Large multimodal models (LMMs) with advanced video analysis capabilities have recently garnered significant attention. However, most evaluations rely on traditional methods like multiple-choice question answering in benchmarks such as VideoMME and LongVideoBench, which are prone to lack the depth needed to capture the complex demands of real-world users. To address this limitation—and due to the prohibitive cost and slow pace of human annotation for video tasks—we introduce VideoAutoArena, an arena-style benchmark inspired by LMSYS Chatbot Arena’s framework, designed to automatically assess LMMs’ video analysis abilities. VideoAutoArena utilizes user simulation to generate open-ended, adaptive questions that rigorously assess model performance in video understanding. The benchmark features an automated, scalable evaluation framework, incorporating a modified ELO Rating System for fair and continuous comparisons across multiple LMMs. To validate our automated judging system, we construct a "gold standard" using a carefully curated subset of human annotations, demonstrating that our arena strongly aligns with human judgment while maintaining scalability. Additionally, we introduce a fault-driven evolution strategy, progressively increasing question complexity to push models toward handling more challenging video analysis scenarios. Experimental results demonstrate that VideoAutoArena effectively differentiates among state-of-the-art LMMs, providing insights into model strengths and areas for improvement. To further streamline our evaluation, we introduce VideoAutoBench as an auxiliary benchmark, where human annotators label winners in a subset of VideoAutoArena battles. We use GPT-4o as a judge to compare responses against these human-validated answers. Together, VideoAu-Toarena and VideoAutoBench offer a cost-effective, and scalable framework for evaluating LMMs in user-centric video analysis.

Achieving effective unified pretraining on large time series corpora remains an open challenge in developing time series foundation models. Existing methods, such as Moirai, introduce multiple projection layers for time series of different frequencies to account for high data heterogeneity. We identify major drawbacks to this human-imposed frequency-level model specialization. First, frequency is not a reliable indicator for grouping pretraining data. Second, time series can display varied distributions even within a short window. Frequency-level specialization overlooks the diversity at this granularity. To address these issues, this paper introduces Moirai-MoE, excluding human-defined data groupings while delegating the modeling of diverse time series patterns to the sparse mixture of experts (MoE) within Transformers. With this design, Moirai-MoE eliminates reliance on heuristics and enables automatic token-level specialization. Extensive evaluations on 39 datasets demonstrate the superiority of Moirai-MoE over state-of-the-art foundation models. This study also conducts comprehensive model analyses to explore the inner workings of time series MoE foundation models.

We introduce Reward-Guided Speculative Decoding (RSD), a novel framework aimed at improving the efficiency of inference in large language models (LLMs). RSD synergistically combines a lightweight draft model with a more powerful target model, incorporating a controlled bias to prioritize high-reward outputs, in contrast to existing speculative decoding methods that enforce strict unbiasedness. RSD employs a process reward model to evaluate intermediate decoding steps and dynamically decide whether to invoke the target model, optimizing the trade-off between computational cost and output quality. We theoretically demonstrate that a threshold-based mixture strategy achieves an optimal balance between resource utilization and performance. Extensive evaluations on challenging reasoning benchmarks, including Olympiad-level tasks, show that RSD delivers significant efficiency gains against decoding with the target model only (up to 4.4X fewer FLOPs), while achieving significant better accuracy than parallel decoding method on average (up to +3.5). These results highlight RSD as a robust and cost-effective approach for deploying LLMs in resource-intensive scenarios.

As large language models (LLMs) continue to advance, their capacity to function effectively across a diverse range of languages has shown marked improvement. Preliminary studies observe that the hidden activations of LLMs often resemble English, even when responding to non-English prompts. This has led to the widespread assumption that LLMs may ``think'' in English. However, more recent results showing strong multilingual performance, even surpassing English performance on specific tasks in other languages, challenge this view. In this work, we find that LLMs progressively develop a core language-agnostic parameter space—a remarkably small subset of parameters whose deactivation results in significant performance degradation across all languages. This compact yet critical set of parameters underlies the model’s ability to generalize beyond individual languages, supporting the emergence of abstract thought that is not tied to any specific linguistic system. Specifically, we identify language-related neurons—those are consistently activated during the processing of particular languages, and categorize them as either shared (active across multiple languages) or exclusive (specific to one). As LLMs undergo continued development over time, we observe a marked increase in both the proportion and functional importance of shared neurons, while exclusive neurons progressively diminish in influence. These shared neurons constitute the backbone of the core language-agnostic parameter space, supporting the emergence of abstract thought. Motivated by these insights, we propose neuron-specific training strategies tailored to LLMs' language-agnostic levels at different development stages. Experiments across diverse LLM families support our approach. Our codes are available at https://anonymous.4open.science/status/S-C393.

Recent advancements in multimodal pretraining have shown promising efficacy in 3D representation learning by aligning multimodal features across 3D shapes, their 2D counterparts, and language descriptions. However, the methods used by existing frameworks to curate such multimodal data, in particular language descriptions for 3D shapes, are not scalable, and the collected language descriptions are not diverse. To address this, we introduce ULIP-2, a simple yet effective tri-modal pretraining framework that leverages large multimodal models to automatically generate holistic language descriptions for 3D shapes. It only needs 3D data as input, eliminating the need for any manual 3D annotations, and is therefore scalable to large datasets. ULIP-2 is also equipped with scaled-up backbones for better multimodal representation learning. We conduct experiments on two large-scale 3D datasets, Objaverse and ShapeNet, and augment them with tri-modal datasets of 3D point clouds, images, and language for training ULIP-2. Experiments show that ULIP-2 demonstrates substantial benefits in three downstream tasks: zero-shot 3D classification, standard 3D classification with fine-tuning, and 3D captioning (3D-to-language generation). It achieves a new SOTA of 50.6% (top-1) on Objaverse-LVIS and 84.7% (top-1) on ModelNet40 in zero-shot classification. In the ScanObjectNN benchmark for standard fine-tuning, ULIP-2 reaches an overall accuracy of 91.5% with a compact model of only 1.4 million parameters. ULIP-2 sheds light on a new paradigm for scalable multimodal 3D representation learning without human annotations and shows significant improvements over existing baselines. The code and datasets are released at https://github.com/salesforce/ULIP.

Anthony Tiong, Junqi Zhao, Boyang Li, Junnan Li, Steven Hoi, Caiming Xiong. 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 models (LMMs) are processing increasingly longer and richer inputs. Albeit the progress, few public benchmark is available to measure such development. To mitigate this gap, we introduce LongVideoBench, a question-answering benchmark that features video-language interleaved inputs up to an hour long. Our benchmark includes 3,763 varying-length web-collected videos with their subtitles across diverse themes, designed to comprehensively evaluate LMMs on long-term multimodal understanding. To achieve this, we interpret the primary challenge as to accurately retrieve and reason over detailed multimodal information from long inputs. As such, we formulate a novel video question-answering task termed referring reasoning. Specifically, as part of the question, it contains a referring query that references related video contexts, called referred context. The model is then required to reason over relevant video details from the referred context. Following the paradigm of referring reasoning, we curate 6,678 human-annotated multiple-choice questions in 17 fine-grained categories, establishing one of the most comprehensive benchmarks for long-form video understanding. Evaluations suggest that the LongVideoBench presents significant challenges even for the most advanced proprietary models (e.g. GPT-4o, Gemini-1.5-Pro), while their open-source counterparts show an even larger performance gap. In addition, our results indicate that model performance on the benchmark improves only when they are capable of processing more frames, positioning LongVideoBench as a valuable benchmark for evaluating future-generation long-context LMMs.

Data heterogeneity across clients in federated learning (FL) settings is a widely acknowledged challenge. In response, personalized federated learning (PFL) emerged as a framework to curate local models for clients' tasks. In PFL, a common strategy is to develop local and global models jointly - the global model (for generalization) informs the local models, and the local models (for personalization) are aggregated to update the global model. A key observation is that if we can improve the generalization ability of local models, then we can improve the generalization of global models, which in turn builds better personalized models. In this work, we consider class imbalance, an overlooked type of data heterogeneity, in the classification setting. We propose FedNH, a novel method that improves the local models' performance for both personalization and generalization by combining the uniformity and semantics of class prototypes. FedNH initially distributes class prototypes uniformly in the latent space and smoothly infuses the class semantics into class prototypes. We show that imposing uniformity helps to combat prototype collapse while infusing class semantics improves local models. Extensive experiments were conducted on popular classification datasets under the cross-device setting. Our results demonstrate the effectiveness and stability of our method over recent works.

Large language models (LLMs) have demonstrated excellent zero-shot generalization to new language tasks. However, effective utilization of LLMs for zero-shot visual question-answering (VQA) remains challenging, primarily due to the modality disconnect and task disconnect between the LLM and VQA tasks. End-to-end training on multimodal data may bridge the disconnects, but is inflexible and computationally expensive. To address this issue, we propose Img2LLM, a plug-and-play module that provides LLM prompts to enable LLMs to perform zeroshot VQA tasks without end-to-end training. We develop LLM-agnostic models describe image content as exemplar question-answer pairs, which prove to be effective LLM prompts. Img2LLM offers the following benefits: 1) It achieves comparable or better performance than methods relying on end-to-end training. For example, we outperform Flamingo [3] by 5.6% on VQAv2. On the challenging A-OKVQA dataset, our method outperforms few-shot methods by as much as 20%. 2) It flexibly interfaces with a wide range of LLMs to perform VQA. 3) It eliminates the need to specialize LLMs using end-to-end finetuning and serve highly specialized LLMs to end users, thereby reducing cost. Code is available via the LAVIS [28] framework at https://github.com/salesforce/LAVIS/tree/main/projects/img2llm-vqa.

Large language models (LLMs) pretrained on vast source code have achieved prominent progress in code intelligence. However, existing code LLMs have two main limitations. First, they often adopt a specific architecture (encoder-only or decoder-only) or rely on a unified encoder-decoder network for different downstream tasks, lacking the flexibility to operate in the optimal architecture for a specific task. Secondly, they often employ a limited set of pretraining objectives which might not be relevant to some tasks and hence result in substantial performance degrade. To address these limitations, we propose “CodeT5+”, a family of encoder-decoder LLMs for code in which component modules can be flexibly combined to suit a wide range of code tasks. Such flexibility is enabled by our proposed mixture of pretraining objectives, which cover span denoising, contrastive learning, text-code matching, and causal LM pretraining tasks, on both unimodal and bimodal multilingual code corpora. Furthermore, we propose to initialize CodeT5+ with frozen off-the-shelf LLMs without training from scratch to efficiently scale up our models, and explore instruction-tuning to align with natural language instructions. We extensively evaluate CodeT5+ on over 20 code-related benchmarks in different settings, including zero-shot, finetuning, and instruction-tuning. We observe state-of-the-art (SoTA) performance on various code-related tasks, and our instruction-tuned CodeT5+ 16B achieves new SoTA results of 35.0% pass@1 and 54.5% pass@10 on the HumanEval code generation task against other open code LLMs, even surpassing the OpenAI code-cushman-001 model.

The cost of vision-and-language pre-training has become increasingly prohibitive due to end-to-end training of large-scale models. This paper proposes BLIP-2, a generic and efficient pre-training strategy that bootstraps vision-language pre-training from off-the-shelf frozen pre-trained image encoders and frozen large language models. BLIP-2 bridges the modality gap with a lightweight Querying Transformer, which is pre-trained in two stages. The first stage bootstraps vision-language representation learning from a frozen image encoder. The second stage bootstraps vision-to-language generative learning from a frozen language model. BLIP-2 achieves state-of-the-art performance on various vision-language tasks, despite having significantly fewer trainable parameters than existing methods. For example, our model outperforms Flamingo80B by 8.7% on zero-shot VQAv2 with 54x fewer trainable parameters. We also demonstrate the model’s emerging capabilities of zero-shot image-to-text generation that can follow natural language instructions.

Large-scale pre-training and instruction tuning have been successful at creating general-purpose language models with broad competence. However, building general-purpose vision-language models is challenging due to the rich input distributions and task diversity resulting from the additional visual input. Although vision-language pretraining has been widely studied, vision-language instruction tuning remains under-explored. In this paper, we conduct a systematic and comprehensive study on vision-language instruction tuning based on the pretrained BLIP-2 models. We gather 26 publicly available datasets, covering a wide variety of tasks and capabilities, and transform them into instruction tuning format. Additionally, we introduce an instruction-aware Query Transformer, which extracts informative features tailored to the given instruction. Trained on 13 held-in datasets, InstructBLIP attains state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and larger Flamingo models. Our models also lead to state-of-the-art performance when finetuned on individual downstream tasks (e.g., 90.7% accuracy on ScienceQA questions with image contexts). Furthermore, we qualitatively demonstrate the advantages of InstructBLIP over concurrent multimodal models. All InstructBLIP models are open-source.

Subject-driven text-to-image generation models create novel renditions of an input subject based on text prompts. Existing models suffer from lengthy fine-tuning and difficulties preserving the subject fidelity. To overcome these limitations, we introduce BLIP-Diffusion, a new subject-driven image generation model that supports multimodal control which consumes inputs of subject images and text prompts. Unlike other subject-driven generation models, BLIP-Diffusion introduces a new multimodal encoder which is pre-trained to provide subject representation. We first pre-train the multimodal encoder following BLIP-2 to produce visual representation aligned with the text. Then we design a subject representation learning task which enables a diffusion model to leverage such visual representation and generates new subject renditions. Compared with previous methods such as DreamBooth, our model enables zero-shot subject-driven generation, and efficient fine-tuning for customized subject with up to 20x speedup. We also demonstrate that BLIP-Diffusion can be flexibly combined with existing techniques such as ControlNet and prompt-to-prompt to enable novel subject-driven generation and editing applications. Implementations are available at: https://github.com/salesforce/LAVIS/tree/main/projects/blip-diffusion.

The goal of semantic code search or text-to-code search is to retrieve a semantically relevant code snippet from an existing code database using a natural language query. When constructing a practical semantic code search system, existing approaches fail to provide an optimal balance between retrieval speed and the relevance of the retrieved results. We propose an efficient and effective text-to-code search framework with cascaded fast and slow models, in which a fast transformer encoder model is learned to optimize a scalable index for fast retrieval followed by learning a slow classification-based re-ranking model to improve the accuracy of the top K results from the fast retrieval. To further reduce the high memory cost of deploying two separate models in practice, we propose to jointly train the fast and slow model based on a single transformer encoder with shared parameters. Empirically our cascaded method is not only efficient and scalable, but also achieves state-of-the-art results with an average mean reciprocal ranking (MRR) score of 0.7795 (across 6 programming languages) on the CodeSearchNet benchmark as opposed to the prior state-of-the-art result of 0.744 MRR. Our codebase can be found at this link.

Yidco-and-language pre-training has shown promising improvements on various downstream tasks. Most previous methods capture cross-modal interactions with a standard transformer-based multimodal encoder, not fully addressing the misalignment between unimodal video and text features. Besides, learning finegrained visual-language alignment usually requires off-the-shelf object detectors to provide object information, which is bottlenecked by the detector's limited vocabulary and expensive computation cost. In this paper, we propose Align and Prompt: a new video-and-language pre-training framework (AlPro), which operates on sparsely-sampled video frames and achieves more effective cross-modal alignment without explicit object detectors. First, we introduce a video-text contrastive (VTC) loss to align unimodal video-text features at the instance level, which eases the modeling of cross-modal interactions. Then, we propose a novel visually-grounded pre-training task, prompting entity modeling (PEM), which learns finegrained alignment between visual region and text entity via an entity prompter module in a self-supervised way. Finally, we pretrain the video-and-language transformer models on large webly-source video-text pairs using the proposed VTC and PEM losses as well as two standard losses of masked language modeling (MLM) and video-text matching (VTM). The resulting pre-trained model achieves state-of-the-art performance on both text-video retrieval and videoQA, outperforming prior work by a substantial margin. Implementation and pre-trained models are available at https://github.com/salesforce/ALPRO.

Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks. However, most existing pre-trained models only excel in either understanding-based tasks or generation-based tasks. Furthermore, performance improvement has been largely achieved by scaling up the dataset with noisy image-text pairs collected from the web, which is a suboptimal source of supervision. In this paper, we propose BLIP, a new VLP framework which transfers flexibly to both vision-language understanding and generation tasks. BLIP effectively utilizes the noisy web data by bootstrapping the captions, where a captioner generates synthetic captions and a filter removes the noisy ones. We achieve state-of-the-art results on a wide range of vision-language tasks, such as image-text retrieval (+2.7% in average recall@1), image captioning (+2.8% in CIDEr), and VQA (+1.6% in VQA score). BLIP also demonstrates strong generalization ability when directly transferred to video-language tasks in a zero-shot manner. Code and models are available at https://github.com/salesforce/BLIP.

Semi-supervised learning has been an effective paradigm for leveraging unlabeled data to reduce the reliance on labeled data. We propose CoMatch, a new semi-supervised learning method that unifies dominant approaches and addresses their limitations. CoMatch jointly learns two representations of the training data, their class probabilities and low-dimensional embeddings. The two representations interact with each other to jointly evolve. The embeddings impose a smoothness constraint on the class probabilities to improve the pseudo-labels, whereas the pseudo-labels regularize the structure of the embeddings through graph-based contrastive learning. CoMatch achieves state-of-the-art performance on multiple datasets. It achieves substantial accuracy improvements on the label-scarce CIFAR-10 and STL-10. On ImageNet with 1% labels, CoMatch achieves a top-1 accuracy of 66.0%, outperforming FixMatch [32] by 12.6%. Furthermore, CoMatch achieves better representation learning performance on downstream tasks, outperforming both supervised learning and self-supervised learning. Code and pre-trained models are available at https://github.com/salesforce/CoMatch/.

Learning from noisy data has attracted much attention, where most methods focus on label noise. In this work, we propose a new learning framework which simultaneously addresses three types of noise commonly seen in real-world data: label noise, out-of-distribution input, and input corruption. In contrast to most existing methods, we combat noise by learning robust representation. Specifically, we embed images into a low-dimensional subspace, and regularize the geometric structure of the subspace with robust contrastive learning, which includes an unsupervised consistency loss and a supervised mixup prototypical loss. We also propose a new noise cleaning method which leverages the learned representation to enforce a smoothness constraint on neighboring samples. Experiments on multiple benchmarks demonstrate state-of-the-art performance of our method and robustness of the learned representation. Code is available at https://github.com/salesforce/RRL/.

Large-scale vision and language representation learning has shown promising improvements on various vision-language tasks. Most existing methods employ a transformer-based multimodal encoder to jointly model visual tokens (region-based image features) and word tokens. Because the visual tokens and word tokens are unaligned, it is challenging for the multimodal encoder to learn image-text interactions. In this paper, we introduce a contrastive loss to ALign the image and text representations BEfore Fusing (ALBEF) them through cross-modal attention, which enables more grounded vision and language representation learning. Unlike most existing methods, our method does not require bounding box annotations nor high-resolution images. In order to improve learning from noisy web data, we propose momentum distillation, a self-training method which learns from pseudo-targets produced by a momentum model. We provide a theoretical analysis of ALBEF from a mutual information maximization perspective, showing that different training tasks can be interpreted as different ways to generate views for an image-text pair. ALBEF achieves state-of-the-art performance on multiple downstream vision-language tasks. On image-text retrieval, ALBEF outperforms methods that are pre-trained on orders of magnitude larger datasets. On VQA and NLVR$^2$, ALBEF achieves absolute improvements of 2.37% and 3.84% compared to the state-of-the-art, while enjoying faster inference speed. Code and models are available at https://github.com/salesforce/ALBEF.

The recent development of commodity 360° cameras have enabled a single video to capture an entire scene, which endows promising potentials in surveillance scenarios. However, research in omnidirectional video analysis has lagged behind the hardware advances. In this work, we address the important problem of action recognition in topview 360° videos. Due to the wide filed-of-view, 360° videos usually capture multiple people performing actions at the same time. Furthermore, the appearance of people are deformed. The proposed framework first transforms top-view omnidirectional videos into panoramic videos using a calibrationfree method. Then spatial-temporal features are extracted using region-based 3D CNNs for action recognition. We propose a weakly-supervised method based on multiinstance multi-label learning, which trains the model to recognize and localize multiple actions in a video using only video-level action labels as supervision. We perform experiments to quantitatively validate the efficacy of the proposed method over state-of-the-art baselines and variants of our model, and qualitatively demonstrate action localization results. To enable research in this direction, we introduce the 360Action dataset. It is the first omnidirectional video dataset for multi-person action recognition with a diverse set of scenes, actors and actions. The dataset is available at https://github.com/ryukenzen/360action.

The computer vision community is witnessing an unprecedented rate of new tasks being proposed and addressed, thanks to the deep convolutional networks' capability to find complex mappings from X to Y. The advent of each task often accompanies the release of a large-scale annotated dataset, for supervised training of deep network. However, it is expensive and time-consuming to manually label sufficient amount of training data. Therefore, it is important to develop algorithms that can leverage off-the-shelf labeled dataset to learn useful knowledge for the target task. While previous works mostly focus on transfer learning from a single source, we study multi-source transfer across domains and tasks (MS-DTT), in a semi-supervised setting. We propose GradMix, a model-agnostic method applicable to any model trained with gradient-based learning rule, to transfer knowledge via gradient descent by weighting and mixing the gradients from all sources during training. GradMix follows a meta-learning objective, which assigns layer-wise weights to the source gradients, such that the combined gradient follows the direction that minimize the loss for a small set of samples from the target dataset. In addition, we propose to adaptively adjust the learning rate for each mini-batch based on its importance to the target task, and a pseudo-labeling method to leverage the unlabeled samples in the target domain. We conduct MS-DTT experiments on two tasks: digit recognition and action recognition, and demonstrate the advantageous performance of the proposed method against multiple baselines.

Bridging vision and natural language is a longstanding goal in computer vision and multimedia research. While earlier works focus on generating a single-sentence description for visual content, recent works have studied paragraph generation. In this paper, we introduce the problem of video storytelling, which aims at generating coherent and succinct stories for long videos. Video storytelling introduces new challenges, mainly due to the diversity of the story and the length and complexity of the video. We propose novel methods to address the challenges. First, we propose a context-aware framework for multimodal embedding learning, where we design a residual bidirectional recurrent neural network to leverage contextual information from past and future. The multimodal embedding is then used to retrieve sentences for video clips. Second, we propose a Narrator model to select clips that are representative of the underlying storyline. The Narrator is formulated as a reinforcement learning agent, which is trained by directly optimizing the textual metric of the generated story. We evaluate our method on the video story dataset, a new dataset that we have collected to enable the study. We compare our method with multiple state-of-the-art baselines and show that our method achieves better performance, in terms of quantitative measures and user study.

The recent advances in instance-level detection tasks lay strong foundation for genuine comprehension of the visual scenes. However, the ability to fully comprehend a social scene is still in its preliminary stage. In this work, we focus on detecting human-object interactions (HOIs) in social scene images, which is demanding in terms of research and increasingly useful for practical applications. To undertake social tasks interacting with objects, humans direct their attention and move their body based on their intention. Based on this observation, we provide a unique computational perspective to explore human intention in HOI detection. Specifically, the proposed human intention-driven HOI detection (iHOI) framework models human pose with the relative distances from body joints to the object instances. It also utilizes human gaze to guide the attended contextual regions in a weakly-supervised setting. In addition, we propose a hard negative sampling strategy to address the problem of mis-grouping. We perform extensive experiments on two benchmark datasets, namely V-COCO and HICO-DET. The efficacy of each proposed component has also been validated.

Despite the success of deep neural networks (DNNs) in image classification tasks, the human-level performance relies on massive training data with high-quality manual annotations, which are expensive and time-consuming to collect. There exist many inexpensive data sources on the web, but they tend to contain inaccurate labels. Training on noisy labeled datasets causes performance degradation because DNNs can easily overfit to the label noise. To overcome this problem, we propose a noise-tolerant training algorithm, where a meta-learning update is performed prior to conventional gradient update. The proposed meta-learning method simulates actual training by generating synthetic noisy labels, and train the model such that after one gradient update using each set of synthetic noisy labels, the model does not overfit to the specific noise. We conduct extensive experiments on the noisy CIFAR-10 dataset and the Clothing1M dataset. The results demonstrate the advantageous performance of the proposed method compared to several state-of-the-art baselines.

The recent advances in instance-level detection tasks lay a strong foundation for automated visual scenes understanding. However, the ability to fully comprehend a social scene still eludes us. In this work, we focus on detecting human-object interactions (HOIs) in images, an essential step towards deeper scene understanding. HOI detection aims to localize human and objects, as well as to identify the complex interactions between them. Innate in practical problems with large label space, HOI categories exhibit a long-tail distribution, i.e., there exist some rare categories with very few training samples. Given the key observation that HOIs contain intrinsic semantic regularities despite they are visually diverse, we tackle the challenge of long-tail HOI categories by modeling the underlying regularities among verbs and objects in HOIs as well as general relationships. In particular, we construct a knowledge graph based on the ground-truth annotations of training dataset and external source. In contrast to direct knowledge incorporation, we address the necessity of dynamic image-specific knowledge retrieval by multi-modal learning, which leads to an enhanced semantic embedding space for HOI comprehension. The proposed method shows improved performance on V-COCO and HICO-DET benchmarks, especially when predicting the rare HOI categories.

The amount of 360-degree panoramas shared online has been rapidly increasing due to the availability of affordable and compact omnidirectional cameras, which offers huge amount of new information unavailable before. In this paper, we present the first work to exploit unlabeled 360-degree data for image representation learning. We propose middle-out, a new self-supervised learning task, which leverages the spatial configuration of normal field-of-view images sampled from a 360-degree image as supervisory signal. We train a Siamese ConvNet model to identify the middle image among three shuffled images sampled from a panorama by perspective projection. Compared to previous self-supervised methods that train models using image patches or video frames with limited field-of-view, our method leverages the rich semantic information contained in 360-degree images and enforces the model to not only learn about objects, but also develop a higher-level understanding about object relationships and scene structures. We quantitatively demonstrate that the feature representation learned using the proposed task is useful for a wide range of vision tasks including object classification, object detection, scene classification, semantic segmentation, and geometry estimation. We also qualitatively show that the proposed method can enforce the ConvNet to extract high-level semantic concepts, an ability which previous self-supervised learning methods have not acquired.

Presentation has been an effective method for delivering information to an audience for many years. Over the past few decades, technological advancements have revolutionized the way humans deliver presentation. Conventionally, the quality of a presentation is usually evaluated through painstaking manual analysis with experts. Although the expert feedback is effective in assisting users to improve their presentation skills, manual evaluation suffers from high cost and is often not available to most individuals. In this work, we propose a novel multi-sensor self-quantification system for presentations, which is designed based on a new proposed assessment rubric. We present our analytics model with conventional ambient sensors (i.e., static cameras and Kinect sensor) and the emerging wearable egocentric sensors (i.e., Google Glass). In addition, we performed a cross-correlation analysis of speaker’s vocal behavior and body language. The proposed framework is evaluated on a new presentation dataset, namely, NUS Multi-Sensor Presentation dataset, which consists of 51 presentations covering a diverse range of topics. To validate the efficacy of the proposed system, we have conducted a series of user studies with the speakers and an interview with an English communication expert, which reveals positive and promising feedback.

The recent success in human action recognition with deep learning methods mostly adopt the supervised learning paradigm, which requires significant amount of manually labeled data to achieve good performance. However, label collection is an expensive and time-consuming process. In this work, we propose an unsupervised learning framework, which exploits unlabeled data to learn video representations. Different from previous works in video representation learning, our unsupervised learning task is to predict 3D motion in multiple target views using video representation from a source view. By learning to extrapolate cross-view motions, the representation can capture view-invariant motion dynamics which is discriminative for the action. In addition, we propose a view-adversarial training method to enhance learning of view-invariant features. We demonstrate the effectiveness of the learned representations for action recognition on multiple datasets.

Since the beginning of early civilizations, social relationships derived from each individual fundamentally form the basis of social structure in our daily life. In the computer vision literature, much progress has been made in scene understanding, such as object detection and scene parsing. Recent research focuses on the relationship between objects based on its functionality and geometrical relations. In this work, we aim to study the problem of social relationship recognition, in still images. We have proposed a dual-glance model for social relationship recognition, where the first glance fixates at the individual pair of interest and the second glance deploys attention mechanism to explore contextual cues. We have also collected a new large scale People in Social Context (PISC) dataset, which comprises of 22,670 images and 76,568 annotated samples from 9 types of social relationship. We provide benchmark results on the PISC dataset, and qualitatively demonstrate the efficacy of the proposed model.

Training deep learning based video classifiers for action recognition requires a large amount of labeled videos. The labeling process is labor-intensive and time-consuming. On the other hand, large amount of weakly-labeled images are uploaded to the Internet by users everyday. To harness the rich and highly diverse set of Web images, a scalable approach is to crawl these images to train deep learning based classifier, such as Convolutional Neural Networks (CNN). However, due to the domain shift problem, the performance of Web images trained deep classifiers tend to degrade when directly deployed to videos. One way to address this problem is to fine-tune the trained models on videos, but sufficient amount of annotated videos are still required. In this work, we propose a novel approach to transfer knowledge from image domain to video domain. The proposed method can adapt to the target domain (i.e. video data) with limited amount of training data. Our method maps the video frames into a low-dimensional feature space using the class-discriminative spatial attention map for CNNs. We design a novel Siamese EnergyNet structure to learn energy functions on the attention maps by jointly optimizing two loss functions, such that the attention map corresponding to a ground truth concept would have higher energy. We conduct extensive experiments on two challenging video recognition datasets (i.e. TVHI and UCF101), and demonstrate the efficacy of our proposed method.

Presentation is one of the most effective methods to disseminate information. Traditional methods to evaluate the quality of a presentation generally involves a human instructor, which is infeasible in many scenarios. Recent studies have focused on the automated assessment of presentations. A variety of systems have been developed that focus on analyzing various aspects of presentations. However, those systems are mainly limited by their performance, as they mostly adopt hand-crafted features and ad-hoc algorithms. In this work, we propose a multi-stream deep learning framework customized for presentation assessment. The framework uses Bidirectional Long Short-Term Memory with attention mechanism for temporal modeling, and fuses information from multiple modalities for the final decision. We also design a novel assessment rubric based on input from a domain expert. Experimental results on the NUS Multi-Sensor Presentation (NUSMAP) dataset show that the proposed framework is computationally efficient and achieves significant improvement in classification accuracy.

This technical demo presents PreSense, an Assistive Presentation Self-Quantification System. Oral presentation is traditionally evaluated by a human instructor, which is cost-ineffective and time-consuming. PreSense allows individuals to self-evaluate their presentation skills. The system is designed to receive multimodal inputs from multiple sources, including webcam, Kinect sensor and Google Glass. The multimodal data is processed by a deep assessment framework, which outputs the evaluation results based on a carefully designed assessment rubric. In order to systematically visualize the individual assessment results, we develop an interactive graphical interface and demonstrate its efficacy using presentation data from the NUS Multi-Sensor Presentation (NUSMAP) dataset.