end timestamps of a moment (part of a video) described by a given natural language query. Fully supervised methods need complete temporal boundary annotations to achieve promising results, which is costly since the annotator needs to watch the whole moment. Weakly supervised methods only rely on the paired video and query, but the performance is relatively poor. In this paper, we look closer into the annotation process and propose a new paradigm called "glance annotation". This paradigm requires the timestamp of only one single random frame, which we refer to as a "glance", within the temporal boundary of the fully supervised counterpart. We argue this is beneficial because comparing to weak supervision, trivial cost is added yet more potential in performance is provided. Under the glance annotation setting, we propose a method named as Video moment retrieval via Glance Annotation (ViGA) based on contrastive learning. ViGA cuts the input video into clips and contrasts between clips and queries, in which glance guided Gaussian distributed weights are assigned to all clips. Our extensive experiments indicate that ViGA achieves better results than the state-of-the-art weakly supervised methods by a large margin, even comparable to fully supervised methods in some cases.

We introduce AdaFrame, a conditional computation framework that adaptively selects relevant frames on a per-input basis for fast video recognition. AdaFrame, which contains a Long Short-Term Memory augmented with a global memory to provide context information, operates as an agent to interact with video sequences aiming to search over time which frames to use. Trained with policy search methods, at each time step, AdaFrame computes a prediction, decides where to observe next, and estimates a utility, i.e., expected future rewards, of viewing more frames in the future. Exploring predicted utilities at testing time, AdaFrame is able to achieve adaptive lookahead inference so as to minimize the overall computational cost without incurring a degradation in accuracy. We conduct extensive experiments on two large-scale video benchmarks, FCVID and ActivityNet. With a vanilla ResNet-101 model, AdaFrame achieves similar performance of using all frames while only requiring, on average, 8.21 and 8.65 frames on FCVID and ActivityNet, respectively. We also demonstrate AdaFrame is compatible with modern 2D and 3D networks for video recognition. Furthermore, we show, among other things, learned frame usage can reflect the difficulty of making prediction decisions both at instance-level within the same class and at class-level among different categories.

The vanilla Few-shot Learning (FSL) learns to build a classifier for a new concept from one or very few target examples, with the general assumption that source and target classes are sampled from the same domain. Recently, the task of Cross-Domain Few-Shot Learning (CD-FSL) aims at tackling the FSL where there is a huge domain shift between the source and target datasets. Extensive efforts on CD-FSL have been made via either directly extending the meta-learning paradigm of vanilla FSL methods, or employing massive unlabeled target data to help learn models. In this paper, we notice that in the CD-FSL task, the few labeled target images have never been explicitly leveraged to inform the model in the training stage. However, such a labeled target example set is very important to bridge the huge domain gap. Critically, this paper advocates a more practical training scenario for CD-FSL. And our key insight is to utilize a few labeled target data to guide the learning of the CD-FSL model. Technically, we propose a novel Generalized Meta-learning based Feature-Disentangled Mixup network, namely GMeta-FDMixup. We make three key contributions of utilizing GMeta-FDMixup to address CD-FSL. Firstly, we present two mixup modules - mixup-P and mixup-M that help facilitate utilizing the unbalanced and disjoint source and target datasets. These two novel modules enable diverse image generation for training the model on the source domain. Secondly, to narrow the domain gap explicitly, we contribute a novel feature disentanglement module that learns to decouple the domain-irrelevant and domain-specific features. By stripping the domain-specific features, we alleviate the negative effects caused by the domain inductive bias. Finally, we repurpose a new contrastive learning module, dubbed ConL. ConL prevents the model from only capturing category-related features via introducing contrastive loss. Thus, the generalization ability on novel categories is improved. Extensive experimental results on two benchmarks show the superiority of our setting and the effectiveness of our method. Code and models will be released.

Cross-modal text to video retrieval aims to find relevant videos for given text queries, which is crucial for various real-world applications. The key to address this task is to build the correspondence between video and text such that the related samples from different modalities can be aligned. As the text (sentence) contains both nouns and verbs representing objects as well as their interactions, retrieving the relevant videos requires a fine-grained understanding of video contents---not only the semantic concepts (i.e., objects) but also the interactions between them. Nevertheless, the current approaches mostly represent videos with aggregated frame-level features for the learning of joint space and ignore the information of object interactions, which usually results in suboptimal retrieval performance. To improve the performance of cross-modal video retrieval, this paper proposes a framework that models videos as spatial-temporal graphs where nodes correspond to visual objects and edges correspond to the relations/interactions between objects. With the spatial-temporal graphs, the object interactions in frame sequences can be captured to enrich the video representations for joint space learning. Specifically, Graph Convolutional Network is introduced to learn the representations on spatial-temporal graphs, aiming to encode the spatial-temporal interactions between objects; while BERT is introduced to dynamically encode the sentence according to the context for cross-modal retrieval. Extensive experiments verify the effectiveness of the proposed framework and it achieves promising performances on both MSR-VTT and LSMDC datasets.

Video recognition aims at understanding semantic contents that normally involve the interactions of humans and related objects under certain scenes. A common practice to improve recognition accuracy is to combine object, scene and action features for classification directly, assuming that they are explicitly complementary. In this paper, we break down the fusion of three features into two pairwise feature relation modeling processes, which mitigates the difficulty of correlation learning in high dimensional features. Towards this goal, we introduce a Semantics Attention Module that captures the relations of a pair of features by refining the relatively “weak” feature with the guidance from the “strong” feature using attention mechanisms. The refined representation is further combined with the “strong” feature using a residual design for downstream tasks. Two SAMs are applied in a Semantics Attention Network (SAN) for improving video recognition. Extensive experiments are conducted on two large-scale video benchmarks, FCVID and ActivityNet v1.3—the proposed approach achieves better results while requiring much less computational effort than alternative methods.

Dense Event Captioning (DEC) aims to jointly localize and describe multiple events of interest in untrimmed videos, which is an advancement of the conventional video captioning task (generating a single sentence description for a trimmed video). Weakly Supervised Dense Event Captioning (WS-DEC) goes one step further by not relying on human-annotated temporal event boundaries. However, there are few methods trying to tackle this task, and how to connect localization and description remains an open problem. In this paper, we demonstrate that under weak supervision, the event captioning module and localization module should be more closely bridged in order to improve description performance. Different from previous approaches, in our method, the event captioner generates a sentence from a video segment and feeds it to the sentence localizer to reconstruct the segment, and the localizer produces word importance weights as a guidance for the captioner to improve event description. To further bridge the sentence localizer and event captioner, a concept learner is adopted as the basis of the sentence localizer, which can be utilized to construct an induced set of concept features to enhance video features and improve the event captioner. Finally, our proposed method outperforms state-of-the-art WS-DEC methods on the ActivityNet Captions dataset.

Video captioning is an important vision task and has been intensively studied in the computer vision community. Existing methods that utilize the fine-grained spatial information have achieved significant improvements, however, they either rely on costly external object detectors or do not sufficiently model the spatial/temporal relations. In this paper, we aim at designing a spatial information extraction and aggregation method for video captioning without the need of external object detectors. For this purpose, we propose a Recurrent Region Attention module to better extract diverse spatial features, and by employing Motion-Guided Cross-frame Message Passing, our model is aware of the temporal structure and able to establish high-order relations among the diverse regions across frames. They jointly encourage information communication and produce compact and powerful video representations. Furthermore, an Adjusted Temporal Graph Decoder is proposed to flexibly update video features and model high-order temporal relations during decoding. Experimental results on three benchmark datasets: MSVD, MSR-VTT, and VATEX demonstrate that our proposed method can outperform state-of-the-art methods.

Label distributions in real-world are oftentimes long-tailed and imbalanced, resulting in biased models towards dominant labels. While long-tailed recognition has been extensively studied for image classification tasks, limited effort has been made for the video domain. In this paper, we introduce VideoLT, a large-scale long-tailed video recognition dataset, as a step toward real-world video recognition. VideoLT contains 256,218 untrimmed videos, annotated into 1,004 classes with a long-tailed distribution. Through extensive studies, we demonstrate that state-of-the-art methods used for long-tailed image recognition do not perform well in the video domain due to the additional temporal dimension in videos. This motivates us to propose FrameStack, a simple yet effective method for long-tailed video recognition. In particular, FrameStack performs sampling at the frame-level in order to balance class distributions, and the sampling ratio is dynamically determined using knowledge derived from the network during training. Experimental results demonstrate that FrameStack can improve classification performance without sacrificing the overall accuracy. Code and dataset are available at: https://github.com/17Skye17/VideoLT.

Adversarial training is one effective approach for training robust deep neural networks against adversarial attacks. While being able to bring reliable robustness, adversarial training (AT) methods in general favor high capacity models, i.e., the larger the model the better the robustness. This tends to limit their effectiveness on small models, which are more preferable in scenarios where storage or computing resources are very limited (e.g., mobile devices). In this paper, we leverage the concept of knowledge distillation to improve the robustness of small models by distilling from adversarially trained large models. We first revisit several state-of-the-art AT methods from a distillation perspective and identify one common technique that can lead to improved robustness: the use of robust soft labels – predictions of a robust model. Following this observation, we propose a novel adversarial robustness distillation method called Robust Soft Label Adversarial Distillation (RSLAD) to train robust small student models. RSLAD fully exploits the robust soft labels produced by a robust (adversarially-trained) large teacher model to guide the student’s learning on both natural and adversarial examples in all loss terms. We empirically demonstrate the effectiveness of our RSLAD approach over existing adversarial training and distillation methods in improving the robustness of small models against state-of-the-art attacks including the AutoAttack. We also provide a set of understandings on our RSLAD and the importance of robust soft labels for adversarial robustness distillation. Code: https://github.com/zibojia/RSLAD.

Object detection is an essential computer vision task that possesses extensive application prospects in on-road applications. Copious novel methods have been proposed in this branch recently. However, the majority of them have high computational cost, making them intractable to be deployed on embedded devices. In this paper, taking YOLOv5s, the smallest model in the YOLOv5 family, as the baseline, we explore a bag of tricks that improve the detection performance for a specified on-road application, under the premise of ensuring that it does not increase the computational cost of YOLOv5s. Specifically, we introduce relevantly external data to deal with the problems of sample imbalance. Meanwhile, knowledge distillation is employed to transfer knowledge from a cumbersome model to a compact model, where a united distillation scheme is developed to enhance the effectiveness. In addition, a pseudo-label based training strategy is utilized to further learn from the biggest YOLOv5 model. We have applied the above tricks to the Embedded Deep Learning Object Detection Model Compression Competition for Traffic in Asian Countries held in conjunction with ICMR 2021. The experiments have shown that all the tricks are useful. Their combination have built an accurate and slim detection model. It is highly competitive and has been ranked 2nd place in the competition. We believe the tricks are also meaningful for building other application-oriented object detectors.

Given a video demonstration, can we imitate the action contained in this video? In this paper, we introduce a novel task, dubbed mesh-based action imitation. The goal of this task is to enable an arbitrary target human mesh to perform the same action shown on the video demonstration. To achieve this, a novel Mesh-based Video Action Imitation (M-VAI) method is proposed by us. M-VAI first learns to reconstruct the meshes from the given source image frames, then the initial recovered mesh sequence is fed into mesh2mesh, a mesh sequence smooth module proposed by us, to improve the temporal consistency. Finally, we imitate the actions by transferring the pose from the constructed human body to our target identity mesh. High-quality and detailed human body meshes can be generated by using our M-VAI. Extensive experiments demonstrate the feasibility of our task and the effectiveness of our proposed method.

A recent study [4] finds that existing few-shot learning methods, trained on the source domain, fail to generalize to the novel target domain when a domain gap is observed. This motivates the task of Cross-Domain Few-Shot Learning (CD-FSL). In this paper, we realize that the labeled target data in CD-FSL has not been leveraged in any way to help the learning process. Thus, we advocate utilizing few labeled target data to guide the model learning. Technically, a novel meta-FDMixup network is proposed. We tackle this problem mainly from two aspects. Firstly, to utilize the source and the newly introduced target data of two different class sets, a mixup module is re-proposed and integrated into the meta-learning mechanism. Secondly, a novel disentangle module together with a domain classifier is proposed to extract the disentangled domain-irrelevant and domain-specific features. These two modules together enable our model to narrow the domain gap thus generalizing well to the target datasets. Additionally, a detailed feasibility and pilot study is conducted to reflect the intuitive understanding of CD-FSL under our new setting. Experimental results show the effectiveness of our new setting and the proposed method. Codes and models are available at https://github.com/lovelyqian/Meta-FDMixup.

Referring Image Segmentation (RIS) aims at segmenting the target object from an image referred by one given natural language expression. The diverse and flexible expressions and complex visual contents in the images raise the RIS model with higher demands for investigating fine-grained matching behaviors between words in expressions and objects presented in images. However, such matching behaviors are hard to be learned and captured when the visual cues of referents (i.e. referred objects) are insufficient, as the referents of weak visual cues tend to be easily confused by cluttered background at boundary or even overwhelmed by salient objects in the image. And the insufficient visual cues issue can not be handled by the cross-modal fusion mechanisms as done in previous work.In this paper, we tackle this problem from a novel perspective of enhancing the visual information for the referents by devising a Two-stage Visual cues enhancement Network (TV-Net), where a novel Retrieval and Enrichment Scheme (RES) and an Adaptive Multi-resolution feature Fusion (AMF) module are proposed. Specifically, RES retrieves the most relevant image from an external data pool with regard to both the visual and textual similarities, and then enriches the visual information of the referent with the retrieved image for better multimodal feature learning. AMF further enhances the visual detailed information by incorporating the high-resolution feature maps from lower convolution layers of the image. Through the two-stage enhancement, our proposed TV-Net enjoys better performances in learning fine-grained matching behaviors between the natural language expression and image, especially when the visual information of the referent is inadequate, thus produces better segmentation results. Extensive experiments are conducted to validate the effectiveness of the proposed method on the RIS task, with our proposed TV-Net surpassing the state-of-the-art approaches on four benchmark datasets.

Video moment retrieval aims to localize the most relevant video moment given the text query. Weakly supervised approaches leverage video-text pairs only for training, without temporal annotations. Most current methods align the proposed video moment and the text in a joint embedding space. However, in lack of temporal annotations, the semantic gap between these two modalities makes it predominant to learn joint feature representation for most methods, with less emphasis on learning visual feature representation. This paper aims to improve the visual feature representation with supervisions in the visual domain, obtaining discriminative visual features for cross-modal learning. Based on the observation that relevant video moments (i.e., share similar activities) from different videos are commonly described by similar sentences; hence the visual features of these relevant video moments should also be similar despite that they come from different videos. Therefore, to obtain more discriminative and robust visual features for video moment retrieval, we propose to align the visual features of relevant video moments from different videos that co-occurred in the same training batch. Besides, a contrastive learning approach is introduced for learning the moment-level alignment of these videos. Through extensive experiments, we demonstrate that the proposed visual co-occurrence alignment learning method outperforms the cross-modal alignment learning counterpart and achieves promising results for video moment retrieval.

There is a growing trend in placing video advertisements on social platforms for online marketing, which demands automatic approaches to understand the contents of advertisements effectively. Taking the 2021 TAAC competition as an opportunity, we developed a multimodal system to improve the ability of structured analysis of advertising video content. In our framework, we break down the video structuring analysis problem into two tasks, i.e., scene segmentation and multi-modal tagging. In scene segmentation, we build upon a temporal convolution module for temporal modeling to predict whether adjacent frames belong to the same scene. In multi-modal tagging, we first compute clip-level visual features by aggregating frame-level features with NeXt-SoftDBoF. The visual features are further complemented with textual features that are derived using a global-local attention mechanism to extract useful information from OCR (Optical Character Recognition) and ASR (Audio Speech Recognition) outputs. Our solution achieved a score of 0.2470 measured in consideration of localization and prediction accuracy, ranking fourth in the 2021 TAAC final leaderboard.

In this paper, we propose an end-to-end deep learning architecture that generates 3D triangular meshes from single color images. Restricted by the nature of prevalent deep learning techniques, the majority of previous works represent 3D shapes in volumes or point clouds. However, it is non-trivial to convert these representations to compact and ready-to-use mesh models. Unlike the existing methods, our network represents 3D shapes in meshes, which are essentially graphs and well suited for graph-based convolutional neural networks. Leveraging perceptual features extracted from an input image, our network produces the correct geometry by progressively deforming an ellipsoid. To make the whole deformation procedure stable, we adopt a coarse-to-fine strategy, and define various mesh/surface related losses to capture properties of various aspects, which benefits producing the visually appealing and physically accurate 3D geometry. In addition, our model by nature can be adapted to objects in specific domains, e.g., human faces, and be easily extended to learn per-vertex properties, e.g., color. Extensive experiments show that our method not only qualitatively produces the mesh model with better details, but also achieves the higher 3D shape estimation accuracy compared against the state-of-the-arts.

Rapid development of mobile devices has led to explosive growth of videos and online platforms, which creates great demand for online advertising in videos. Existing advertising methods often aim to randomly select a time point as insertion position, which means that the video content is likely not related to the ad content, resulting in unsatisfactory user experience. While previous works have neglected to understand rich semantics as well as multimodal information in video advertising, in contrast to previous works, we present an innovative method for video-in-video advertising using multimodal modeling. First, different pre-trained models are used to extract multimodal representations. Then, through multimodal modeling, we learn the complementarity among different representations and obtain a unified video-level description. Finally, the unified representations of ads and videos are utilized to find the best matching result for each advertisement. Our method emphasizes the content similarity between ad and video, which would make the transition between video and ad more natural. Comprehensive experiments with both objective and subjective evaluations demonstrate the effectiveness and user-friendliness of our proposed video-in-video advertising framework.

Food recognition has captured numerous research attention for its importance for health-related applications. The existing approaches mostly focus on the categorization of food according to dish names, while ignoring the underlying ingredient composition. In reality, two dishes with the same name do not necessarily share the exact list of ingredients. Therefore, the dishes under the same food category are not mandatorily equal in nutrition content. Nevertheless, due to limited datasets available with ingredient labels, the problem of ingredient recognition is often overlooked. Furthermore, as the number of ingredients is expected to be much less than the number of food categories, ingredient recognition is more tractable in the real-world scenario. This paper provides an insightful analysis of three compelling issues in ingredient recognition. These issues involve recognition in either image-level or region level, pooling in either single or multiple image scales, learning in either single or multi-task manner. The analysis is conducted on a large food dataset, Vireo Food-251, contributed by this paper. The dataset is composed of 169,673 images with 251 popular Chinese food and 406 ingredients. The dataset includes adequate challenges in scale and complexity to reveal the limit of the current approaches in ingredient recognition.

Hashtags are keywords describing a topic or a theme and are usually chosen by microblogging users. Hence, the hashtags can be used to categorize microblog posts. With the fast development of the social network, the task of recommending suitable hashtags has received considerable attention in recent years. Recently, most neural network methods have treated the task as a multi-class classification problem. In fact, users are constantly introducing new hashtags in a highly dynamic way. Treating the task as a multi-class classification problem with a fixed number of target categories does not allow the method to deal with the new hashtags. To address this problem, the task is reinterpreted as a matching problem and a novel co-attention memory network is proposed to represent the multimodal microblogs and hashtags. We utilize a co-attention mechanism to model the multimodal mircroblogs, and utilize the post history to represent the hashtags. Experimental results on a Twitter-based dataset demonstrated that the proposed method can achieve better performance than the current state-of-the-art methods that treat the task as a multi-class classification problem.

This paper proposes a novel multimedia task: story-driven video editing. Given a story paragraph, this task aims to retrieve related video segments from a gallery of collected video segments and compose them into a video sequence by the storyline order. Our proposed baseline solution consists of three modules: a retrieval module, which returns lists of candidate segments for all query sentences in the story paragraph using an object-aware sentence-segment matching method; a sequence candidate proposal module, which aggregates the retrieved segment sets into a sequence proposal by the submodular optimization method; a sorting module, which arranges the candidates according to the storyline of the paragraph using the Sinkhorn network. We build a benchmark for this task including a reorganized version of the ActivityNet Captions dataset, a well-defined quantitative metric called Evaluation of Segment-to-Sequence Matching (ESSM) for measuring the difference between the generated video segment sequence and the ground truth. Quantitative results of the proposed baseline solution are reported. We hope this new task and benchmark will bring broad research attention and push forward a lot of novel online short video editing applications.

This paper studies the task of image captioning with novel objects, which only exist in testing images. Intrinsically, this task can reflect the generalization ability of models in understanding and captioning the semantic meanings of visual concepts and objects unseen in training set, sharing the similarity to one/zero-shot learning. The critical difficulty thus comes from that no paired images and sentences of the novel objects can be used to help train the captioning model. Inspired by recent work (Chen et al. 2019b) that boosts one-shot learning by learning to generate various image deformations, we propose learning meta-networks for deforming features for novel object captioning. To this end, we introduce the feature deformation meta-networks (FDM-net), which is trained on source data, and learn to adapt to the novel object features detected by the auxiliary detection model. FDM-net includes two sub-nets: feature deformation, and scene graph sentence reconstruction, which produce the augmented image features and corresponding sentences, respectively. Thus, rather than directly deforming images, FDM-net can efficiently and dynamically enlarge the paired images and texts by learning to deform image features. Extensive experiments are conducted on the widely used novel object captioning dataset, and the results show the effectiveness of our FDM-net. Ablation study and qualitative visualization further give insights of our model.

We study the problem of attacking video recognition models in the black-box setting, where the model information is unknown and the adversary can only make queries to detect the predicted top-1 class and its probability. Compared with the black-box attack on images, attacking videos is more challenging as the computation cost for searching the adversarial perturbations on a video is much higher due to its high dimensionality. To overcome this challenge, we propose a heuristic black-box attack model that generates adversarial perturbations only on the selected frames and regions. More specifically, a heuristic-based algorithm is proposed to measure the importance of each frame in the video towards generating the adversarial examples. Based on the frames' importance, the proposed algorithm heuristically searches a subset of frames where the generated adversarial example has strong adversarial attack ability while keeps the perturbations lower than the given bound. Besides, to further boost the attack efficiency, we propose to generate the perturbations only on the salient regions of the selected frames. In this way, the generated perturbations are sparse in both temporal and spatial domains. Experimental results of attacking two mainstream video recognition methods on the UCF-101 dataset and the HMDB-51 dataset demonstrate that the proposed heuristic black-box adversarial attack method can significantly reduce the computation cost and lead to more than 28% reduction in query numbers for the untargeted attack on both datasets.

Previous researches of sketches often considered sketches in pixel format and leveraged CNN based models in the sketch understanding. Fundamentally, a sketch is stored as a sequence of data points, a vector format representation, rather than the photo-realistic image of pixels. SketchRNN studied a generative neural representation for sketches of vector format by Long Short Term Memory networks (LSTM). Unfortunately, the representation learned by SketchRNN is primarily for the generation tasks, rather than the other tasks of recognition and retrieval of sketches. To this end and inspired by the recent BERT model, we present a model of learning Sketch Bidirectional Encoder Representation from Transformer (Sketch-BERT). We generalize BERT to sketch domain, with the novel proposed components and pre-training algorithms, including the newly designed sketch embedding networks, and the self-supervised learning of sketch gestalt. Particularly, towards the pre-training task, we present a novel Sketch Gestalt Model (SGM) to help train the Sketch-BERT. Experimentally, we show that the learned representation of Sketch-BERT can help and improve the performance of the downstream tasks of sketch recognition, sketch retrieval, and sketch gestalt.

This paper proposes a Hyperbolic Visual Embedding Learning Network for zero-shot recognition. The network learns image embeddings in hyperbolic space, which is capable of preserving the hierarchical structure of semantic classes in low dimensions. Comparing with existing zeroshot learning approaches, the network is more robust because the embedding feature in hyperbolic space better represents class hierarchy and thereby avoid misleading resulted from unrelated siblings. Our network outperforms exiting baselines under hierarchical evaluation with an extremely challenging setting, i.e., learning only from 1,000 categories to recognize 20,841 unseen categories. While under flat evaluation, it has competitive performance as state-of-the-art methods but with five times lower embedding dimensions. Our code is publicly available.

It is challenging in learning a makeup-invariant face verification model, due to (1) insufficient makeup/non-makeup face training pairs, (2) the lack of diverse makeup faces, and (3) the significant appearance changes caused by cosmetics. To address these challenges, we propose a unified Face Morphological Multi-branch Network (FMMu-Net) for makeup-invariant face verification, which can simultaneously synthesize many diverse makeup faces through face morphology network (FM-Net) and effectively learn cosmetics-robust face representations using attention-based multi-branch learning network (AttM-Net). For challenges (1) and (2), FM-Net (two stacked auto-encoders) can synthesize realistic makeup face images by transferring specific regions of cosmetics via cycle consistent loss. For challenge (3), AttM-Net, consisting of one global and three local (task-driven on two eyes and mouth) branches, can effectively capture the complementary holistic and detailed information. Unlike DeepID2 which uses simple concatenation fusion, we introduce a heuristic method AttM-FM, attached to AttM-Net, to adaptively weight the features of different branches guided by the holistic information. We conduct extensive experiments on makeup face verification benchmarks (M-501, M-203, and FAM) and general face recognition datasets (LFW and IJB-A). Our framework FMMu-Net achieves state-of-the-art performances.

Deep neural networks (DNNs) are vulnerable to backdoor attacks which can hide backdoor triggers in DNNs by poisoning training data. A backdoored model behaves normally on clean test images, yet consistently predicts a particular target class for any test examples that contain the trigger pattern. As such, backdoor attacks are hard to detect, and have raised severe security concerns in real-world applications. Thus far, backdoor research has mostly been conducted in the image domain with image classification models. In this paper, we show that existing image backdoor attacks are far less effective on videos, and outline 4 strict conditions where existing attacks are likely to fail: 1) scenarios with more input dimensions (eg. videos), 2) scenarios with high resolution, 3) scenarios with a large number of classes and few examples per class (a ``sparse dataset"), and 4) attacks with access to correct labels (eg. clean-label attacks). We propose the use of a universal adversarial trigger as the backdoor trigger to attack video recognition models, a situation where backdoor attacks are likely to be challenged by the above 4 strict conditions. We show on benchmark video datasets that our proposed backdoor attack can manipulate state-of-the-art video models with high success rates by poisoning only a small proportion of training data (without changing the labels). We also show that our proposed backdoor attack is resistant to state-of-the-art backdoor defense/detection methods, and can even be applied to improve image backdoor attacks. Our proposed video backdoor attack not only serves as a strong baseline for improving the robustness of video models, but also provides a new perspective for more understanding more powerful backdoor attacks.

Retrieving relevant samples across multiple-modalities is a primary topic that receives consistently research interests in multimedia communities, and has benefited various real-world multimedia applications (e.g., text-based image searching). Current models mainly focus on learning a unified visual semantic embedding space to bridge visual contents & text query, targeting at aligning relevant samples from different modalities as neighbors in the embedding space. However, these models did not consider relations between visual components in learning visual representations, resulting in their incapability of distinguishing images with the same visual components but different relations (i.e., Figure 1). To precisely modeling visual contents, we introduce a novel framework that enhanced visual representation with relations between components. Specifically, visual relations are represented by the scene graph extracted from an image, then encoded by the graph convolutional neural networks for learning visual relational features. We combine the relational and compositional representation together for image-text retrieval. Empirical results conducted on the challenging MS-COCO and Flicker 30K datasets demonstrate the effectiveness of our proposed model for cross-modal retrieval task.

Humans can easily recognize actions with only a few examples given, while the existing video recognition models still heavily rely on the large-scale labeled data inputs. This observation has motivated an increasing interest in few-shot video action recognition, which aims at learning new actions with only very few labeled samples. In this paper, we propose a depth guided Adaptive Meta-Fusion Network for few-shot video recognition which is termed as AMeFu-Net. Concretely, we tackle the few-shot recognition problem from three aspects: firstly, we alleviate this extremely data-scarce problem by introducing depth information as a carrier of the scene, which will bring extra visual information to our model; secondly, we fuse the representation of original RGB clips with multiple non-strictly corresponding depth clips sampled by our temporal asynchronization augmentation mechanism, which synthesizes new instances at feature-level; thirdly, a novel Depth Guided Adaptive Instance Normalization (DGAdaIN) fusion module is proposed to fuse the two-stream modalities efficiently. Additionally, to better mimic the few-shot recognition process, our model is trained in the meta-learning way. Extensive experiments on several action recognition benchmarks demonstrate the effectiveness of our model.

The task of person-level action recognition in complex events aims to densely detect pedestrians and individually predict their actions from surveillance videos. In this paper, we present a simple yet efficient pipeline for this task, referred to as TSD-TSM networks. Firstly, we adopt the TSD detector for the pedestrian localization on each single keyframe. Secondly, we generate the sequential ROIs for a person proposal by replicating the adjusted bounding box coordinates around the keyframe. Particularly, we propose to conduct straddling expansion and region squaring on the original bounding box of a person proposal to widen the potential space of motion and interaction and lead to a square box for ROI detection. Finally, we adapt the TSM classifier on the generated ROI sequences to perform action classification and further adopt late fusion to promote the prediction. Our proposed pipeline achieved the 3rd place in the ACM-MM 2020 grand challenge, i.e., Large-scale Human-centric Video Analysis in Complex Events (Track-4), obtaining final 15.31% [email protected] and 20.63% [email protected] on the testing set.

Understanding food recipe requires anticipating the implicit causal effects of cooking actions, such that the recipe can be converted into a graph describing the temporal workflow of the recipe. This is a non-trivial task that involves common-sense reasoning. However, existing efforts rely on hand-crafted features to extract the workflow graph from recipes due to the lack of large-scale labeled datasets. Moreover, they fail to utilize the cooking images, which constitute an important part of food recipes. In this paper, we build MM-ReS, the first large-scale dataset for cooking workflow construction, consisting of 9,850 recipes with human-labeled workflow graphs. Cooking steps are multi-modal, featuring both text instructions and cooking images. We then propose a neural encoder-decoder model that utilizes both visual and textual information to construct the cooking workflow, which achieved over 20% performance gain over existing hand-crafted baselines.

Video visual relation detection (VidVRD) aims at obtaining not only the trajectories of objects but also the dynamic visual relations between them. It provides abundant information for video understanding and can serve as a bridge between vision and language. Compared with visual relation detection on image, VidVRD requires one more step at last called visual relation association which associates relation segments across time dimension into video relations. This step plays an important role in the task but is less studied. Nevertheless, visual relation association is a difficult task as the association process is easily affected by inaccurate tracklet detection and relation prediction in the former steps. In this paper, we propose a novel relation association method called Multiple Hypothesis Association (MHA). It maintains multiple possible relation hypothesis during the association process in order to tolerate and handle the inaccurate or missing problem in the former steps and generate more accurate video relations. Our experiments on the benchmark datasets (Imagenet-VidVRD and VidOR) show that our method outperforms the state-of-the-art methods.

In recent years, the abuse of a face swap technique called deepfake has raised enormous public concerns. So far, a large number of deepfake videos (known as "deepfakes") have been crafted and uploaded to the internet, calling for effective countermeasures. One promising countermeasure against deepfakes is deepfake detection. Several deepfake datasets have been released to support the training and testing of deepfake detectors, such as DeepfakeDetection [1] and FaceForensics++ [23]. While this has greatly advanced deepfake detection, most of the real videos in these datasets are filmed with a few volunteer actors in limited scenes, and the fake videos are crafted by researchers using a few popular deepfake softwares. Detectors developed on these datasets may become less effective against real-world deepfakes on the internet. To better support detection against real-world deepfakes, in this paper, we introduce a new dataset WildDeepfake, which consists of 7,314 face sequences extracted from 707 deepfake videos collected completely from the internet. WildDeepfake is a small dataset that can be used, in addition to existing datasets, to develop and test the effectiveness of deepfake detectors against real-world deepfakes. We conduct a systematic evaluation of a set of baseline detection networks on both existing and our WildDeepfake datasets, and show that WildDeepfake is indeed a more challenging dataset, where the detection performance can decrease drastically. We also propose two (eg. 2D and 3D) Attention-based Deepfake Detection Networks (ADDNets) to leverage the attention masks on real/fake faces for improved detection. We empirically verify the effectiveness of ADDNets on both existing datasets and WildDeepfake. The dataset is available at: https://github.com/deepfakeinthewild/deepfake-in-the-wild.

Expert finding is a task designed to enable recommendation of the right person who can provide high-quality answers to a requester's question. Most previous works try to involve a content-based recommendation, which only superficially comprehends the relevance between a requester's question and the expertise of candidate experts by exploring the content or topic similarity between the requester's question and the candidate experts' historical answers. However, if a candidate expert has never answered a question similar to the requester's question, then existing methods have difficulty making a correct recommendation. Therefore, exploring the implicit relevance between a requester's question and a candidate expert's historical records by perception and reasoning should be taken into consideration. In this study, we propose a novel \textslrecurrent memory reasoning network (RMRN) to perform this task. This method focuses on different parts of a question, and accordingly retrieves information from the histories of the candidate expert.Since only a small percentage of historical records are relevant to any requester's question, we introduce a Gumbel-Softmax-based mechanism to select relevant historical records from candidate experts' answering histories. To evaluate the proposed method, we constructed two large-scale datasets drawn from Stack Overflow and Yahoo! Answer. Experimental results on the constructed datasets demonstrate that the proposed method could achieve better performance than existing state-of-the-art methods.

Despite significant progress in object categorization, in recent years, a number of important challenges remain; mainly, the ability to learn from limited labeled data and to recognize object classes within large, potentially open, set of labels. Zero-shot learning is one way of addressing these challenges, but it has only been shown to work with limited sized class vocabularies and typically requires separation between supervised and unsupervised classes, allowing former to inform the latter but not vice versa. We propose the notion of vocabulary-informed learning to alleviate the above mentioned challenges and address problems of supervised, zero-shot, generalized zero-shot and open set recognition using a unified framework. Specifically, we propose a weighted maximum margin framework for semantic manifold-based recognition that incorporates distance constraints from (both supervised and unsupervised) vocabulary atoms. Distance constraints ensure that labeled samples are projected closer to their correct prototypes, in the embedding space, than to others. We illustrate that resulting model shows improvements in supervised, zero-shot, generalized zero-shot, and large open set recognition, with up to 310K class vocabulary on Animal with Attributes and ImageNet datasets.

Person re-identification (re-id) aims to match people across non-overlapping camera views in a public space. This is a challenging problem because the people captured in surveillance videos often wear similar clothing. Consequently, the differences in their appearance are typically subtle and only detectable at particular locations and scales. In this paper, we propose a deep re-id network (MuDeep) that is composed of two novel types of layers - a multi-scale deep learning layer, and a leader-based attention learning layer. Specifically, the former learns deep discriminative feature representations at different scales, while the latter utilizes the information from multiple scales to lead and determine the optimal weightings for each scale. The importance of different spatial locations for extracting discriminative features is learned explicitly via our leader-based attention learning layer. Extensive experiments are carried out to demonstrate that the proposed MuDeep outperforms the state-of-the-art on a number of benchmarks and has a better generalization ability under a domain generalization setting.

In this paper, we propose Deeply Supervised Object Detectors (DSOD), an object detection framework that can be trained from scratch. Recent advances in object detection heavily depend on the off-the-shelf models pre-trained on large-scale classification datasets like ImageNet and OpenImage. However, one problem is that adopting pre-trained models from classification to detection task may incur learning bias due to the different objective function and diverse distributions of object categories. Techniques like fine-tuning on detection task could alleviate this issue to some extent but are still not fundamental. Furthermore, transferring these pre-trained models across discrepant domains will be more difficult (e.g., from RGB to depth images). Thus, a better solution to handle these critical problems is to train object detectors from scratch, which motivates our proposed method. Previous efforts on this direction mainly failed by reasons of the limited training data and naive backbone network structures for object detection. In DSOD, we contribute a set of design principles for learning object detectors from scratch. One of the key principles is the deep supervision, enabled by layer-wise dense connections in both backbone networks and prediction layers, plays a critical role in learning good detectors from scratch. After involving several other principles, we build our DSOD based on the single-shot detection framework (SSD). We evaluate our method on PASCAL VOC 2007, 2012 and COCO datasets. DSOD achieves consistently better results than the state-of-the-art methods with much more compact models. Specifically, DSOD outperforms baseline method SSD on all three benchmarks, while requiring only 1/2 parameters. We also observe that DSOD can achieve comparable/slightly better results than Mask RCNN [1] + FPN [2] (under similar input size) with only 1/3 parameters, using no extra data or pre-trained models.

In this paper, we propose a novel multimodal matching model for the image and sentence based on their multiple representations. Each representation of the image or sentence undergoes an independent neural network, consisting of multiple layers of nonlinear mappings to yield the corresponding embedding. Besides exploiting the image and sentence relationship based on their embeddings, we propose one novel loss to further exploit the relationship within each single modality, namely, image and sentence based on the yielded multiple embeddings, which is used to train the neural networks simultaneously. The experimental results demonstrate that multiple representations can help to capture the image contents and the sentence semantic meaning more precisely, thus making comprehensive exploitations of the complicated image and sentence matching relationship. More concretely, the proposed matching model significantly outperforms the state-of-the-art approaches in bidirectional image-sentence retrieval on the Flickr8K, Flickr30K, and Microsoft COCO datasets.

This paper aims at learning to score the figure skating sports videos. To address this task, we propose a deep architecture that includes two complementary components, i.e., Self-Attentive LSTM and Multi-scale Convolutional Skip LSTM. These two components can efficiently learn the local and global sequential information in each video. Furthermore, we present a large-scale figure skating sports video dataset - FisV dataset. This dataset includes 500 figure skating videos with the average length of 2 minutes and 50 seconds. Each video is annotated by two scores of nine different referees, i.e., Total Element Score(TES) and Total Program Component Score (PCS). Our proposed model is validated on FisV and MIT-skate datasets. The experimental results show the effectiveness of our models in learning to score the figure skating videos. The codes and datasets would be downloaded from https://github.com/loadder/MS_LSTM.git.

During the past decade, both multi-label learning and zero-shot learning have attracted huge research attention, and significant progress has been made. Multi-label learning algorithms aim to predict multiple labels given one instance, while most existing zero-shot learning approaches target at predicting a single testing label for each unseen class via transferring knowledge from auxiliary seen classes to target unseen classes. However, relatively less effort has been made on predicting multiple labels in the zero-shot setting, which is nevertheless a quite challenging task. In this work, we investigate and formalize a flexible framework consisting of two components, i.e., visual-semantic embedding and zero-shot multi-label prediction. First, we present a deep regression model to project the visual features into the semantic space, which explicitly exploits the correlations in the intermediate semantic layer of word vectors and makes label prediction possible. Then, we formulate the label prediction problem as a pairwise one and employ Ranking SVM to seek the unique multi-label correlations in the embedding space. Furthermore, we provide a transductive multi-label zeroshot prediction approach that exploits the testing data manifold structure. We demonstrate the effectiveness of the proposed approach on three popular multi-label datasets with state-of-theart performance obtained on both conventional and generalized ZSL settings.

The process of learning good representations for machine learning tasks can be very computationally expensive. Typically, we facilitate the same backbones learned on the training set to infer the labels of testing data. Interestingly, This learning and inference paradigm, however, is quite different from the typical inference scheme of human biological visual systems. Essentially, neuroscience studies have shown that the right hemisphere of the human brain predominantly makes a fast processing of low-frequency spatial signals, while the left hemisphere more focuses on analyzing high-frequency information in a slower way. And the low-pass analysis helps facilitate the high-pass analysis via a feedback form. Inspired by this biological vision mechanism, this paper explores the possibility of learning a layer-skippable inference network. Specifically, we propose a layer-skippable network that dynamically carries out coarse-tofine object categorization. Such a network has two branches to jointly deal with both coarse and fine-grained classification tasks. The layer-skipping mechanism is proposed to learn a gating network by generating dynamic inference graphs, and reducing the computational cost by detouring the inference path from some layers. This adaptive path inference strategy endows the network with better flexibility and larger capacity and makes the high-performance deep networks with dynamic structures. To efficiently train the gating network, a novel ranking-based loss function is presented. Furthermore, the learned representations are enhanced by the proposed top-down feedback facilitation and feature-wise affine transformation, individually. The former one employs features of a coarse branch to help the finegrained object recognition task, while the latter one encodes the selected path to enhance the final feature representations. Extensive experiments are conducted on several widely used coarse-to-fine object categorization benchmarks, and promising results are achieved by our proposed model. Quite surprisingly, our layer-skipping mechanism improves the network robustness to adversarial attacks. The codes and models are released on https://github.com/avalonstrel/DSN.

Generating realistic images with the guidance of reference images and human poses is challenging. Despite the success of previous works on synthesizing person images in the iconic views, no efforts are made towards the task of poseguided image synthesis in the non-iconic views. Particularly, we find that previous models cannot handle such a complex task, where the person images are captured in the non-iconic views by commercially-available digital cameras. To this end, we propose a new framework - Multi-branch Refinement Network (MR-Net), which utilizes several visual cues, including target person poses, foreground person body and scene images parsed. Furthermore, a novel Region of Interest (RoI) perceptual loss is proposed to optimize the MR-Net. Extensive experiments on two non-iconic datasets, Penn Action and BBC-Pose, as well as an iconic dataset - Market-1501, show the efficacy of the proposed model that can tackle the problem of pose-guided person image generation from the non-iconic views. The data, models, and codes are downloadable from https://github.com/loadder/MR-Net.

Visual and semantic saliency are important in image captioning. However, single-phase image captioning benefits little from limited saliency without a saliency predictor. In this paper, a novel saliency-enhanced re-captioning framework via two-phase learning is proposed to enhance the single-phase image captioning. In the framework, visual saliency and semantic saliency are distilled from the first-phase model and fused with the second-phase model for model self-boosting. The visual saliency mechanism can generate a saliency map and a saliency mask for an image without learning a saliency map predictor. The semantic saliency mechanism sheds some lights on the properties of words with part-of-speech Noun in a caption. Besides, another type of saliency, sample saliency is proposed to explicitly compute the saliency degree of each sample, which helps for more robust image captioning. In addition, how to combine the above three types of saliency for further performance boost is also examined. Our framework can treat an image captioning model as a saliency extractor, which may benefit other captioning models and related tasks. The experimental results on both the Flickr30k and MSCOCO datasets show that the saliency-enhanced models can obtain promising performance gains.

Emotional content is a crucial ingredient in user-generated videos. However, the sparsity of emotional expressions in the videos poses an obstacle to visual emotion analysis. In this paper, we propose a new neural approach, Bi-stream Emotion Attribution-Classification Network (BEAC-Net), to solve three related emotion analysis tasks: emotion recognition, emotion attribution, and emotion-oriented summarization, in a single integrated framework. BEAC-Net has two major constituents, an attribution network and a classification network. The attribution network extracts the main emotional segment that classification should focus on in order to mitigate the sparsity issue. The classification network utilizes both the extracted segment and the original video in a bi-stream architecture. We contribute a new dataset for the emotion attribution task with human-annotated ground-truth labels for emotion segments. Experiments on two video datasets demonstrate superior performance of the proposed framework and the complementary nature of the dual classification streams.

Given one or a few training instances of novel classes, oneshot learning task requires that the classifier generalizes to these novel classes. Directly training one-shot classifier may suffer from insufficient training instances in one-shot learning. Previous one-shot learning works investigate the metalearning or metric-based algorithms; in contrast, this paper proposes a Self-Training Jigsaw Augmentation (Self-Jig) method for one-shot learning. Particularly, we solve one-shot learning by directly augmenting the training images through leveraging the vast unlabeled instances. Precisely our proposed Self-Jig algorithm can synthesize new images from the labeled probe and unlabeled gallery images. The labels of gallery images are predicted to help the augmentation process, which can be taken as a self-training scheme. Intrinsically, we argue that we provide a very useful way of directly generating massive amounts of training images for novel classes. Extensive experiments and ablation study not only evaluate the efficacy but also reveal the insights, of the proposed Self-Jig method.

Sequence-to-sequence models incorporated with attention mechanism have shown promising improvements on video captioning. While there is rich information both inside and between frames, spatial attention is rarely explored and motion information is usually handled by 3D-CNNs as just another modality for fusion. On the other hand, researches about human perception suggest that apparent motion can attract attention. Motivated by this, we aim to learn spatial attention on video frames under the guidance of motion information for caption generation. We present a novel video captioning framework by utilizing Motion Guided Spatial Attention (MGSA). The proposed MGSA exploits the motion between video frames by learning spatial attention from stacked optical flow images with a custom CNN. To further relate the spatial attention maps of video frames, we designed a Gated Attention Recurrent Unit (GARU) to adaptively incorporate previous attention maps. The whole framework can be trained in an end-to-end manner. We evaluate our approach on two benchmark datasets, MSVD and MSR-VTT. The experiments show that our designed model can generate better video representation and state of the art results are obtained under popular evaluation metrics such as BLEU@4, CIDEr, and METEOR.

This paper presents an efficient algorithm to tackle temporal localization of activities in videos via sentence queries. The task differs from traditional action localization in three aspects: (1) Activities are combinations of various kinds of actions and may span a long period of time. (2) Sentence queries are not limited to a predefined list of classes. (3) The videos usually contain multiple different activity instances. Traditional proposal-based approaches for action localization that only consider the class-agnostic “actionness” of video snippets are insufficient to tackle this task. We propose a novel Semantic Activity Proposal (SAP) which integrates the semantic information of sentence queries into the proposal generation process to get discriminative activity proposals. Visual and semantic information are jointly utilized for proposal ranking and refinement. We evaluate our algorithm on the TACoS dataset and the Charades-STA dataset. Experimental results show that our algorithm outperforms existing methods on both datasets, and at the same time reduces the number of proposals by a factor of at least 10.

Undersampling has been widely used in the class-imbalance learning area. The main deficiency of most existing undersampling methods is that their data sampling strategies are heuristic-based and independent of the used classifier and evaluation metric. Thus, they may discard informative instances for the classifier during the data sampling. In this work, we propose a meta-learning method built on the undersampling to address this issue. The key idea of this method is to parametrize the data sampler and train it to optimize the classification performance over the evaluation metric. We solve the non-differentiable optimization problem for training the data sampler via reinforcement learning. By incorporating evaluation metric optimization into the data sampling process, the proposed method can learn which instance should be discarded for the given classifier and evaluation metric. In addition, as a data level operation, this method can be easily applied to arbitrary evaluation metric and classifier, including non-parametric ones (e.g., C4.5 and KNN). Experimental results on both synthetic and realistic datasets demonstrate the effectiveness of the proposed method.

Recent efforts have been made on continuously learning concepts over time, i.e., class-incremental learning, which, however, is still an open question to the task of object detection. For the first time, this paper systematically studies the task of learning an end-to-end class-incremental object detection model. Remarkably, this work tailors Faster R-CNN and proposes a Class-Incremental Faster R-CNN (CIFRCN) model that can dynamically add new classes by only using a few labeled images of new objects. Particularly, the domain of "foreground" in a Region Proposal Network is expanded to generate accurate bounding box proposals, and the classifier in Fast R-CNN is enlarged using knowledge distillation to accurately classify the proposals. Extensive experiments on two challenging datasets are conducted to demonstrate the efficacy of the proposed model for adding new objects in a class-incremental way, whereas other detectors quickly fail.

Deep learning has achieved great successes in solving specific artificial intelligence problems recently. Substantial progresses are made on Computer Vision (CV) and Natural Language Processing (NLP). As a connection between the two worlds of vision and language, video captioning is the task of producing a natural-language utterance (usually a sentence) that describes the visual content of a video. The task is naturally decomposed into two sub-tasks. One is to encode a video via a thorough understanding and learn visual representation. The other is caption generation, which decodes the learned representation into a sequential sentence, word by word. In this survey, we first formulate the problem of video captioning, then review state-of-the-art methods categorized by their emphasis on vision or language, and followed by a summary of standard datasets and representative approaches. Finally, we highlight the challenges which are not yet fully understood in this task and present future research directions.

Character-level Chinese named entity recognition (NER) that applies long short-term memory (LSTM) to incorporate lexicons has achieved great success. However, this method fails to fully exploit GPU parallelism and candidate lexicons can conflict. In this work, we propose a faster alternative to Chinese NER: a convolutional neural network (CNN)-based method that incorporates lexicons using a rethinking mechanism. The proposed method can model all the characters and potential words that match the sentence in parallel. In addition, the rethinking mechanism can address the word conflict by feeding back the high-level features to refine the networks. Experimental results on four datasets show that the proposed method can achieve better performance than both word-level and character-level baseline methods. In addition, the proposed method performs up to 3.21 times faster than state-of-the-art methods, while realizing better performance.

Object detection for automatic visual checkout in self-service vending machines is attracting significant attention in the retail industry. However, several critical challenges have not received enough attention. First, large-scale, high-quality retail image datasets are urgently demanded to train and evaluate the detection models. Second, the trained models should be able to cope with the frequently added new goods at low cost, while most cutting-edge models cannot. In this paper, we propose a new hierarchical large-scale object detection dataset, called Take Goods from Shelves (TGFS), containing 38K images of 24 fine-grained and 3 coarse classes. A preliminary method for solving the goods-adding problem, called Faster R-CNN Class-incremental Object Detector (FCIOD), is also described and evaluated. In addition, several popular methods are benchmarked on the TGFS dataset.

Face frontalization has always been an important field. Recently, with the introduction of generative adversarial networks (GANs), face frontalization has achieved remarkable success. A critical challenge during face frontalization is to ensure the features of the original profile image are retained. Even though some state-of-the-art methods can preserve identity features while rotating the face to the frontal view, they still have difficulty preserving facial expression features. Therefore, we propose the novel triangle cycle-consistent generative adversarial networks for the face frontalization task, termed TC-GAN. Our networks contain two generators and one discriminator. One of the generators generates the frontal contour, and the other generates the facial features. They work together to generate a photo-realistic frontal view of the face. We also introduce cycle-consistent loss to retain feature information effectively. To validate the advantages of TC-GAN, we apply it to the face frontalization task on two datasets. The experimental results demonstrate that our method can perform large-pose face frontalization while preserving the facial features (both identity and expression). To the best of our knowledge, TC-GAN outperforms the state-of-the-art methods in the preservation of facial identity and expression features during face frontalization.

Recently, spatio-temporal convolutional networks have achieved prominent performance in action classification. However, debates on the importance of temporal information lead to the rethinking of these architectures. In this work, we propose to employ sparse temporal convolutional operations in networks for efficient action modeling. We demonstrate that the explicit temporal feature interactions can be largely reduced without any degradation. And towards better scalability, we use causal convolutions for temporal feature learning. Under causality constraints, we replenish the model with auxiliary self-supervised tasks, namely video prediction and frame order discrimination. Besides, a gradient based multi-task learning algorithm is introduced for guaranteeing the dominance of action recognition task. The proposed model matches or outperforms the state-of-the-art methods on Kinetics, Something-Something V2, UCF101 and HMDB51 datasets.

One-shot learning aims to recognize novel target classes from few examples by transferring knowledge from source classes, under a general assumption that the source and target classes are semantically related but not exactly the same. Based on this assumption, recent work has focused on image-based one-shot learning, while little work has addressed video-based one shot learning. One of the challenges lies in that it is difficult to maintain the disjoint-class assumption for videos, since video clips of target classes may potentially appear in the videos of source classes. To address this issue, we introduce a novel setting, termed as embodied agents based one-shot learning, which leverages synthetic videos produced in a virtual environment to understand realistic videos of target classes. In this setting, we further propose two types of learning tasks: embodied one-shot video domain adaptation and embodied one-shot video transfer recognition. These tasks serve as a testbed for evaluating video related one-shot learning tasks. In addition, we propose a general video segment augmentation method, which significantly facilitates a variety of one-shot learning tasks. Experimental results validate the soundness of our setting and learning tasks, and also show the effectiveness of our augmentation approach to video recognition in the small-sample size regime.

Instance-level image retrieval is a long-standing and challenging problem in multimedia. Recently, fine-tuning Convolutional Neural Networks (CNNs) has become a promising direction, and a number of successful strategies based on global CNN descriptors have been proposed. However, it is difficult to make direct comparisons and draw conclusions due to different settings and/or datasets. The goal of this paper is two-fold. Firstly, we present a unified implementation of modern global-CNN-based retrieval systems, break such a system into six major components, and investigate each part individually as well as globally when considering different configurations. We conduct a systematic series of experiments on a component-by-component basis and find an optimal solution in designing such a system. Secondly, we introduce a novel joint loss function with learnable parameter for fine-tuning for retrieval tasks and show, with extensive experiments, significant improvement over previous works. On the new and challenging large-scale Google-Landmarks-Dataset, we set a baseline for future research and comparisons, while on traditional retrieval benchmarks such as Oxford5k and Paris6k, as well as their recent revised versions ROxford5k and RParis6k, we achieve state-of-the-art performance under all three (Easy, Medium, and Hard) evaluation protocals by a large margin compared to competing methods.

Deep neural networks (DNNs) are known for their vulnerability to adversarial examples. These are examples that have undergone small, carefully crafted perturbations, and which can easily fool a DNN into making misclassifications at test time. Thus far, the field of adversarial research has mainly focused on image models, under either a white-box setting, where an adversary has full access to model parameters, or a black-box setting where an adversary can only query the target model for probabilities or labels. Whilst several white-box attacks have been proposed for video models, black-box video attacks are still unexplored. To close this gap, we propose the first black-box video attack framework, called V-BAD. V-BAD utilizestentative perturbations transferred from image models andpartition-based rectifications found by the NES to obtain good adversarial gradient estimates with fewer queries to the target model. V-BAD is equivalent to estimating the projection of the adversarial gradient on a selected subspace. Using three benchmark video datasets, we demonstrate that V-BAD can craft both untargeted and targeted attacks to fool two state-of-the-art deep video recognition models. For the targeted attack, it achieves $>$93% success rate using only an average of $3.4 \sim 8.4 \times 10^4$ queries, a similar number of queries to state-of-the-art black-box image attacks. This is despite the fact that videos often have two orders of magnitude higher dimensionality than static images. We believe that V-BAD is a promising new tool to evaluate and improve the robustness of video recognition models to black-box adversarial attacks.

Sketch has been employed as an effective communication tool to express the abstract and intuitive meaning of object. While content-based sketch recognition has been studied for several decades, the instance-level Sketch Based Image Retrieval (iSBIR) task has attracted significant research attention recently. In many previous iSBIR works -- TripletSN, and DSSA, edge maps were employed as intermediate representations in bridging the cross-domain discrepancy between photos and sketches. However, it is nontrivial to efficiently train and effectively use the edge maps in an iSBIR system. Particularly, we find that such an edge map based iSBIR system has several major limitations. First, the system has to be pre-trained on a significant amount of edge maps, either from large-scale sketch datasets, e.g., TU-Berlin~\citeeitz2012hdhso, or converted from other large-scale image datasets, e.g., ImageNet-1K\citedeng2009imagenet dataset. Second, the performance of such an iSBIR system is very sensitive to the quality of edge maps. Third and empirically, the multi-cropping strategy is essentially very important in improving the performance of previous iSBIR systems. To address these limitations, this paper advocates an end-to-end iSBIR system without using the edge maps. Specifically, we present a Triplet Classification Network (TC-Net) for iSBIR which is composed of two major components: triplet Siamese network, and auxiliary classification loss. Our TC-Net can break the limitations existed in previous works. Extensive experiments on several datasets validate the efficacy of the proposed network and system.

Facial expression is important in understanding our social interaction. Thus the ability to recognize facial expression enables the novel multimedia applications. With the advance of recent deep architectures, research on facial expression recognition has achieved great progress. However, these models are still suffering from the problems of lacking sufficient and diverse high quality training faces, vulnerability to the facial variations, and recognizing a limited number of basic types of emotions. To tackle these problems, this paper proposes a novel end-to-end Compositional Generative Adversarial Network (Comp-GAN) that is able to synthesize new face images with specified poses and desired facial expressions; and such synthesized images can be further utilized to help train a robust and generalized expression recognition model. Essentially, Comp-GAN can dynamically change the expression and pose of faces according to the input images while keeping the identity information. Specifically, the generator has two major components: one for generating images with desired expression and the other for changing the pose of faces. Furthermore, a face reconstruction learning process is applied to re-generate the input image and constrains the generator for preserving the key information such as facial identity. For the first time, various one/zero-shot facial expression recognition tasks have been created. We conduct extensive experiments to show that the images generated by Comp-GAN are helpful to improve the performance of one/zero-shot facial expression recognition.

This paper presents LiteEval, a simple yet effective coarse-to-fine framework for resource efficient video recognition, suitable for both online and offline scenarios. Exploiting decent yet computationally efficient features derived at a coarse scale with a lightweight CNN model, LiteEval dynamically decides on-the-fly whether to compute more powerful features for incoming video frames at a finer scale to obtain more details. This is achieved by a coarse LSTM and a fine LSTM operating cooperatively, as well as a conditional gating module to learn when to allocate more computation. Extensive experiments are conducted on two large-scale video benchmarks, FCVID and ActivityNet, and the results demonstrate LiteEval requires substantially less computation while offering excellent classification accuracy for both online and offline predictions.

Social media users create millions of microblog entries on various topics each day. Retweet behaviour play a crucial role in spreading topics on social media. Retweet prediction task has received considerable attention in recent years. The majority of existing retweet prediction methods are focus on modeling user preference by utilizing various information, such as user profiles, user post history, user following relationships, etc. Yet, the users exposures towards real-time posting from their followees contribute significantly to making retweet predictions, considering that the users may participate into the hot topics discussed by their followees rather than be limited to their previous interests. To make efficient use of hot topics, we propose a novel masked self-attentive model to perform the retweet prediction task by perceiving the hot topics discussed by the users' followees. We incorporate the posting histories of users with external memory and utilize a hierarchical attention mechanism to construct the users' interests. Hence, our model can be jointly hot-topic aware and user interests aware to make a final prediction. Experimental results on a dataset collected from Twitter demonstrated that the proposed method can achieve better performance than state-of-the-art methods.

Image retagging aims to improve the tag quality of social images by completing the missing tags, rectifying the noise-corrupted tags, and assigning new high-quality tags. Recent approaches simultaneously explore visual, user and tag information to improve the performance of image retagging by mining the tag-image-user associations. However, such methods will become computationally infeasible with the rapidly increasing number of images, tags and users. It has been proven that the anchor graph can significantly accelerate large-scale graph-based learning by exploring only a small number of anchor points. Inspired by this, we propose a novel Social anchor-Unit GrAph Regularized Tensor Completion (SUGAR-TC) method to efficiently refine the tags of social images, which is insensitive to the scale of data. First, we construct an anchor-unit graph across multiple domains (e.g., image and user domains) rather than traditional anchor graph in a single domain. Second, a tensor completion based on Social anchor-Unit GrAph Regularization (SUGAR) is implemented to refine the tags of the anchor images. Finally, we efficiently assign tags to non-anchor images by leveraging the relationship between the non-anchor units and the anchor units. Experimental results on a real-world social image database well demonstrate the effectiveness and efficiency of SUGAR-TC, outperforming the state-of-the-art methods.

The ability to quickly recognize and learn new visual concepts from limited samples enables humans to quickly adapt to new tasks and environments. This ability is enabled by semantic association of novel concepts with those that have already been learned and stored in memory. Computers can start to ascertain similar abilities by utilizing a semantic concept space. A concept space is a high-dimensional semantic space in which similar abstract concepts appear close and dissimilar ones far apart. In this paper, we propose a novel approach to one-shot learning that builds on this core idea. Our approach learns to map a novel sample instance to a concept, relates that concept to the existing ones in the concept space and, using these relationships, generates new instances, by interpolating among the concepts, to help learning. Instead of synthesizing new image instance, we propose to directly synthesize instance features by leveraging semantics using a novel auto-encoder network we call dual TriNet. The encoder part of the TriNet learns to map multi-layer visual features from CNN to a semantic vector. In semantic space, we search for related concepts, which are then projected back into the image feature spaces by the decoder portion of the TriNet. Two strategies in the semantic space are explored. Notably, this seemingly simple strategy results in complex augmented feature distributions in the image feature space, leading to substantially better performance. The codes and models are released in the github: https://github.com/tankche1/ Semantic-Feature-Augmentation-in-Few-shot-Learning.

The ability to recognize actions throughout a video is essential for surveillance, self-driving, and many other applications. Although many researchers have investigated deep neural networks to get a better result in video action recognition, these networks usually require a large number of well-labeled data to train. In this paper, we introduce a dense dilated network to collect action information from snippet-level to global-level. The dilated dense network is composed of the blocks with densely connected dilated convolutions layers. Our proposed framework is capable of fusing outputs from each layer to learn high-level representations, and these representations are robust even with only a few training snippets. We study different spatial and temporal modality fusing configurations and introduce a novel temporal guided fusion upon the dense dilated network which can further boost the performance. We conduct extensive experiments on two popular video action datasets: UCF101 and HMDB51. The experiments demonstrate the effectiveness of our proposed framework.

Recent studies have shown that spatial relationships among objects are very important for visual recognition, since they can provide rich clues on object contexts within the images. In this article, we introduce a novel method to learn the Semantic Feature Map (SFM) with attention-based deep neural networks for image and video classification in an end-to-end manner, aiming to explicitly model the spatial object contexts within the images. In particular, we explicitly apply the designed gate units to the extracted object features for important objects selection and noise removal. These selected object features are then organized into the proposed SFM, which is a compact and discriminative representation with the spatial information among objects preserved. Finally, we employ either Fully Convolutional Networks (FCN) or Long-Short Term Memory (LSTM) as the classifiers on top of the SFM for content recognition. A novel multi-task learning framework with image classification loss, object localization loss, and grid labeling loss are also introduced to help better learn the model parameters. We conduct extensive evaluations and comparative studies to verify the effectiveness of the proposed approach on Pascal VOC 2007/2012 and MS-COCO benchmarks for image classification. In addition, the experimental results also show that the SFMs learned from the image domain can be successfully transferred to CCV and FCVID benchmarks for video classification.

The task of adopting a model with good performance to a target domain that is different from the source domain used for training has received considerable attention in sentiment analysis.Most existing approaches mainly focus on learning representations that are domain-invariant in both the source and target domains.Few of them pay attention to domain specific information, which should also be informative.In this work, we propose a method to simultaneously extract domain specific and invariant representations and train a classifier on each of the representation, respectively.And we introduce a few target domain labeled data for learning domain-specific information.To effectively utilize the target domain labeled data, we train the domain-invariant representation based classifier with both the source and target domain labeled data and train the domain-specific representation based classifier with only the target domain labeled data.These two classifiers then boost each other in a co-training style.Extensive sentiment analysis experiments demonstrated that the proposed method could achieve better performance than state-of-the-art methods.

In recent years, the task of reformulating natural language queries has received considerable attention from both industry and academic communities. Because of the lexical chasm problem between natural language queries and web documents, if we directly use natural language queries as inputs for retrieval, the results are usually unsatisfactory. In this work, we formulated the task as a translation problem to convert natural language queries into keyword queries. Since the nature language queries users input are diverse and multi-faceted, general encoder-decoder models cannot effectively handle low-frequency words and out-of-vocabulary words. We propose a novel encoder-decoder method with two decoders: the pointer decoder firstly extracts query terms directly from the source text via copying mechanism, then the generator decoder generates query terms using two attention modules simultaneously considering the source text and extracted query terms. For evaluation and training, we also proposed a semi-automatic method to construct a large-scale dataset about natural language query-keyword query pairs. Experimental results on this dataset demonstrated that our model could achieve better performance than the previous state-of-the-art methods.

Inspired by the recent neuroscience studies on the left-right asymmetry of the human brain in processing low and high spatial frequency information, this paper introduces a dual skipping network which carries out coarse-to-fine object categorization. Such a network has two branches to simultaneously deal with both coarse and fine-grained classification tasks. Specifically, we propose a layer-skipping mechanism that learns a gating network to predict which layers to skip in the testing stage. This layer-skipping mechanism endows the network with good flexibility and capability in practice. Evaluations are conducted on several widely used coarse-to-fine object categorization benchmarks, and promising results are achieved by our proposed network model.

Facial attribute recognition is an important and yet challenging research topic. Different from most previous approaches which predict attributes only based on the whole images, this paper leverages facial parts locations for better attribute prediction. A facial abstraction image which contains both local facial parts and facial texture information is introduced. This abstraction image is generated by a Generative Adversarial Network (GAN). Then we build a dual-path facial attribute recognition network to utilize features from the original face images and facial abstraction images. Empirically, the features of facial abstraction images are complementary to features of original face images. With the facial parts localized by the abstraction images, our method improves facial attributes recognition, especially the attributes located on small face regions. Extensive evaluations conducted on CelebA and LFWA benchmark datasets show that state-of-the-art performance is achieved.

Recently, video action recognition has been widely studied. Training deep neural networks requires a large amount of well-labeled videos. On the other hand, videos in the same class share high-level semantic similarity. In this paper, we introduce a novel neural network architecture to simultaneously capture local and long-term spatial temporal information. The dilated dense network is proposed with the blocks being composed of densely-connected dilated convolutions layers. The proposed framework is capable of fusing each layer's outputs to learn high-level representations, and the representations are robust even with only few training snippets. The aggregations of dilated dense blocks are also explored. We conduct extensive experiments on UCF101 and demonstrate the effectiveness of our proposed method, especially with few training examples.

In this paper, we study the challenging problem of categorizing videos according to high-level semantics such as the existence of a particular human action or a complex event. Although extensive efforts have been devoted in recent years, most existing works combined multiple video features using simple fusion strategies and neglected the utilization of inter-class semantic relationships. This paper proposes a novel unified framework that jointly exploits the feature relationships and the class relationships for improved categorization performance. Specifically, these two types of relationships are estimated and utilized by imposing regularizations in the learning process of a deep neural network (DNN). Through arming the DNN with better capability of harnessing both the feature and the class relationships, the proposed regularized DNN (rDNN) is more suitable for modeling video semantics. We show that rDNN produces better performance over several state-of-the-art approaches. Competitive results are reported on the well-known Hollywood2 and Columbia Consumer Video benchmarks. In addition, to stimulate future research on large scale video categorization, we collect and release a new benchmark dataset, called FCVID, which contains 91,223 Internet videos and 239 manually annotated categories.

Emotion is a key element in user-generated video. However, it is difficult to understand emotions conveyed in such videos due to the complex and unstructured nature of user-generated content and the sparsity of video frames expressing emotion. In this paper, for the first time, we propose a technique for transferring knowledge from heterogeneous external sources, including image and textual data, to facilitate three related tasks in understanding video emotion: emotion recognition, emotion attribution and emotion-oriented summarization. Specifically, our framework (1) learns a video encoding from an auxiliary emotional image dataset in order to improve supervised video emotion recognition, and (2) transfers knowledge from an auxiliary textual corpora for zero-shot recognition of emotion classes unseen during training. The proposed technique for knowledge transfer facilitates novel applications of emotion attribution and emotion-oriented summarization. A comprehensive set of experiments on multiple datasets demonstrate the effectiveness of our framework.

In this paper, we propose a novel fine-grained dictionary learning method for image classification. To learn a high-quality discriminative dictionary, three types of multispecific subdictionaries, i.e., class-specific dictionaries (CSDs), universal dictionary (UD), and family-specific dictionaries (FSDs), are simultaneously uncovered. Here, CSDs and UD, respectively, model the patterns for each class and the patterns irrespective of any class. FSDs can help reveal the shared patterns between multiple image classes, by filling the gap between the patterns in CSDs and UD. The dependence among image classes is revealed by the shared FSDs, and a common FSD can be assigned to several classes to represent their residual. Finally, the most discriminative FSD for each class is identified by minimizing the sparse reconstruction error. Extensive experiments are conducted on different widely used data sets for image classification. The results demonstrate the superior performance of the proposed method over some state-of-the-art methods.

As one of the most common human helminths, hookworm is a leading cause of maternal and child morbidity, which seriously threatens human health. Recently, wireless capsule endoscopy (WCE) has been applied to automatic hookworm detection. Unfortunately, it remains a challenging task. In recent years, deep convolutional neural network (CNN) has demonstrated impressive performance in various image and video analysis tasks. In this paper, a novel deep hookworm detection framework is proposed for WCE images, which simultaneously models visual appearances and tubular patterns of hookworms. This is the first deep learning framework specifically designed for hookworm detection in WCE images. Two CNN networks, namely edge extraction network and hookworm classification network, are seamlessly integrated in the proposed framework, which avoid the edge feature caching and speed up the classification. Two edge pooling layers are introduced to integrate the tubular regions induced from edge extraction network and the feature maps from hookworm classification network, leading to enhanced feature maps emphasizing the tubular regions. Experiments have been conducted on one of the largest WCE datasets with WCE images, which demonstrate the effectiveness of the proposed hookworm detection framework. It significantly outperforms the state-of-the-art approaches. The high sensitivity and accuracy of the proposed method in detecting hookworms shows its potential for clinical application.

Item-to-item collaborative filtering (aka.item-based CF) has been long used for building recommender systems in industrial settings, owing to its interpretability and efficiency in real-time personalization. It builds a user's profile as her historically interacted items, recommending new items that are similar to the user's profile. As such, the key to an item-based CF method is in the estimation of item similarities. Early approaches use statistical measures such as cosine similarity and Pearson coefficient to estimate item similarities, which are less accurate since they lack tailored optimization for the recommendation task. In recent years, several works attempt to learn item similarities from data, by expressing the similarity as an underlying model and estimating model parameters by optimizing a recommendation-aware objective function. While extensive efforts have been made to use shallow linear models for learning item similarities, there has been relatively less work exploring nonlinear neural network models for item-based CF. In this work, we propose a neural network model named Neural Attentive Item Similarity model (NAIS) for item-based CF. The key to our design of NAIS is an attention network, which is capable of distinguishing which historical items in a user profile are more important for a prediction. Compared to the state-of-the-art item-based CF method Factored Item Similarity Model (FISM) [1] , our NAIS has stronger representation power with only a few additional parameters brought by the attention network. Extensive experiments on two public benchmarks demonstrate the effectiveness of NAIS. This work is the first attempt that designs neural network models for item-based CF, opening up new research possibilities for future developments of neural recommender systems.

Videos are inherently multimodal. This paper studies the problem of exploiting the abundant multimodal clues for improved video classification performance. We introduce a novel hybrid deep learning framework that integrates useful clues from multiple modalities, including static spatial appearance information, motion patterns within a short time window, audio information, as well as long-range temporal dynamics. More specifically, we utilize three Convolutional Neural Networks (CNNs) operating on appearance, motion, and audio signals to extract their corresponding features. We then employ a feature fusion network to derive a unified representation with an aim to capture the relationships among features. Furthermore, to exploit the long-range temporal dynamics in videos, we apply two long short-term memory (LSTM) networks with extracted appearance and motion features as inputs. Finally, we also propose refining the prediction scores by leveraging contextual relationships among video semantics. The hybrid deep learning framework is able to exploit a comprehensive set of multimodal features for video classification. Through an extensive set of experiments, we demonstrate that: 1) LSTM networks that model sequences in an explicitly recurrent manner are highly complementary to the CNN models; 2) the feature fusion network that produces a fused representation through modeling feature relationships outperforms a large set of alternative fusion strategies; and 3) the semantic context of video classes can help further refine the predictions for improved performance. Experimental results on two challenging benchmarks-the UCF-101 and the Columbia Consumer Videos (CCV)-provide strong quantitative evidence that our framework can produce promising results: 93.1% on the UCF-101 and 84.5% on the CCV, outperforming several competing methods with clear margins.

The papers in this special section focus on the topic of video analytics. Also known as video content analysis, video analytics refer to the capability of automatically analyzing video to extract knowledge/information and detect and determine temporal and spatial events. The algorithms designed for these analytics can be implemented as software on general-purpose machines, or as hardware in specialized video processing units. Video analytics is still an emerging technology with techniques that are continuously being developed to help make widespread implementation feasible in the years ahead. Such analytics has been typically used in semantic categorization and retrieval of video databases. A goal of this special issue is to focus on video analytics beyond categorization and retrieval. With increasing hardware capability and advances in algorithms used, real-time video analytics is now being used in a wide range of domains including entertainment, health-care, retail, automotive, transport, home automation, emotion analysis, aesthetics, inappropriate content detection, safety and security. For instance, video analytics are increasingly being deployed for real-time alerts in situation monitoring systems such as traffic surveillance (vehicle counting), counting people in lines (some hospitals are using this to get more nurses from a less busy department to serve thewaiting patients) and in manufacturing (for monitoring and counting). From the sensing aspect, 3-D cameras such as RGB-D and LiDAR (Light Detection and Ranging) cameras are becoming more and more affordable, enabling additional areas of research and applications, such as self-driving cars employing video analytics on LiDAR captured data for path planning as well as obstacle detection.

Features extracted by deep networks have been popular in many visual search tasks. This article studies deep network structures and training schemes for mobile visual search. The goal is to learn an effective yet portable feature representation that is suitable for bridging the domain gap between mobile user photos and (mostly) professionally taken product images while keeping the computational cost acceptable for mobile-based applications. The technical contributions are twofold. First, we propose an alternative of the contrastive loss popularly used for training deep Siamese networks, namely robust contrastive loss, where we relax the penalty on some positive and negative pairs to alleviate overfitting. Second, a simple multitask fine-tuning scheme is leveraged to train the network, which not only utilizes knowledge from the provided training photo pairs but also harnesses additional information from the large ImageNet dataset to regularize the fine-tuning process. Extensive experiments on challenging real-world datasets demonstrate that both the robust contrastive loss and the multitask fine-tuning scheme are effective, leading to very promising results with a time cost suitable for mobile product search scenarios.

We propose a fast first-order method to solve multi-term nonsmooth composite convex minimization problems by employing a recent proximal average approximation technique and a novel adaptive parameter tuning technique. Thanks to this powerful parameter tuning technique, the proximal gradient step can be performed with a much larger stepsize in the algorithm implementation compared with the prior PA-APG method, which is the core to enable significant improvements in practical performance. Moreover, by choosing the approximation parameter adaptively, the proposed method is shown to enjoy the O(1/k) iteration complexity theoretically without needing any extra computational cost, while the PA-APG method incurs much more iterations for convergence. The preliminary experimental results on overlapping group Lasso and graph-guided fused Lasso problems confirm our theoretic claim well, and indicate that the proposed method is almost five times faster than the state-of-the-art PA-APG method and therefore suitable for higher-precision required optimization.

This paper focuses on a novel and challenging vision task, dense video captioning, which aims to automatically describe a video clip with multiple informative and diverse caption sentences. The proposed method is trained without explicit annotation of fine-grained sentence to video region-sequence correspondence, but is only based on weak video-level sentence annotations. It differs from existing video captioning systems in three technical aspects. First, we propose lexical fully convolutional neural networks (Lexical-FCN) with weakly supervised multi-instance multi-label learning to weakly link video regions with lexical labels. Second, we introduce a novel submodular maximization scheme to generate multiple informative and diverse region-sequences based on the Lexical-FCN outputs. A winner-takes-all scheme is adopted to weakly associate sentences to region-sequences in the training phase. Third, a sequence-to-sequence learning based language model is trained with the weakly supervised information obtained through the association process. We show that the proposed method can not only produce informative and diverse dense captions, but also outperform state-of-the-art single video captioning methods by a large margin.

Person Re-identification (re-id) aims to match people across non-overlapping camera views in a public space. It is a challenging problem because many people captured in surveillance videos wear similar clothes. Consequently, the differences in their appearance are often subtle and only detectable at the right location and scales. Existing re-id models, particularly the recently proposed deep learning based ones match people at a single scale. In contrast, in this paper, a novel multi-scale deep learning model is proposed. Our model is able to learn deep discriminative feature representations at different scales and automatically determine the most suitable scales for matching. The importance of different spatial locations for extracting discriminative features is also learned explicitly. Experiments are carried out to demonstrate that the proposed model outperforms the state-of-the art on a number of benchmarks.

We present Deeply Supervised Object Detector (DSOD), a framework that can learn object detectors from scratch. State-of-the-art object objectors rely heavily on the off the-shelf networks pre-trained on large-scale classification datasets like Image Net, which incurs learning bias due to the difference on both the loss functions and the category distributions between classification and detection tasks. Model fine-tuning for the detection task could alleviate this bias to some extent but not fundamentally. Besides, transferring pre-trained models from classification to detection between discrepant domains is even more difficult (e.g. RGB to depth images). A better solution to tackle these two critical problems is to train object detectors from scratch, which motivates our proposed DSOD. Previous efforts in this direction mostly failed due to much more complicated loss functions and limited training data in object detection. In DSOD, we contribute a set of design principles for training object detectors from scratch. One of the key findings is that deep supervision, enabled by dense layer-wise connections, plays a critical role in learning a good detector. Combining with several other principles, we develop DSOD following the single-shot detection (SSD) framework. Experiments on PASCAL VOC 2007, 2012 and MS COCO datasets demonstrate that DSOD can achieve better results than the state-of-the-art solutions with much more compact models. For instance, DSOD outperforms SSD on all three benchmarks with real-time detection speed, while requires only 1/2 parameters to SSD and 1/10 parameters to Faster RCNN.

The aim of fine-grained recognition is to identify sub-ordinate categories in images like different species of birds. Existing works have confirmed that, in order to capture the subtle differences across the categories, automatic localization of objects and parts is critical. Most approaches for object and part localization rehed on the bottom-up pipeline, where thousands of region proposals are generated and then filtered by pre-trained object/part models. This is computationally expensive and not scalable once the number of objects/parts becomes large. In this paper, we propose a nonparametric data-driven method for object and part localization. Given an unlabeled test image, our approach transfers annotations from a few similar images retrieved in the training set. In particular, we propose an iterative transfer strategy that gradually refine the predicted bounding boxes. Based on the located objects and parts, deep convolutional features are extracted for recognition. We evaluate our approach on the widely-used CUB200-2011 dataset and a new and large dataset called Birdsnap. On both datasets, we achieve better results than many state-of-the-art approaches, including a few using oracle (manually annotated) bounding boxes in the test images.

Emotional content is a key ingredient in user-generated videos. However, due to the emotion sparsely expressed in the user-generated video, it is very difficult to analayze emotions in videos. In this paper, we propose a new architecture--Frame-Transformer Emotion Classification Network (FT-EC-net) to solve three highly correlated emotion analysis tasks: emotion recognition, emotion attribution and emotion-oriented summarization. We also contribute a new dataset for emotion attribution task by annotating the ground-truth labels of attribution segments. A comprehensive set of experiments on two datasets demonstrate the effectiveness of our framework.

Deep neural networks have significantly improved the performance of face recognition and facial attribute prediction, which however are still very challenging on the million scale dataset, i.e. MegaFace. In this paper, we for the first time, advocate a multi-task deep neural network for jointly learning face recognition and facial attribute prediction tasks. Extensive experimental evaluation clearly demonstrates the effectiveness of our architecture. Remarkably, on the largest face recognition benchmark -- MegaFace dataset, our networks can achieve the Rank-1 identication accuracy of 77.74% and face verication accuracy 79.24% TAR at 10-6 FAR, which are the best performance on the small protocol among all the publicly released methods.

The ubiquity of online fashion shopping demands effective recommendation services for customers. In this paper, we study two types of fashion recommendation: (i) suggesting an item that matches existing components in a set to form a stylish outfit (a collection of fashion items), and (ii) generating an outfit with multimodal (images/text) specifications from a user. To this end, we propose to jointly learn a visual-semantic embedding and the compatibility relationships among fashion items in an end-to-end fashion. More specifically, we consider a fashion outfit to be a sequence (usually from top to bottom and then accessories) and each item in the outfit as a time step. Given the fashion items in an outfit, we train a bidirectional LSTM (Bi-LSTM) model to sequentially predict the next item conditioned on previous ones to learn their compatibility relationships. Further, we learn a visual-semantic space by regressing image features to their semantic representations aiming to inject attribute and category information as a regularization for training the LSTM. The trained network can not only perform the aforementioned recommendations effectively but also predict the compatibility of a given outfit. We conduct extensive experiments on our newly collected Polyvore dataset, and the results provide strong qualitative and quantitative evidence that our framework outperforms alternative methods.

Multi-task learning aims to boost the performance of multiple prediction tasks by appropriately sharing relevant information among them. However, it always suffers from the negative transfer problem. And due to the diverse learning difficulties and convergence rates of different tasks, jointly optimizing multiple tasks is very challenging. To solve these problems, we present a weighted multi-task deep convolutional neural network for person attribute analysis. A novel validation loss trend algorithm is, for the first time proposed to dynamically and adaptively update the weight for learning each task in the training process. Extensive experiments on CelebA, Market-1501 attribute and Duke attribute datasets clearly show that state-of-the-art performance is obtained; and this validates the effectiveness of our proposed framework.

In this paper, a deep end-to-end network for sketch recognition, named Deep Visual-Sequential Fusion model (DVSF) is proposed to model the visual and sequential patterns of the strokes. To capture the intermediate states of sketches, a three-way representation learner is first utilized to extract the visual features. These deep features are simultaneously fed into the visual and sequential networks to capture spatial and temporal properties, respectively. More specifically, visual networks are novelly proposed to learn the stroke patterns by stacking the Residual Fully-Connected (R-FC) layers, which integrate ReLU and Tanh activation functions to achieve the sparsity and generalization ability. To learn the patterns of stroke order, sequential networks are constructed by Residual Long Short-Term Memory (R-LSTM) units, which optimize the network architecture by skip connection. Finally, the visual and sequential representations of the sketches are seamlessly integrated with a fusion layer to obtain the final results. Experiments conducted on the benchmark sketch dataset TU-Berlin demonstrate the effectiveness of the proposed method, which outperforms the state-of-the-art approaches.

This paper presents a brief summary of the first workshop on Visual Analysis for Smart and Connected Communities (VSCC'2017), which is held in conjunction with the ACM Conference on Multimedia, 2017. VSCC'2017 is the first workshop on visual analysis for smart and connected communities, and is aimed at the creation of a multi-discipline community with the concentration on visual data analysis. The topics addressed in VSCC'2017 cover all aspects of smart communities, including safety, security, retrieval, transportation, information technologies, Internet of Things, etc. The focus is on the development of fundamental theories, algorithms, models towards the parsing of visual data generated by various digital devices (e.g., smart phones, vehicles) in smart communities.

Recognizing visual contents in unconstrained videos has become a very important problem for many applications, such as Web video search and recommendation, smart advertising, robotics, etc. This workshop and challenge aims at exploring new challenges and approaches for large-scale video classification with large number of classes from open source videos in a realistic setting, based upon an extension of Fudan-Columbia Video Dataset (FCVID). This newly collected dataset contains over 8000 hours of video data from YouTube and Flicker, annotated into 500 categories. We hope this dataset can stimulate innovative research on this challenging and important problem.

Modeling hairstyles for classification, synthesis and image editing has many practical applications. However, existing hairstyle datasets, such as the Beauty e-Expert dataset, are too small for developing and evaluating computer vision models, especially the recent deep generative models such as generative adversarial network (GAN). In this paper, we contribute a new large-scale hairstyle dataset called Hairstyle30k, which is composed of 30k images containing 64 different types of hairstyles. To enable automated generating and modifying hairstyles in images, we also propose a novel GAN model termed Hairstyle GAN (H-GAN) which can be learned efficiently. Extensive experiments on the new dataset as well as existing benchmark datasets demonstrate the effectiveness of proposed H-GAN model

The spatial relationship among objects provide rich clues to object contexts for visual recognition. In this paper, we propose to learn Semantic Feature Map (SFM) by deep neural networks to model the spatial object contexts for better understanding of image and video contents. Specifically, we first extract high-level semantic object features on input image with convolutional neural networks for every object proposals, and organize them to the designed SFM so that spatial information among objects are preserved. To fully exploit the spatial relationship among objects, we employ either Fully Convolutional Networks (FCN) or Long-Short Term Memory (LSTM) on top of SFM for final recognition. For better training, we also introduce a multi-task learning framework to train the model in an end-to-end manner. It is composed of an overall image classification loss as well as a grid labeling loss, which predicts the objects label at each SFM grid. Extensive experiments are conducted to verify the effectiveness of the proposed approach. For image classification, very promising results are obtained on Pascal VOC 2007/2012 and MS-COCO benchmarks. We also directly transfer the SFM learned on image domain to the video classification task. The results on CCV benchmark demonstrate the robustness and generalization capability of the proposed approach.

Large-scale action recognition and video categorization are important problems in computer vision. To address these problems, we propose a novel object-and scene-based semantic fusion network and representation. Our semantic fusion network combines three streams of information using a three-layer neural network: (i) frame-based low-level CNN features, (ii) object features from a state-of-the-art large-scale CNN object-detector trained to recognize 20K classes, and (iii) scene features from a state-of-the-art CNN scene-detector trained to recognize 205 scenes. The trained network achieves improvements in supervised activity and video categorization in two complex large-scale datasets - ActivityNet and FCVID, respectively. Further, by examining and back propagating information through the fusion network, semantic relationships (correlations) between video classes and objects/scenes can be discovered. These video class-object/video class-scene relationships can in turn be used as semantic representation for the video classes themselves. We illustrate effectiveness of this semantic representation through experiments on zero-shot action/video classification and clustering.

We consider a wide spectrum of regularized stochastic minimization problems, where the regularization term is composite with a linear function. Examples of this formulation include graph-guided regularized minimization, generalized Lasso and a class of ℓ1 regularized problems. The computational challenge is that the closed-form solution of the proximal mapping associated with the regularization term is not available due to the imposed linear composition. Fortunately, the structure of the regularization term allows us to reformulate it as a new convex-concave saddle point problem which can be solved using the Primal-Dual Hybrid Gradient (PDHG) approach. However, this approach may be inefficient in realistic applications as computing the full gradient of the expected objective function could be very expensive when the number of input data samples is considerably large. To address this issue, we propose a Stochastic PDHG (SPDHG) algorithm with either uniformly or non-uniformly averaged iterates. Through uniformly averaged iterates, the SPDHG algorithm converges in expectation withrate for general convex objectives and O(log (t)/t) rate for strongly convex objectives, respectively. While with non-uniformly averaged iterates, the SPDHG algorithm is expected to converge with O(1/t) rate for strongly convex objectives. Numerical experiments on different genres of datasets demonstrate that our proposed algorithm outperforms other competing algorithms.

This paper focuses on a practically very important problem of matching a real-world product photo to exactly the same item(s) in online shopping sites. The task is extremely challenging because the user photos (i.e., the queries in this scenario) are often captured in uncontrolled environments, while the product images in online shops are mostly taken by professionals with clean backgrounds and perfect lighting conditions. To tackle the problem, we study deep network architectures and training schemes, with the goal of learning a robust deep feature representation that is able to bridge the domain gap between the user photos and the online product images. Our contributions are two-fold. First, we propose an alternative of the popular contrastive loss used in siamese deep networks, namely robust contrastive loss, where we "relax" the penalty on positive pairs to alleviate over-fitting. Second, a multi-task fine-tuning approach is introduced to learn a better feature representation, which not only incorporates knowledge from the provided training photo pairs, but also explores additional information from the large ImageNet dataset to regularize the fine-tuning procedure. Experiments on two challenging real-world datasets demonstrate that both the robust contrastive loss and the multi-task fine-tuning approach are effective, leading to very promising results with a time cost suitable for real-time retrieval.

Despite growing research interest, emotion understanding for user-generated videos remains a challenging problem. Major obstacles include the diversity and complexity of video content, as well as the sparsity of expressed emotions. For the first time, we systematically study large-scale video emotion recognition by transferring deep feature encodings. In addition to the traditional, supervised recognition, we study the problem of zero-shot emotion recognition, where emotions in the test set are unseen during training. To cope with this task, we utilize knowledge transferred from auxiliary image and text corpora. A novel auxiliary Image Transfer Encoding (ITE) process is proposed to efficiently encode and generate video representation. We also thoroughly investigate different configurations of convolutional neural networks. Comprehensive experiments on multiple datasets demonstrate the effectiveness of our framework.

Human emotions demonstrate high correlations with certain events that consist of the interaction of objects, and are usually constrained by particular scenes. In this paper, we exploit these abundant context clues for video emotion recognition. We first compute event, object and scene scores with state-of-the-art detectors based on deep neural networks. The extracted high-level features serve as effective contextual information demonstrating what is occurring in the video, and are further integrated in a context fusion network to generate a unified representation aiming to bridge the affective gap. The contribution of this paper is three-fold: (a) we are the first to incorporate event as context for emotion recognition, and we demonstrate its superiority for emotion understanding, (b) we utilize a context fusion network to exploit a comprehensive set of high-level semantics features as contextual clues and (c) the proposed framework can realize recognition in real-time and achieves state-of-the-art performance on two challenging emotion recognition benchmarks, 50.6 on VideoEmotion and 51.8 on Ekman.

This paper studies the problem of learning to hash, which is essentially a mixed integer optimization problem, containing both the binary hash code output and the (continuous) parameters forming the hash functions. Different from existing relaxation methods in hashing, which have no theoretical guarantees for the error bound of the relaxations, we propose binary optimized hashing (BOH), in which we prove that if the loss function is Lipschitz continuous, the binary optimization problem can be relaxed to a bound-constrained continuous optimization problem. Then we introduce a surrogate objective function, which only depends on unbinarized hash functions and does not need the slack variables transforming unbinarized hash functions to discrete functions, to approximate the relaxed objective function. We show that the approximation error is bounded and the bound is small when the problem is optimized. We apply the proposed approach to learn hash codes from either handcraft feature inputs or raw image inputs. Extensive experiments are carried out on three benchmarks, demonstrating that our approach outperforms state-of-the-arts with a significant margin on search accuracies.

Temporal dynamics play an important role for video classification. In this paper, we propose to leverage high-level semantic features to open the "black box" of the state-of-the-art temporal model, Long Short Term Memory (LSTM), with an aim to understand what is learned. More specifically, we first extract object features from a state-of-the-art CNN model that is trained to recognize 20K objects. Then we leverage LSTM with the extracted features as inputs to capture the temporal dynamics in videos. In combination with spatial and motion information, we achieve improvements for supervised video categorization. Furthermore, by masking the inputs, we demonstrate what is learned by LSTM, namely (i) which objects are crucial for recognizing a class-of-interest; (ii) how the LSTM model could assist the temporal localization of these detected objects.

This paper studies deep network architectures to address the problem of video classification. A multi-stream framework is proposed to fully utilize the rich multimodal information in videos. Specifically, we first train three Convolutional Neural Networks to model spatial, short-term motion and audio clues respectively. Long Short Term Memory networks are then adopted to explore long-term temporal dynamics. With the outputs of the individual streams on multiple classes, we propose to mine class relationships hidden in the data from the trained models. The automatically discovered relationships are then leveraged in the multi-stream multi-class fusion process as a prior, indicating which and how much information is needed from the remaining classes, to adaptively determine the optimal fusion weights for generating the final scores of each class. Our contributions are two-fold. First, the multi-stream framework is able to exploit multimodal features that are more comprehensive than those previously attempted. Second, our proposed fusion method not only learns the best weights of the multiple network streams for each class, but also takes class relationship into account, which is known as a helpful clue in multi-class visual classification tasks. Our framework produces significantly better results than the state of the arts on two popular benchmarks, 92.2% on UCF-101 (without using audio) and 84.9% on Columbia Consumer Videos.

The goal of partial video copy detection is to find one or more segments of a query video which have (transformed) copies in a large dataset. Previous related research in this field used either small-scale datasets or large datasets with simulated partial copies by imposing several pre-defined transformations (e.g., photometric changes) due to the extremely time-consuming annotation of real copies. It is still unknown how well the techniques developed on simulated datasets perform on real copies, which are much more challenging and too complex to be simulated. In this paper, we introduce a large-scale video copy database (VCDB) with over 100,000 videos, and more than 9,000 copy pairs found by manual annotation. A state-of-the-art system of video copy detection is evaluated on VCDB to show the limitations of existing techniques. We also evaluate deep learning features learned by two neural networks: one is independently trained on a different dataset and the other is tailored to deal with the copy detection task. Our evaluation suggests that all the existing techniques, including the deep learning features, are far from satisfactory in detecting complex real copies.

Existing video search engines return a ranked list of videos for each user query, which is not convenient for browsing the results of query topics that have multiple facets, such as the “early life,” “personal life,” and “presidency” of a query “Barack Obama.” Organizing video search results into semantically structured hierarchies with nodes covering different topic facets can significantly improve the browsing efficiency for such queries. In this paper, we introduce a hierarchical visualization approach for video search result browsing, which can help users quickly understand the multiple facets of a query topic in a very well-organized manner. Given a query, our approach starts from the hierarchy of its textual descriptions normally available on Wikipedia and then adjusts the hierarchical structure by analyzing the video information to reflect the topic structure of the search result. After that, a simple optimization problem is formulated to perform the video-to-node association considering three important criteria. Furthermore, additional topic facets are mined to complement the contents of the existing semantic hierarchies. A large YouTube video dataset is constructed to evaluate our approach both quantitatively and qualitatively. A demo system is also developed for users to interact with the proposed browsing approach.

Videos contain very rich semantic information. Traditional hand-crafted features are known to be inadequate in analyzing complex video semantics. Inspired by the huge success of the deep learning methods in analyzing image, audio and text data, significant efforts are recently being devoted to the design of deep nets for video analytics. Among the many practical needs, classifying videos (or video clips) based on their major semantic categories (e.g.,"skiing") is useful in many applications. In this paper, we conduct an in-depth study to investigate important implementation options that may affect the performance of deep nets on video classification. Our evaluations are conducted on top of a recent two-stream convolutional neural network (CNN) pipeline, which uses both static frames and motion optical flows, and has demonstrated competitive performance against the state-of-the-art methods. In order to gain insights and to arrive at a practical guideline, many important options are studied, including network architectures, model fusion, learning parameters and the final prediction methods. Based on the evaluations, very competitive results are attained on two popular video classification benchmarks. We hope that the discussions and conclusions from this work can help researchers in related fields to quickly set up a good basis for further investigations along this very promising direction.

No abstract available.

Classifying videos according to content semantics is an important problem with a wide range of applications. In this paper, we propose a hybrid deep learning framework for video classification, which is able to model static spatial information, short-term motion, as well as long-term temporal clues in the videos. Specifically, the spatial and the short-term motion features are extracted separately by two Convolutional Neural Networks (CNN). These two types of CNN-based features are then combined in a regularized feature fusion network for classification, which is able to learn and utilize feature relationships for improved performance. In addition, Long Short Term Memory (LSTM) networks are applied on top of the two features to further model longer-term temporal clues. The main contribution of this work is the hybrid learning framework that can model several important aspects of the video data. We also show that (1) combining the spatial and the short-term motion features in the regularized fusion network is better than direct classification and fusion using the CNN with a softmax layer, and (2) the sequence-based LSTM is highly complementary to the traditional classification strategy without considering the temporal frame orders. Extensive experiments are conducted on two popular and challenging benchmarks, the UCF-101 Human Actions and the Columbia Consumer Videos (CCV). On both benchmarks, our framework achieves very competitive performance: 91.3% on the UCF-101 and 83.5% on the CCV.

The last decade has witnessed a dramatic growth of multimedia content and applications, which in turn requires an increasing demand of computational resources. Meanwhile, the high-performance computing world undergoes a trend toward heterogeneity. However, it is never easy to develop domain-specific applications on heterogeneous systems while maximizing the system efficiency. In this paper, a novel framework, namely, cloud-based heterogeneous computing framework (CHCF), is proposed with a set of tools and techniques for compilation, optimization, and execution of multimedia mining applications on heterogeneous systems. With the aid of the compiler and the utility library provided by CHCF, users are able to develop multimedia mining applications rapidly and efficiently. The proposed framework employs a number of techniques, including adaptive data partitioning, knowledge-based hierarchical scheduling, and performance estimation, to achieve high computing performance. As one of the most important multimedia mining applications, large-scale image retrieval is investigated based on the proposed CHCF. The scalability, computing performance, and programmability of CHCF are studied for large-scale image retrieval by case studies and experimental evaluations. The experimental results demonstrate that CHCF can achieve good scalability and significant computing performance improvements for image retrieval.

Human action recognition in unconstrained videos is a challenging problem with many applications. Most state-of-the-art approaches adopted the well-known bag-of-features representations, generated based on isolated local patches or patch trajectories, where motion patterns, such as object-object and object-background relationships are mostly discarded. In this paper, we propose a simple representation aiming at modeling these motion relationships. We adopt global and local reference points to explicitly characterize motion information, so that the final representation is more robust to camera movements, which widely exist in unconstrained videos. Our approach operates on the top of visual codewords generated on dense local patch trajectories, and therefore, does not require foreground-background separation, which is normally a critical and difficult step in modeling object relationships. Through an extensive set of experimental evaluations, we show that the proposed representation produces a very competitive performance on several challenging benchmark data sets. Further combining it with the standard bag-of-features or Fisher vector representations can lead to substantial improvements.

Techniques for recognizing high-level events in consumer videos on the Internet have many applications. Systems that produced state-of-the-art recognition performance usually contain modules requiring extensive computation, such as the extraction of the temporal motion trajectories, which cannot be deployed on large-scale datasets. In this paper, we provide a comprehensive study on efficient methods in this area and identify technical options for super fast event recognition in Internet videos. We start from analyzing a multimodal baseline that has produced good performance on popular benchmarks, by systematically evaluating each component in terms of both computational cost and contribution to recognition accuracy. After that, we identify alternative features, classifiers, and fusion strategies that can all be efficiently computed. In addition, we also provide a study on the following interesting question: for event recognition in Internet videos, what is the minimum number of visual and audio frames needed to obtain a comparable accuracy to that of using all the frames? Results on two rigorously designed datasets indicate that similar results can be maintained by using only a small portion of the visual frames. We also find that, different from the visual frames, the soundtracks contain little redundant information and thus sampling is always harmful. Integrating all the findings, our suggested recognition system is 2,350-fold faster than a baseline approach with even higher recognition accuracies. It recognizes 20 classes on a 120-second video sequence in just 1.78 seconds, using a regular desktop computer.

User-generated video collections are expanding rapidly in recent years, and systems for automatic analysis of these collections are in high demands. While extensive research efforts have been devoted to recognizing semantics like "birthday party" and "skiing", little attempts have been made to understand the emotions carried by the videos, e.g., "joy" and "sadness". In this paper, we propose a comprehensive computational framework for predicting emotions in user-generated videos. We first introduce a rigorously designed dataset collected from popular video-sharing websites with manual annotations, which can serve as a valuable benchmark for future research. A large set of features are extracted from this dataset, ranging from popular low-level visual descriptors, audio features, to high-level semantic attributes. Results of a comprehensive set of experiments indicate that combining multiple types of features---such as the joint use of the audio and visual clues---is important, and attribute features such as those containing sentiment-level semantics are very effective.

Benefiting from its openness, collaboration and real-time features, Micro blog has become one of the most important news communication media in modern society. However, it is also filled with fake news. Without verification, such information could spread promptly through social network and result in serious consequences. To evaluate news credibility on Micro blog, we propose a hierarchical propagation model. We detect sub-events within a news event to describe its detailed aspects. Thus, for a news event, a three-layer credibility network consisting of event, sub-events and messages can represent it from different scale and reveal vital information for credibility evaluation. After linking these entities with their semantic and social associations, the credibility value of each entity is propagated on this network to achieve the final evaluation result. By formulating this propagation process as a graph optimization problem, we provide a globally optimal solution with an iterative algorithm. Experiments conducted on two real-world datasets show that the proposed model boosts the accuracy by more than 6% and the F-score by more than 16% over a baseline method.

User-generated contents play an important role in the Internet video-sharing activities. Techniques for summarizing the user-generated videos (UGVs) into short representative clips are useful in many applications. This paper introduces an approach for UGV summarization based on semantic recognition. Different from other types of videos like movies or broadcasting news, where the semantic contents may vary greatly across different shots, most UGVs have only a single long shot with relatively consistent high-level semantics. Therefore, a few semantically representative segments are generally sufficient for a UGV summary, which can be selected based on the distribution of semantic recognition scores. In addition, due to the poor shooting quality of many UGVs, factors such as camera shaking and lighting condition are also considered to achieve more pleasant summaries. Experiments on over 100 UGVs with both subjective and objective evaluations show that our approach clearly outperforms several alternative methods and is highly efficient. Using a regular laptop, it can produce a summary for a 2-minute video in just 10 seconds.

Organizing video search results into semantically structured hierarchies can greatly improve the efficiency of browsing complex query topics. Traditional hierarchical clustering techniques are inadequate since they lack the ability to generate semantically interpretable structures. In this paper, we introduce an approach to organize video search results to an adapted semantic hierarchy. As many hot search topics such as celebrities and famous cities have Wikipedia pages where hierarchical topic structures are available, we start from the Wikipedia hierarchies and adjust the structures according to the characteristics of the returned videos from a search engine. Ordinary clustering based on textual information of the videos is performed to discover the hidden topic structures in the video search results, which are used to adapt the hierarchy extracted from Wikipedia. After that, a simple optimization problem is formulated to assign the videos to each node of the hierarchy considering three important criteria. Experiments conducted on a Youtube video dataset verify the effectiveness of our approach.

Videos contain very rich semantics and are intrinsically multimodal. In this paper, we study the challenging task of classifying videos according to their high-level semantics such as human actions or complex events. Although extensive efforts have been paid to study this problem, most existing works combined multiple features using simple fusion strategies and neglected the exploration of inter-class semantic relationships. In this paper, we propose a novel unified framework that jointly learns feature relationships and exploits the class relationships for improved video classification performance. Specifically, these two types of relationships are learned and utilized by rigorously imposing regularizations in a deep neural network (DNN). Such a regularized DNN can be efficiently launched using a GPU implementation with an affordable training cost. Through arming the DNN with better capability of exploring both the inter-feature and the inter-class relationships, the proposed regularized DNN is more suitable for identifying video semantics. With extensive experimental evaluations, we demonstrate that the proposed framework exhibits superior performance over several state-of-the-art approaches. On the well-known Hollywood2 and Columbia Consumer Video benchmarks, we obtain to-date the best reported results: 65.7% and 70.6% respectively in terms of mean average precision.

Semantic attributes have been recognized as a more spontaneous manner to describe and annotate image content. It is widely accepted that image annotation using semantic attributes is a significant improvement to the traditional binary or multiclass annotation due to its naturally continuous and relative properties. Though useful, existing approaches rely on an abundant supervision and high-quality training data, which limit their applicability. Two standard methods to overcome small amounts of guidance and low-quality training data are transfer and active learning. In the context of relative attributes, this would entail learning multiple relative attributes simultaneously and actively querying a human for additional information. This paper addresses the two main limitations in existing work: 1) it actively adds humans to the learning loop so that minimal additional guidance can be given and 2) it learns multiple relative attributes simultaneously and thereby leverages dependence amongst them. In this paper, we formulate a joint active learning to rank framework with pairwise supervision to achieve these two aims, which also has other benefits such as the ability to be kernelized. The proposed framework optimizes over a set of ranking functions (measuring the strength of the presence of attributes) simultaneously and dependently on each other. The proposed pairwise queries take the form of which one of these two pictures is more natural? These queries can be easily answered by humans. Extensive empirical study on real image data sets shows that our proposed method, compared with several state-of-the-art methods, achieves superior retrieval performance while requires significantly less human inputs.

The four papers in this special section cover several hot and emerging topics in socio-mobile media analysis and retrieval.

Event detection plays an essential role in video content analysis. In this paper, we present our approach based on motion relativity and feature selection for video event detection. First, we propose a new motion feature, namely Expanded Relative Motion Histogram of Bag-of-Visual-Words (ERMH-BoW) to employ motion relativity for event detection. In ERMH-BoW, by representing what aspect of an event with Bag-of-Visual-Words (BoW), we construct relative motion histograms between different visual words to depict the objects' activities or how aspect of the event. ERMH-BoW thus integrates both what and how aspects for a complete event description. Meanwhile, we show that by employing motion relativity, ERMH-BoW is invariant to the varying camera movement and able to honestly describe the object activities in an event. Furthermore, compared with other motion features, ERMH-BoW encodes not only the motion of objects, but also the interactions between different objects/scenes. Second, to address the high-dimensionality problem of the ERMH-BoW feature, we further propose an approach based on information gain and informativeness weighting to select a cleaner and more discriminative set of features. Our experiments carried out on several challenging datasets provided by TRECVID for the MED (Multimedia Event Detection) task demonstrate that our proposed approach outperforms the state-of-the-art approaches for video event detection.

Organizing video search results in a list view is widely adopted by current commercial search engines, which cannot support efficient browsing for complex search topics that have multiple semantic facets. In this article, we propose to organize video search results in a highly structured way. Specifically, videos are placed on a semantic hierarchy that accurately organizes various facets of a given search topic. To pick the most suitable videos for each node of the hierarchy, we define and utilize three important criteria: relevance, uniqueness, and diversity. Extensive evaluations on a large YouTube video dataset demonstrate the effectiveness of our approach.

The amount of videos available on the Web is growing explosively. While some videos are very interesting and receive high rating from viewers, many of them are less interesting or even boring. This paper conducts a pilot study on the understanding of human perception of video interestingness, and demonstrates a simple computational method to identify more interesting videos. To this end we first construct two datasets of Flickr and YouTube videos respectively. Human judgements of interestingness are collected and used as the ground-truth for training computational models. We evaluate several off-the-shelf visual and audio features that are potentially useful for predicting interestingness on both datasets. Results indicate that audio and visual features are equally important and the combination of both modalities shows very promising results.

Hashing techniques have been intensively investigated in the design of highly efficient search engines for large-scale computer vision applications. Compared with prior approximate nearest neighbor search approaches like tree-based indexing, hashing-based search schemes have prominent advantages in terms of both storage and computational efficiencies. Moreover, the procedure of devising hash functions can be easily incorporated into sophisticated machine learning tools, leading to data-dependent and task-specific compact hash codes. Therefore, a number of learning paradigms, ranging from unsupervised to supervised, have been applied to compose appropriate hash functions. However, most of the existing hash function learning methods either treat hash function design as a classification problem or generate binary codes to satisfy pair wise supervision, and have not yet directly optimized the search accuracy. In this paper, we propose to leverage list wise supervision into a principled hash function learning framework. In particular, the ranking information is represented by a set of rank triplets that can be used to assess the quality of ranking. Simple linear projection-based hash functions are solved efficiently through maximizing the ranking quality over the training data. We carry out experiments on large image datasets with size up to one million and compare with the state-of-the-art hashing techniques. The extensive results corroborate that our learned hash codes via list wise supervision can provide superior search accuracy without incurring heavy computational overhead.

The aesthetics of videos can be used as a useful clue to improve user satisfaction in many applications such as search and recommendation. In this paper, we demonstrate a computational approach to automatically evaluate the aesthetics of videos, with particular emphasis on identifying beautiful scenes. Using a standard classification pipeline, we analyze the effectiveness of a comprehensive set of features, ranging from low-level visual features, mid-level semantic attributes, to style descriptors. In addition, since there is limited public training data with manual labels of video aesthetics, we explore freely available resources with a simple assumption that people tend to share more aesthetically appealing works than unappealing ones. Specifically, we use images from DPChallenge and videos from Flickr as positive training data and the Dutch documentary videos as negative data, where the latter contain mostly old materials of low visual quality. Our extensive evaluations show that combining multiple features is helpful, and very promising results can be obtained using the noisy but annotation-free training data. On the NHK Multimedia Challenge dataset, we attain a Spearman's rank correlation coefficient of 0.41.

The state-of-the-art partial-duplicate image search systems reply heavily on the match of local features like SIFT. Independently matching local features across two images ignores the overall geometry structure and therefore may incur many false matches. To reduce such matches, several geometry verification methods have been proposed. This paper introduces a new geometry verification method named as Strong Geometry Consistency (SGC), which uses the orientation, scale and location information of the local feature points to accurately and quickly remove the false matches. We also propose a simple scale weighting (SW) strategy, which gives feature points with larger scales greater weights, based on the intuition that a larger-scale feature point tends to be more robust for image search as it occupies a larger area of an image. Extensive experiments performed on three popular datasets show that SGC significantly outperforms state-of-the-art geometry verification methods, and SW can further boost the performance with marginal additional computation.

Scalable image search based on visual similarity has been an active topic of research in recent years. State-of-the-art solutions often use hashing methods to embed high-dimensional image features into Hamming space, where search can be performed in real-time based on Hamming distance of compact hash codes. Unlike traditional metrics (e.g., Euclidean) that offer continuous distances, the Hamming distances are discrete integer values. As a consequence, there are often a large number of images sharing equal Hamming distances to a query, which largely hurts search results where fine-grained ranking is very important. This paper introduces an approach that enables query-adaptive ranking of the returned images with equal Hamming distances to the queries. This is achieved by firstly offline learning bitwise weights of the hash codes for a diverse set of predefined semantic concept classes. We formulate the weight learning process as a quadratic programming problem that minimizes intra-class distance while preserving inter-class relationship captured by original raw image features. Query-adaptive weights are then computed online by evaluating the proximity between a query and the semantic concept classes. With the query-adaptive bitwise weights, returned images can be easily ordered by weighted Hamming distance at a finer-grained hash code level rather than the original Hamming distance level. Experiments on a Flickr image dataset show clear improvements from our proposed approach.

Recent years have witnessed the growing popularity of hashing in large-scale vision problems. It has been shown that the hashing quality could be boosted by leveraging supervised information into hash function learning. However, the existing supervised methods either lack adequate performance or often incur cumbersome model training. In this paper, we propose a novel kernel-based supervised hashing model which requires a limited amount of supervised information, i.e., similar and dissimilar data pairs, and a feasible training cost in achieving high quality hashing. The idea is to map the data to compact binary codes whose Hamming distances are minimized on similar pairs and simultaneously maximized on dissimilar pairs. Our approach is distinct from prior works by utilizing the equivalence between optimizing the code inner products and the Hamming distances. This enables us to sequentially and efficiently train the hash functions one bit at a time, yielding very short yet discriminative codes. We carry out extensive experiments on two image benchmarks with up to one million samples, demonstrating that our approach significantly outperforms the state-of-the-arts in searching both metric distance neighbors and semantically similar neighbors, with accuracy gains ranging from 13% to 46%.

Event recognition in unconstrained Internet videos has great potential in many applications. State-of-the-art systems usually include modules that need extensive computation, such as the extraction of spatial-temporal interest points, which poses a big challenge for large-scale video processing. This paper presents SUPER, a Speeded UP Event Recognition framework for efficient Internet video analysis. We take a multimodal baseline that has produced strong performance on popular benchmarks, and systematically evaluate each component in terms of both computational cost and contribution to recognition accuracy. We show that, by choosing suitable features, classifiers, and fusion strategies, recognition speed can be greatly improved with minor performance degradation. In addition, we also evaluate how many visual and audio frames are needed for event recognition in Internet videos, a question left unanswered in the literature. Results on a rigorously designed dataset indicate that similar recognition accuracy can be attained using only 14 frames per video on average. We also observe that, different from the visual channel, the soundtracks contains little redundant information for video event recognition. Integrating all the findings, our suggested SUPER framework is 220-fold faster than the baseline approach with merely 3.8% drop in recognition accuracy. It classifies an 80-second video sequence using models of 20 classes in just 4.56 seconds.

Joint audio-visual patterns often exist in videos and provide strong multi-modal cues for detecting multimedia events. However, conventional methods generally fuse the visual and audio information only at a superficial level, without adequately exploring deep intrinsic joint patterns. In this paper, we propose a joint audio-visual bi-modal representation, called bi-modal words. We first build a bipartite graph to model relation across the quantized words extracted from the visual and audio modalities. Partitioning over the bipartite graph is then applied to construct the bi-modal words that reveal the joint patterns across modalities. Finally, different pooling strategies are employed to re-quantize the visual and audio words into the bi-modal words and form bi-modal Bag-of-Words representations that are fed to subsequent multimedia event classifiers. We experimentally show that the proposed multi-modal feature achieves statistically significant performance gains over methods using individual visual and audio features alone and alternative multi-modal fusion methods. Moreover, we found that average pooling is the most suitable strategy for bi-modal feature generation.

Techniques for recognizing complex events in diverse Internet videos are important in many applications. State-of-the-art video event recognition approaches normally involve modules that demand extensive computation, which prevents their application to large scale problems. In this demonstration, we present a fast video event recognition system, which requires just a few seconds to process a general YouTube video with a few minutes of duration. The development of this system is grounded on several important findings from a large set of empirical studies, where we systematically evaluated many technical options for each critical module of a present-day video event recognition framework. Pooling the insights gained from this study leads to a speeded-up event recognition system that is 220-times faster than a decent baseline while still has a high degree of recognition accuracy. We also demonstrate the technical feasibility of using event recognition results as the sole clue for video search, where the similarity of videos is determined based on the consistency of the event recognition confidence scores. We showcase this capability using an Internet video dataset containing about 10 thousands of YouTube videos. Very promising results were observed.

Exploring context information for visual recognition has recently received significant research attention. This paper proposes a novel and highly efficient approach, which is named semantic diffusion, to utilize semantic context for large-scale image and video annotation. Starting from the initial annotation of a large number of semantic concepts (categories), obtained by either machine learning or manual tagging, the proposed approach refines the results using a graph diffusion technique, which recovers the consistency and smoothness of the annotations over a semantic graph. Different from the existing graph-based learning methods that model relations among data samples, the semantic graph captures context by treating the concepts as nodes and the concept affinities as the weights of edges. In particular, our approach is capable of simultaneously improving annotation accuracy and adapting the concept affinities to new test data. The adaptation provides a means to handle domain change between training and test data, which often occurs in practice. Extensive experiments are conducted to improve concept annotation results using Flickr images and TV program videos. Results show consistent and significant performance gain (10 +% on both image and video data sets). Source codes of the proposed algorithms are available online.

Sufficient training examples are essential for effective learning of semantic visual concepts. In practice, however, acquiring noise-free training examples has always been expensive. Recently the rapid popularization of social media websites, such as Flickr, has made it possible to collect training exemplars without human assistance. This paper proposes a novel and efficient approach to collect training samples from the noisily tagged Web images for visual concept learning, where we try to maximize two important criteria, relevancy and coverage, of the automatically generated training sets. For the former, a simple method named semantic field is introduced to handle the imprecise and incomplete image tags. Specifically, the relevancy of an image to a target concept is predicted by collectively analyzing the associated tag list of the image using two knowledge sources: WordNet corpus and statistics from Flickr.com. To boost the coverage or diversity of the training sets, we further propose an ontology-based hierarchical pooling method to collect samples not only based on the target concept alone, but also from ontologically neighboring concepts. Extensive experiments on three different datasets (NUS-WIDE, PASCAL VOC, and ImageNet) demonstrate the effectiveness of our proposed approach, producing competitive performance even when comparing with concept classifiers learned using expert- labeled training examples.

In this paper, we exploit a novel ranking mechanism that processes query samples with noisy labels, motivated by the practical application of web image search re-ranking where the originally highest ranked images are usually posed as pseudo queries for subsequent re-ranking. Availing ourselves of the low-frequency spectrum of a neighborhood graph built on the samples, we propose a graph-theoretical framework amenable to noise resistant ranking. The proposed framework consists of two components: spectral filtering and graph-based ranking. The former leverages sparse bases, progressively selected from a pool of smooth eigenvectors of the graph Laplacian, to reconstruct the noisy label vector associated with the query sample set and accordingly filter out the query samples with less authentic positive labels. The latter applies a canonical graph ranking algorithm with respect to the filtered query sample set. Quantitative image re-ranking experiments carried out on two public web image databases bear out that our re-ranking approach compares favorably with the state-of-the-arts and improves web image search engines by a large margin though we harvest the noisy queries from the top-ranked images returned by these search engines.

With the proliferation of images on the Web, fast search of visually similar images has attracted significant attention. State-of-the-art techniques often embed high-dimensional visual features into low-dimensional Hamming space, where search can be performed in real-time based on Hamming distance of compact binary codes. Unlike traditional metrics (e.g., Euclidean) of raw image features that produce continuous distance, the Hamming distances are discrete integer values. In practice, there are often a large number of images sharing equal Hamming distances to a query, resulting in a critical issue for image search where ranking is very important. In this paper, we propose a novel approach that facilitates query-adaptive ranking for the images with equal Hamming distance. We achieve this goal by firstly offline learning bit weights of the binary codes for a diverse set of predefined semantic concept classes. The weight learning process is formulated as a quadratic programming problem that minimizes intra-class distance while preserving interclass relationship in the original raw image feature space. Query-adaptive weights are then rapidly computed by evaluating the proximity between a query and the concept categories. With the adaptive bit weights, the returned images can be ordered by weighted Hamming distance at a finer-grained binary code level rather than at the original integer Hamming distance level. Experimental results on a Flickr image dataset show clear improvements from our query-adaptive ranking approach.

Recognizing visual content in unconstrained videos has become a very important problem for many applications. Existing corpora for video analysis lack scale and/or content diversity, and thus limited the needed progress in this critical area. In this paper, we describe and release a new database called CCV, containing 9,317 web videos over 20 semantic categories, including events like "baseball" and "parade", scenes like "beach", and objects like "cat". The database was collected with extra care to ensure relevance to consumer interest and originality of video content without post-editing. Such videos typically have very little textual annotation and thus can benefit from the development of automatic content analysis techniques.

Automatically generating compact textual descriptions of complex video contents has wide applications. With the recent advancements in automatic audio-visual content recognition, in this paper we explore the technical feasibility of the challenging issue of precisely recounting video contents. Based on cutting-edge automatic recognition techniques, we start from classifying a variety of visual and audio concepts in video contents. According to the classification results, we apply simple rule-based methods to generate textual descriptions of video contents. Results are evaluated by conducting carefully designed user studies. We find that the state-of-the-art visual and audio concept classification, although far from perfect, is able to provide very useful clues indicating what is happening in the videos. Most users involved in the evaluation confirmed the informativeness of our machine-generated descriptions.

A common obstacle in effective learning of visual concept classifiers is the scarcity of positive training examples due to expensive labeling cost. This paper explores the sampling of weakly tagged web images for concept learning without human assistance. In particular, ontology knowledge is incorporated for semantic pooling of positive examples from ontologically neighboring concepts. This effectively widens the coverage of the positive samples with visually more diversified content, which is important for learning a good concept classifier. We experiment with two learning strategies: aggregate and incremental. The former strategy re-trains a new classifier by combining existing and newly collected examples, while the latter updates the existing model using the new samples incrementally. Extensive experiments on NUS-WIDE and VOC 2010 datasets show very encouraging results, even when comparing with classifiers learnt using expert labeled training examples.

The use of context knowledge is critical for understanding human actions, which typically occur under particular scene settings with certain object interactions. For instance, driving car usually happens outdoors, and kissing involves two people moving toward each other. In this paper, we investigate the problem of context modeling for human action retrieval. We first identify ten simple object-level action atoms relevant to many human actions, e.g., people getting closer. With the action atoms and several background scene classes, we show that action retrieval can be improved through modeling action-scene-object dependency. An algorithm inspired by the popular semi-supervised learning paradigm is introduced for this purpose. One important contribution of this paper is to show that modeling the dependencies among actions, objects, and scenes can be efficiently achieved with very few examples. Such a solution has tremendous potential in practice as it is often expensive to acquire large sets of training data. Experiments were performed on the challenging Hollywood2 dataset containing 89 movies. The results validate the effectiveness of our approach, achieving a mean average precision of 26% with just ten examples per action.

As it is true for human perception that we gather information from different sources in natural and multi-modality forms, learning from multi-modalities has become an effective scheme for various information retrieval problems. In this paper, we propose a novel multi-modality fusion approach for video search, where the search modalities are derived from a diverse set of knowledge sources, such as text transcript from speech recognition, low-level visual features from video frames, and high-level semantic visual concepts from supervised learning. Since the effectiveness of each search modality greatly depends on specific user queries, prompt determination of the importance of a modality to a user query is a critical issue in multi-modality search. Our proposed approach, named concept-driven multi-modality fusion (CDMF), explores a large set of predefined semantic concepts for computing multi-modality fusion weights in a novel way. Specifically, in CDMF, we decompose the query-modality relationship into two components that are much easier to compute: query-concept relatedness and concept-modality relevancy. The former can be efficiently estimated online using semantic and visual mapping techniques, while the latter can be computed offline based on concept detection accuracy of each modality. Such a decomposition facilitates the need of adaptive learning of fusion weights for each user query on-the-fly, in contrast to the existing approaches which mostly adopted predefined query classes and/or modality weights. Experimental results on TREC video-retrieval evaluation 2005-2008 dataset validate the effectiveness of our approach, which outperforms the existing multi-modality fusion methods and achieves near-optimal performance (from oracle fusion) for many test queries.

Based on the local keypoints extracted as salient image patches, an image can be described as a ¿bag-of-visual-words (BoW)¿ and this representation has appeared promising for object and scene classification. The performance of BoW features in semantic concept detection for large-scale multimedia databases is subject to various representation choices. In this paper, we conduct a comprehensive study on the representation choices of BoW, including vocabulary size, weighting scheme, stop word removal, feature selection, spatial information, and visual bi-gram. We offer practical insights in how to optimize the performance of BoW by choosing appropriate representation choices. For the weighting scheme, we elaborate a soft-weighting method to assess the significance of a visual word to an image. We experimentally show that the soft-weighting outperforms other popular weighting schemes such as TF-IDF with a large margin. Our extensive experiments on TRECVID data sets also indicate that BoW feature alone, with appropriate representation choices, already produces highly competitive concept detection performance. Based on our empirical findings, we further apply our method to detect a large set of 374 semantic concepts. The detectors, as well as the features and detection scores on several recent benchmark data sets, are released to the multimedia community.

Semi-supervised learning (SSL) relies on partial supervision information for prediction, where only a small set of samples are associated with labels. Performance of SSL is significantly degraded if the given labels are not reliable. Such problems arise in realistic applications such as web image search using noisy textual tags. This paper proposes a novel and efficient graph based SSL method with the unique capacity of pruning contradictory labels and inferring new labels through a bidirectional and alternating optimization process. The objective is to automatically identify the most suitable samples for manipulation, labeling or unlabeling, and meanwhile estimate a smooth classification function over a weighted graph. Different from other graph based SSL approaches, the proposed method employs a bivariate objective function and iteratively modifies label variables on both labeled and unlabeled samples. Starting from such a SSL setting, we present a relearning framework to improve the performance of base learner, particularly for the application of web image search. Besides the toy demonstration on artificial data, we evaluated the proposed method on Flickr image search with unreliable textual labels. Experimental results confirm the significant improvements of the method over the baseline text based search engine and the state-of-the-art SSL methods.

Learning to cope with domain change has been known as a challenging problem in many real-world applications. This paper proposes a novel and efficient approach, named domain adaptive semantic diffusion (DASD), to exploit semantic context while considering the domain-shift-of-context for large scale video concept annotation. Starting with a large set of concept detectors, the proposed DASD refines the initial annotation results using graph diffusion technique, which preserves the consistency and smoothness of the annotation over a semantic graph. Different from the existing graph learning methods which capture relations among data samples, the semantic graph treats concepts as nodes and the concept affinities as the weights of edges. Particularly, the DASD approach is capable of simultaneously improving the annotation results and adapting the concept affinities to new test data. The adaptation provides a means to handle domain change between training and test data, which occurs very often in video annotation task. We conduct extensive experiments to improve annotation results of 374 concepts over 340 hours of videos from TRECVID 2005-2007 data sets. Results show consistent and significant performance gain over various baselines. In addition, the proposed approach is very efficient, completing DASD over 374 concepts within just 2 milliseconds for each video shot on a regular PC.

Automatic video search based on semantic concept detectors has recently received significant attention. Since the number of available detectors is much smaller than the size of human vocabulary, one major challenge is to select appropriate detectors to response user queries. In this paper, we propose a novel approach that leverages heterogeneous knowledge sources for domain adaptive video search. First, instead of utilizing WordNet as most existing works, we exploit the context information associated with Flickr images to estimate query-detector similarity. The resulting measurement, named Flickr context similarity (FCS), reflects the co-occurrence statistics of words in image context rather than textual corpus. Starting from an initial detector set determined by FCS, our approach novelly transfers semantic context learned from test data domain to adaptively refine the query-detector similarity. The semantic context transfer process provides an effective means to cope with the domain shift between external knowledge source (e.g., Flickr context) and test data, which is a critical issue in video search. To the best of our knowledge, this work represents the first research aiming to tackle the challenging issue of domain change in video search. Extensive experiments on 120 textual queries over TRECVID 2005-2008 data sets demonstrate the effectiveness of semantic context transfer for domain adaptive video search. Results also show that the FCS is suitable for measuring query-detector similarity, producing better performance to various other popular measures.

Human visual perception is able to recognize a wide range of targets under challenging conditions, but has limited throughput. Machine vision and automatic content analytics can process images at a high speed, but suffers from inadequate recognition accuracy for general target classes. In this paper, we propose a new paradigm to explore and combine the strengths of both systems. A single trial EEG-based brain machine interface (BCI) subsystem is used to detect objects of interest of arbitrary classes from an initial subset of images. The EEG detection outcomes are used as input to a graph-based pattern mining subsystem to identify, refine, and propagate the labels to retrieve relevant images from a much larger pool. The combined strategy is unique in its generality, robustness, and high throughput. It has great potential for advancing the state of the art in media retrieval applications. We have evaluated and demonstrated significant performance gains of the proposed system with multiple and diverse image classes over several data sets, including those from Internet (Caltech 101) and remote sensing images. In this paper, we will also present insights learned from the experiments and discuss future research directions.

This paper proposes a novel approach to exploit the ontological relationship of visual words by linguistic reasoning. A visual word ontology is constructed to facilitate the rigorous evaluation of linguistic similarity across visual words. The linguistic similarity measurement enables cross-bin matching of visual words, compromising the effectiveness and speed of conventional keypoint matching and bag-of-word approaches. A constraint EMD is proposed and experimented to efficiently match visual words. Empirical findings indicate that the proposed approach offers satisfactory performance to near-duplicate retrieval, while still enjoying the merit of speed efficiency compared with other techniques.

Event detection plays an essential role in video content analysis. However, the existing features are still weak in event detection because: i) most features just capture what is involved in an event or how the event evolves separately, and thus cannot completely describe the event; ii) to capture event evolution information, only motion distribution over the whole frame is used which proves to be noisy in unconstrained videos; iii) the estimated object motion is usually distorted by camera movement. To cope with these problems, in this paper, we propose a new motion feature, namely Expanded Relative Motion Histogram of Bag-of-Visual-Words (ERMH-BoW) to employ motion relativity and visual relatedness for event detection. In ERMH-BoW, by representing what aspect of an event with Bag-of-Visual-Words (BoW), we construct relative motion histograms between visual words to depict the object activities or how aspect of the event. ERMH-BoW thus integrates both what and how aspects for a complete event description. Instead of motion distribution features, local motion of visual words is employed which is more discriminative in event detection. Meanwhile, we show that by employing relative motion, ERMH-BoW is able to honestly describe object activities in an event regardless of varying camera movement. Besides, to alleviate the visual word correlation problem in BoW, we propose a novel method to expand the relative motion histogram. The expansion is achieved by diffusing the relative motion among correlated visual words measured by visual relatedness. To validate the effectiveness of the proposed feature, ERMH-BoW is used to measure video clip similarity with Earth Mover's Distance (EMD) for event detection. We conduct experiments for detecting LSCOM events in TRECVID 2005 video corpus, and performance is improved by 74% and 24% compared with existing motion distribution feature and BoW feature respectively.

Bag-of-visual-words (BoW) has been popular for visual classification in recent years. In this paper, we propose a novel BoW expansion method to alleviate the effect of visual word correlation problem. We achieve this by diffusing the weights of visual words in BoW based on visual word relatedness, which is rigorously defined within a visual ontology. The proposed method is tested in video indexing experiment on TRECVID-2006 video retrieval benchmark, and an improvement of 7% over the traditional BoW is reported.

This paper describes the construction and utilization of two novel semantic spaces, namely ontology-enriched semantic space (OSS) and ontology-enriched orthogonal semantic space (OS2), to facilitate the selection of concept detectors for video search. These two semantic spaces are enriched with ontology knowledge, while emphasizing consistent and uniform comparison of ontological relatedness among concepts for query-to-concept mapping. OS2, in addition to being a linear space like OSS, also guarantees orthogonality of the semantic space. Compared with other ontology reasoning measures, both spaces are capable of providing platforms that offer a global view of concept inter-relatedness, by allowing evaluation of concept similarity in metric spaces. We simulate OSS and OS2by using LSCOM concepts and experiment search effectiveness with VIREO-374 concept detectors. Empirical observations indicate that the proposed semantic spaces enable more effective selection of concept detectors than eight other existing ontology measures. OS2, in particular, is better in providing a viable and reasonable solution for fusion of multiple concept detectors.

The identification of near-duplicate keyframe (NDK) pairs is a useful task for a variety of applications such as news story threading and content-based video search. In this paper, we propose a novel approach for the discovery and tracking of NDK pairs and threads in the broadcast domain. The detection of NDKs in a large data set is a challenging task due to the fact that when the data set increases linearly, the computational cost increases in a quadratic speed, and so does the number of false alarms. This paper explores the symmetric and transitive nature of near-duplicate for the effective detection and fast tracking of NDK pairs based upon the matching of local keypoints in frames. In the detection phase, we propose a robust measure, namely pattern entropy (PE), to measure the coherency of symmetric keypoint matching across the space of two keyframes. This measure is shown to be effective in discovering the NDK identity of a frame. In the tracking phase, the NDK pairs and threads are rapidly propagated and linked with sitivity without the need of detection. This step ends up a significant boost in speed efficiency. We evaluate proposed approach against a month of the 2004 broadcast videos. The experimental results indicate our approach outperforms other techniques in terms of recall and precision with a large margin. In addition, by considering the transitivity and the underlying distribution of NDK pairs along time span, a speed up of 3 to 5 times is achieved when keeping the performance close enough to the optimal one obtained by exhaustive evaluation.