Photorealistic self-avatars with facial tracking bring VR communication closer to face-to-face interaction, but also expose a fundamental limitation: people often lack accurate awareness and control of their own facial expressions. As a result, directly mirroring facial movements may not reliably convey users’ intended expressions in social VR. We propose modulating rendered expressions on avatars to improve social interaction in VR (e.g., exaggerating positive ones). To design effective modulation strategies and understand how modulation impacts user behavior and perception, we invited 18 participants to perform emotional expressions under varying modulation levels. We collected their self-reported ratings on accuracy and social appropriateness, as well as naturalness ratings and emotion recognition results from 18 additional observers who viewed the expressions. Results show that suppressing rendered expressions amplified participants’ actual facial movements, with effects differing by expression type. Based on our findings, we calculated the modulation ranges for different expressions and developed a real-time modulation system that exaggerates users’ facial expressions with a mild amplitude. We evaluated the system through a VR speech task where 12 participants delivered speeches with and without modulation. Results showed that the applied modulation improved emotional expressiveness and performance. Finally, we developed three application prototypes to illustrate the broader practical implications of our findings.

In this work, we introduce CAT-V (Caption Anything in Video), a training-free framework for fine-grained object-centric video captioning of user-selected instances. CAT-V combines (i) a SAMURAI-based Segmenter for precise object masks across frames, (ii) a TRACE-Uni Temporal Analyzer for event boundary detection and coarse event descriptions, and (iii) an InternVL-2.5 Captioner that, conditioned on spatiotemporal visual prompts and chain-of-thought (CoT) guidance, produces detailed, temporally coherent captions about object attributes, actions, states, interactions, and context. The system supports point, box, and region prompts and maintains temporal sensitivity by tracking object states across segments. In contrast to vanilla video captioning that is overly abstract and dense video captioning that is often terse, CAT-V enables object-level specificity with spatial accuracy and temporal coherence, without additional training data.

With the rapid growth of online video platforms and the escalating volume of video content, the need for proficient video understanding tools has increased significantly. Given the remarkable capabilities of large language models (LLMs) in language and multimodal tasks, this survey provides a detailed overview of recent advances in video understanding that harness the power of LLMs (Vid-LLMs). The emergent capabilities of Vid-LLMs are surprisingly advanced, particularly their ability for open-ended multi-granularity (abstract, temporal, and spatiotemporal) reasoning combined with common-sense knowledge, suggesting a promising path for future video understanding. We examine the unique characteristics and capabilities of Vid-LLMs, categorizing the approaches into three main types:Video Analyzer × LLM, Video Embedder × LLM, and (Analyzer + Embedder) × LLM. We identify five subtypes based on the functions of LLMs in Vid-LLMs:LLMas Summarizer,LLMas Manager,LLMas Text Decoder,LLMas Regressor, andLLMas Hidden Layer. This survey also presents a comprehensive study of the tasks, datasets, benchmarks, and evaluation methods for Vid-LLMs. Additionally, it explores the extensive applications of Vid-LLMs in various domains, highlighting their remarkable scalability and versatility in real-world video understanding challenges. Additionally, it summarizes the limitations of existing Vid-LLMs and outlines directions for future research. For more information, readers are encouraged to visit the repository at https://github.com/yunlong10/Awesome-LLMs-for-Video-Understanding.

Existing methods like Neural Radiation Fields (NeRF) and 3D Gaussian Splatting (3DGS) have made significant strides in facial attribute control such as facial animation and components editing, yet they struggle with fine-grained representation and scalability in dynamic head modeling. To address these limitations, we propose GaussianStyle. a novel framework that integrates the volumetric strengths of 3DGS with the powerful implicit representation of Style-GAN. The GaussianStyle preserves structural information, such as expressions and poses, using Gaussian points, while projecting the implicit volumetric representation into Style-GAN to capture high-frequency details and mitigate the over-smoothing commonly observed in neural texture rendering. Experimental outcomes indicate that our method achieves state-of-the-art performance in reenactment, novel view synthesis, and animation.

Large language models (LLMs) have demonstrated remarkable capabilities in natural language and multimodal domains. By fine-tuning multimodal LLMs with temporal annotations from well-annotated datasets, e.g., dense video captioning datasets, their temporal understanding capacity in video-language tasks can be obtained. However, there is a notable lack of untrimmed audio-visual video datasets with precise temporal annotations for events. This deficiency hinders LLMs from learning the alignment between time, audio-visual events, and text tokens, thus impairing their ability to localize audio-visual events in videos temporally. To address this gap, we introduce PU-VALOR, a comprehensive audio-visual dataset comprising over 114,081 pseudo-untrimmed videos with detailed temporal annotations. PU-VALOR is derived from the large-scale but coarse-annotated audio-visual dataset VALOR, through a subtle method involving event-based video clustering, random temporal scaling, and permutation. By fine-tuning a multimodal LLM on PU-VALOR, we developed AVicuna, a model capable of aligning audio-visual events with temporal intervals and corresponding text tokens. AVicuna excels in temporal localization and time-aware dialogue capabilities. Our experiments demonstrate that AVicuna effectively handles temporal understanding in audio-visual videos and achieves state-of-the-art performance on open-ended video QA, audio-visual QA, and audio-visual event dense localization tasks.

Video summarization aims to create short, accurate, and cohesive summaries of longer videos. Despite the existence of various video summarization datasets, a notable limitation is their limited amount of source videos, which hampers the effective training of advanced large vision-language models (VLMs). Additionally, most existing datasets are created for video-to-video summarization, overlooking the contemporary need for multimodal video content summarization. Recent efforts have been made to expand from unimodal to multimodal video summarization, categorizing the task into three sub-tasks based on the summary's modality: video-to-video (V2V), video-to-text (V2T), and a combination of video and text summarization (V2VT). However, the textual summaries in previous multimodal datasets are inadequate. To address these issues, we introduce Instruct-V2Xum, a cross-modal video summarization dataset featuring 30,000 diverse videos sourced from YouTube, with lengths ranging from 40 to 940 seconds and an average summarization ratio of 16.39%. Each video summary in Instruct-V2Xum is paired with a textual summary that references specific frame indexes, facilitating the generation of aligned video and textual summaries. In addition, we propose a new video summarization framework named V2Xum-LLM. V2Xum-LLM, specifically V2Xum-LLaMA in this study, is the first framework that unifies different video summarization tasks into one large language model's (LLM) text decoder and achieves task-controllable video summarization with temporal prompts and task instructions. Experiments show that V2Xum-LLaMA outperforms strong baseline models on multiple video summarization tasks. Furthermore, we propose an enhanced evaluation metric for V2V and V2VT summarization tasks.

Video saliency prediction aims to identify the regions in a video that attract human attention and gaze, driven by bottom-up features from the video and top-down processes like memory and cognition. Among these top-down influences, language plays a crucial role in guiding attention by shaping how visual information is interpreted. Existing methods primarily focus on modeling perceptual information while neglecting the reasoning process facilitated by language, where ranking cues are crucial outcomes of this process and practical guidance for saliency prediction. In this paper, we propose CaRDiff (Caption, Rank, and generate with Diffusion), a framework that imitates the process by integrating multimodal large language model (MLLM), a grounding module, and a diffusion model, to enhance video saliency prediction. Specifically, we introduce a novel prompting method VSOR-CoT (Video Slient Object Ranking Chain of Thought), which utilizes an MLLM with a grounding module to caption video content and infer salient objects along with their rankings and positions. This process derives ranking maps that can be sufficiently leveraged by the diffusion model to accurately decode the saliency maps for the given video. Extensive experiments showcase the effectiveness of VSOR-CoT in improving the performance of video saliency prediction. The proposed CaRDiff performs better than state-of-the-art models on the MVS dataset and demonstrates cross-dataset capabilities on the DHF1k dataset through zero-shot evaluation.

Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated remarkable progress in visual understanding. This impressive leap raises a compelling question: how can language models, initially trained solely on linguistic data, effectively interpret and process visual content? This paper aims to address this question with systematic investigation across 4 model families and 4 model scales, uncovering a unique class of attention heads that focus specifically on visual content. Our analysis reveals a strong correlation between the behavior of these attention heads, the distribution of attention weights, and their concentration on visual tokens within the input. These findings enhance our understanding of how LLMs adapt to multimodal tasks, demonstrating their potential to bridge the gap between textual and visual understanding. This work paves the way for the development of AI systems capable of engaging with diverse modalities.

The advancement of Multimodal Large Language Models (MLLMs) has enabled significant progress in multi-modal understanding, expanding their capacity to analyze video content. However, existing evaluation benchmarks for MLLMs primarily focus on abstract video comprehension, lacking a detailed assessment of their ability to understand video compositions, the nuanced interpretation of how visual elements combine and interact within highly compiled video contexts. We introduce VidComposition, a new benchmark specifically designed to evaluate the video composition understanding capabilities of MLLMs using carefully curated compiled videos and cinematic-level annotations. VidComposition includes 982 videos with 1706 multiple-choice questions, covering various compositional aspects such as camera movement, angle, shot size, narrative structure, character actions and emotions, etc. Our comprehensive evaluation of 33 open-source and proprietary MLLMs reveals a significant performance gap between human and model capabilities. This highlights the limitations of current MLLMs in understanding complex, compiled video compositions and offers insights into areas for further improvement. Our benchmark is publicly available at https://yunlong10.github.io/VidComposition/.

Recent years have seen a significant increase in video content creation and consumption. Crafting engaging content requires the careful curation of both visual and audio elements. While visual cue curation, through techniques like optimal viewpoint selection or post-editing, has been central to media production, its natural counterpart, audio, has not undergone equivalent advancements. This often results in a disconnect between visual and acoustic saliency. To bridge this gap, we introduce a novel task: visually-guided acoustic highlighting, which aims to transform audio to deliver appropriate highlighting effects guided by the accompanying video, ultimately creating a more harmonious audio-visual experience. We propose a flexible, transformer-based multimodal framework to solve this task. To train our model, we also introduce a new dataset—THE MUDDY MIX DATASET, leveraging the meticulous audio and video crafting found in movies, which provides a form of free supervision. We develop a pseudo-data generation process to simulate poorly mixed audio, mimicking real-world scenarios through a three-step process—separation, adjustment, and remixing. Our approach consistently outperforms several baselines in both quantitative and subjective evaluation. We also systematically study the impact of different types of contextual guidance and difficulty levels of the dataset. Our project page is here: https://wikichao.github.io/VisAH/.

Foundation models (FMs) such as CLIP have demonstrated impressive zero-shot performance across various tasks by leveraging large-scale, unsupervised pre-training. However, they often inherit harmful or unwanted knowledge from noisy internet-sourced datasets, compromising their reliability in real-world applications. Existing model unlearning methods either rely on access to pre-trained datasets or focus on coarse-grained unlearning (e.g., entire classes), leaving a critical gap for fine-grained unlearning. In this paper, we address the challenging scenario of selectively forgetting specific portions of knowledge within a class—without access to pre-trained data—while preserving the model’s overall performance. We propose a novel three-stage approach that progressively unlearns targeted knowledge while mitigating over-forgetting. It consists of (1) a forgetting stage to fine-tune the CLIP on samples to be forgotten, (2) a reminding stage to restore performance on retained samples, and (3) a restoring stage to recover zero-shot capabilities using model souping. Additionally, we introduce knowledge distillation to handle the distribution disparity between forgetting/retaining samples and unseen pre-trained data. Extensive experiments on CIFAR-10, ImageNet-1K, and style datasets demonstrate that our approach effectively unlearns specific subgroups while maintaining strong zero-shot performance on semantically similar subgroups and other categories, significantly outperforming baseline unlearning methods, which lose effectiveness under the CLIP unlearning setting.

Generating full-body human gestures based on speech signals remains challenges on quality and speed. Existing approaches model different body regions such as body, legs and hands separately, which fail to capture the spatial interactions between them and result in unnatural and disjointed movements. Additionally, their autoregressive/diffusion-based pipelines show slow generation speed due to dozens of inference steps. To address these two challenges, we propose GestureLSM, a flow-matching-based approach for Co-Speech Gesture Generation with spatial-temporal modeling. Our method i) explicitly model the interaction of tokenized body regions through spatial and temporal attention, for generating coherent full-body gestures. ii) introduce the flow matching to enable more efficient sampling by explicitly modeling the latent velocity space. To overcome the suboptimal performance of flow matching baseline, we propose latent shortcut learning and beta distribution time stamp sampling during training to enhance gesture synthesis quality and accelerate inference. Combining the spatial-temporal modeling and improved flow matching-based framework, GestureLSM achieves state-of-the-art performance on BEAT2 while significantly reducing inference time compared to existing methods, highlighting its potential for enhancing digital humans and embodied agents in real-world applications. Project Page: https://andypinxinliu.github.io/GestureLSM

While audio-visual learning equips models with a richer understanding of the real world by leveraging multiple sensory modalities, this integration also introduces new vulnerabilities to adversarial attacks. In this paper, we present a comprehensive study of the adversarial robustness of audio-visual models, considering both temporal and modality-specific vulnerabilities. We propose two powerful adversarial attacks: 1) a temporal invariance attack that exploits the inherent temporal redundancy across consecutive time segments and 2) a modality misalignment attack that introduces incongruence between the audio and visual modalities. These attacks are designed to thoroughly assess the robustness of audio-visual models against diverse threats. Furthermore, to defend against such attacks, we introduce a novel audio-visual adversarial training framework. This framework addresses key challenges in vanilla adversarial training by incorporating efficient adversarial perturbation crafting tailored to multi-modal data and an adversarial curriculum strategy. Extensive experiments in the Kinetics-Sounds dataset demonstrate that our proposed temporal and modality-based attacks in degrading model performance can achieve state-of-the-art performance, while our adversarial training defense largely improves the adversarial robustness as well as the adversarial training efficiency.

Binaural rendering aims to synthesize binaural audio that mimics natural hearing based on a mono audio and the locations of the speaker and listener. Although many methods have been proposed to solve this problem, they struggle with rendering quality and streamable inference. Synthesizing high-quality binaural audio that is indistinguishable from real-world recordings requires precise modeling of binaural cues, room reverb, and ambient sounds. Additionally, real-world applications demand streaming inference. To address these challenges, we propose a flow matching based streaming binaural speech synthesis framework called BinauralFlow. We consider binaural rendering to be a generation problem rather than a regression problem and design a conditional flow matching model to render high-quality audio. Moreover, we design a causal U-Net architecture that estimates the current audio frame solely based on past information to tailor generative models for streaming inference. Finally, we introduce a continuous inference pipeline incorporating streaming STFT/ISTFT operations, a buffer bank, a midpoint solver, and an early skip schedule to improve rendering continuity and speed. Quantitative and qualitative evaluations demonstrate the superiority of our method over SOTA approaches. A perceptual study further reveals that our model is nearly indistinguishable from real-world recordings, with a 42% confusion rate.

Audio source separation is fundamental for machines to understand complex acoustic environments and underpins numerous audio applications. Current supervised deep learning approaches, while powerful, are limited by the need for extensive, task-specific labeled data and struggle to generalize to the immense variability and open-set nature of real-world acoustic scenes. Inspired by the success of generative foundation models, we investigate whether pre-trained text-guided audio diffusion models can overcome these limitations. We make a surprising discovery: zero-shot source separation can be achieved purely through a pre-trained text-guided audio diffusion model under the right configuration. Our method, named ZeroSep, works by inverting the mixed audio into the diffusion model's latent space and then using text conditioning to guide the denoising process to recover individual sources. Without any task-specific training or fine-tuning, ZeroSep repurposes the generative diffusion model for a discriminative separation task and inherently supports open-set scenarios through its rich textual priors. ZeroSep is compatible with a variety of pre-trained text-guided audio diffusion backbones and delivers strong separation performance on multiple separation benchmarks, surpassing even supervised methods.

Vision Transformers (ViTs) have demonstrated impressive performance across a range of applications, including many safety-critical tasks. Many previous studies have observed that adversarial examples crafted on ViTs exhibit higher transferability than those crafted on CNNs, indicating that ViTs contain structural characteristics favorable for transferable attacks. In this work, we take a further step to deeply investigate the role of computational redundancy brought by its unique characteristics in ViTs and its impact on adversarial transferability. Specifically, we identify two forms of redundancy, including the data-level and model-level, that can be harnessed to amplify attack effectiveness. Building on this insight, we design a suite of techniques, including attention sparsity manipulation, attention head permutation, clean token regularization, ghost MoE diversification, and learn to robustify before the attack. A dynamic online learning strategy is also proposed to fully leverage these operations to enhance the adversarial transferability. Extensive experiments on the ImageNet-1k dataset validate the effectiveness of our approach, showing that our methods significantly outperform existing baselines in both transferability and generality across diverse model architectures, including different variants of ViTs and mainstream Vision Large Language Models (VLLMs).

Understanding perspective is fundamental to human visual perception, yet the extent to which multimodal large language models (MLLMs) internalize perspective geometry remains unclear. We introduce MMPerspective, the first benchmark specifically designed to systematically evaluate MLLMs' understanding of perspective through 10 carefully crafted tasks across three complementary dimensions: Perspective Perception, Reasoning, and Robustness. Our benchmark comprises 2,711 real-world and synthetic image instances with 5,083 question-answer pairs that probe key capabilities, such as vanishing point perception and counting, perspective type reasoning, line relationship understanding in 3D space, invariance to perspective-preserving transformations, etc. Through a comprehensive evaluation of 43 state-of-the-art MLLMs, we uncover significant limitations: while models demonstrate competence on surface-level perceptual tasks, they struggle with compositional reasoning and maintaining spatial consistency under perturbations. Our analysis further reveals intriguing patterns between model architecture, scale, and perspective capabilities, highlighting both robustness bottlenecks and the benefits of chain-of-thought prompting. MMPerspective establishes a valuable testbed for diagnosing and advancing spatial understanding in vision-language systems. Resources are available at https://yunlong10.github.io/MMPerspective/

Machine learning models can perform well on indistribution data but often fail on biased subgroups that are underrepresented in the training data, hindering the robustness of models for reliable applications. Such subgroups are typically unknown due to the absence of subgroup labels. Discovering biased subgroups is the key to understanding models' failure modes and further improving models' robustness. Most previous works of subgroup discovery make an implicit assumption that models only underperform on a single biased subgroup, which does not hold on in-the-wild data where multiple biased subgroups exist. In this work, we propose Decomposition, Interpretation, and Mitigation (DIM), a novel method to address a more challenging but also more practical problem of discovering multiple biased subgroups in image classifiers. Our approach decomposes the image features into multiple components that represent multiple subgroups. This decomposition is achieved via a bilinear dimension reduction method, Partial Least Square (PLS), guided by useful supervision from the image classifier. We further interpret the semantic meaning of each subgroup component by generating natural language descriptions using vision-language foundation models. Finally, DIM mitigates multiple biased subgroups simultaneously via two strategies, including the data and model-centric strategies. Extensive experiments on CIFAR-100 and Breeds datasets demonstrate the effectiveness of DIM in discovering and mitigating multiple biased subgroups. Furthermore, DIM uncovers the failure modes of the classifier on Hard ImageNet, showcasing its broader applicability to understanding model bias in image classifiers. The code is available at https://github.com/ZhangAIPI/DIM.

Adversarial examples, crafted by adding perturbations imperceptible to humans, can deceive neural networks. Re-cent studies identify the adversarial transferability across various models, i.e., the cross-model attack ability of adversarial samples. To enhance such adversarial transferability, existing input transformation-based methods diversify input data with transformation augmentation. However, their effectiveness is limited by the finite number of available transformations. In our study, we introduce a novel approach named Learning to Transform (L2T). L2T increases the diversity of transformed images by selecting the optimal combination of operations from a pool of candidates, consequently im-proving adversarial transferability. We conceptualize the selection of optimal transformation combinations as a tra-jectory optimization problem and employ a reinforcement learning strategy to effectively solve the problem. Comprehensive experiments on the ImageNet dataset, as well as practical tests with Google Vision and GPT-4V, reveal that L2T surpasses current methodologies in enhancing adversarial transferability, thereby confirming its effectiveness and practical significance. The code is available at https://github.com/ZhangAIPI/TransferAttack.

The one-shot talking-head generation learns to synthesize a talking-head video with one source portrait image under the driving of same or different identity video. Usually these methods require plane-based pixel transformations via Jacobin matrices or facial image warps for novel poses generation. The constraints of using a single image source and pixel displacements often compromise the clarity of the synthesized images. Some methods try to improve the quality of synthesized videos by introducing additional super-resolution modules, but this will undoubtedly increase computational consumption and destroy the original data distribution. In this work, we propose an adaptive high-quality talking-head video generation method, which synthesizes high-resolution video without additional pre-trained modules. Specifically, inspired by existing super-resolution methods, we down-sample the one-shot source image, and then adaptively reconstruct high-frequency details via an encoder-decoder module, resulting in enhanced video clarity. Our method consistently improves the quality of generated videos through a straightforward yet effective strategy, substantiated by quantitative and qualitative evaluations. The code and demo video are available on: https://github.com/Songluchuan/AdaSR-TalkingHead/.

Conventional audio classification relied on predefined classes, lacking the ability to learn from free-form text. Recent methods unlock learning joint audio-text embeddings from raw audio-text pairs describing audio in natural language. Despite recent advancements, there is little exploration of systematic methods to train models for recognizing sound events and sources in alternative scenarios, such as distinguishing fireworks from gunshots at outdoor events in similar situations. This study introduces causal reasoning and counterfactual analysis in the audio domain. We use counterfactual instances and include them in our model across different aspects. Our model considers acoustic characteristics and sound source information from human-annotated reference texts. To validate the effectiveness of our model, we conducted pre-training utilizing multiple audio captioning datasets. We then evaluate with several common downstream tasks, demonstrating the merits of the proposed method as one of the first works leveraging counterfactual information in audio domain. Specifically, the top-1 accuracy in open-ended language-based audio retrieval task increased by more than 43%.

While backpropagation (BP) is the mainstream approach for gradient computation in neural network training, its heavy reliance on the chain rule of differentiation constrains the designing flexibility of network architecture and training pipelines. We avoid the recursive computation in BP and develop a unified likelihood ratio (ULR) method for gradient estimation with only one forward propagation. Not only can ULR be extended to train a wide variety of neural network architectures, but the computation flow in BP can also be rearranged by ULR for better device adaptation. Moreover, we propose several variance reduction techniques to further accelerate the training process. Our experiments offer numerical results across diverse aspects, including various neural network training scenarios, computation flow rearrangement, and fine-tuning of pre-trained models. All findings demonstrate that ULR effectively enhances the flexibility of neural network training by permitting localized module training without compromising the global objective and significantly boosts the network robustness.

The rapid evolution of egocentric video analysis brings new insights into understanding human activities and intentions from a first-person perspective. Despite this progress, the fragmentation in tasks like action recognition, procedure learning, and moment retrieval, \etc, coupled with inconsistent annotations and isolated model development, hinders a holistic interpretation of video content. In response, we introduce the EAGLE (Egocentric AGgregated Language-video Engine) model and the EAGLE-400K dataset to provide a unified framework that integrates various egocentric video understanding tasks. EAGLE-400K, the \textit{first} large-scale instruction-tuning dataset tailored for egocentric video, features 400K diverse samples to enhance a broad spectrum of tasks from activity recognition to procedure knowledge learning. Moreover, EAGLE, a strong video multimodal large language model (MLLM), is designed to effectively capture both spatial and temporal information. In addition, we propose a set of evaluation metrics designed to facilitate a thorough assessment of MLLM for egocentric video understanding. Our extensive experiments demonstrate EAGLE's superior performance over existing models, highlighting its ability to balance task-specific understanding with holistic video interpretation. With EAGLE, we aim to pave the way for research opportunities and practical applications in real-world scenarios.

To increase the generalization capability of VQA systems, many recent studies have tried to de-bias spurious language or vision associations that shortcut the question or image to the answer. Despite these efforts, the literature fails to address the confounding effect of vision and language simultaneously. As a result, when they reduce bias learned from one modality, they usually increase bias from the other. In this paper, we first model a confounding effect that causes language and vision bias simultaneously, then propose a counterfactual inference to remove the influence of this effect. The model trained in this strategy can concurrently and efficiently reduce vision and language bias. To the best of our knowledge, this is the first work to reduce biases resulting from confounding effects of vision and language in VQA, leveraging causal explain-away relations. We accompany our method with an explain-away strategy, pushing the accuracy of the questions with numerical answers results compared to existing methods that have been an open problem. The proposed method outperforms the state-of-the-art methods in VQA-CP v2 datasets. R2: Providing brief insights into the experimental setup and results would add valuable context for readers. In response to R2, we released the code and documentation for the implementation as follows. Our codes are available at https://github.com/ali-vosoughi/PW-VQA.

Humans naturally perceive surrounding scenes by unifying sound and sight from a first-person view. Likewise, machines are advanced to approach human intelligence by learning with multisensory inputs from an egocentric perspective. In this paper, we explore the challenging egocentric audio-visual object localization task and observe that 1) egomotion commonly exists in first-person recordings, even within a short duration; 2) The out-of-view sound components can be created when wearers shift their attention. To address the first problem, we propose a geometryaware temporal aggregation module that handles the egomotion explicitly. The effect of egomotion is mitigated by estimating the temporal geometry transformation and exploiting it to update visual representations. Moreover, we propose a cascaded feature enhancement module to over-come the second issue. It improves cross-modal localization robustness by disentangling visually-indicated audio representation. During training, we take advantage of the naturally occurring audio-visual temporal synchronization as the “free” self-supervision to avoid costly labeling. We also annotate and create the Epic Sounding Object dataset for evaluation purposes. Extensive experiments show that our method achieves state-of-the-art localization performance in egocentric videos and can be generalized to diverse audio-visual scenes. Code is available at https://github.com/WikiChao/Ego-AV-Loc.

Machine learning models have been found to learn shortcuts—unintended decision rules that are unable to generalize—undermining models' reliability. Previous works address this problem under the tenuous assumption that only a single shortcut exists in the training data. Real-world images are rife with multiple visual cues from background to texture. Key to advancing the reliability of vision systems is understanding whether existing methods can overcome multiple shortcuts or struggle in a Whac-A-Mole game, i.e., where mitigating one shortcut amplifies reliance on others. To address this shortcoming, we propose two benchmarks: 1) UrbanCars, a dataset with precisely controlled spurious cues, and 2) ImageNet-W, an evaluation set based on ImageNet for watermark, a shortcut we discovered affects nearly every modern vision model. Along with texture and background, ImageNet-W allows us to study multiple shortcuts emerging from training on natural images. We find computer vision models, including large foundation models—regardless of training set, architecture, and supervision—struggle when multiple shortcuts are present. Even methods explicitly designed to combat shortcuts struggle in a Whac-A-Mole dilemma. To tackle this challenge, we propose Last Layer Ensemble, a simple-yet-effective method to mitigate multiple shortcuts without Whac-A-Mole behavior. Our results surface multi-shortcut mitigation as an overlooked challenge critical to advancing the reliability of vision systems. The datasets and code are released: https://github.com/facebookresearch/Whac-A-Mole.

Listener head generation centers on generating non-verbal behaviors (e.g., smile) of a listener in reference to the information delivered by a speaker. A significant challenge when generating such responses is the non-deterministic nature of fine-grained facial expressions during a conversation, which varies depending on the emotions and attitudes of both the speaker and the listener. To tackle this problem, we propose the Emotional Listener Portrait (ELP), which treats each fine-grained facial motion as a composition of several discrete motion-codewords and explicitly models the probability distribution of the motions under different emotion in conversation. Benefiting from the "explicit" and "discrete" design, our ELP model can not only automatically generate natural and diverse responses toward a given speaker via sampling from the learned distribution but also generate controllable responses with a predetermined attitude. Under several quantitative metrics, our ELP exhibits significant improvements compared to previous methods.

Can machines recording an audio-visual scene produce realistic, matching audio-visual experiences at novel positions and novel view directions? We answer it by studying a new task---real-world audio-visual scene synthesis---and a first-of-its-kind NeRF-based approach for multimodal learning. Concretely, given a video recording of an audio-visual scene, the task is to synthesize new videos with spatial audios along arbitrary novel camera trajectories in that scene. We propose an acoustic-aware audio generation module that integrates prior knowledge of audio propagation into NeRF, in which we implicitly associate audio generation with the 3D geometry and material properties of a visual environment. Furthermore, we present a coordinate transformation module that expresses a view direction relative to the sound source, enabling the model to learn sound source-centric acoustic fields. To facilitate the study of this new task, we collect a high-quality Real-World Audio-Visual Scene (RWAVS) dataset. We demonstrate the advantages of our method on this real-world dataset and the simulation-based SoundSpaces dataset. Notably, we refer readers to view our demo videos for convincing comparisons.

Audio-visual learning seeks to enhance the computer’s multi-modal perception leveraging the correlation between the auditory and visual modalities. Despite their many useful downstream tasks, such as video retrieval, AR/VR, and accessibility, the performance and adoption of existing audio-visual models have been impeded by the availability of high-quality datasets. Annotating audio-visual datasets is laborious, expensive, and time-consuming. To address this challenge, we designed and developed an efficient audio-visual annotation tool called Peanut. Peanut’s human-AI collaborative pipeline separates the multi-modal task into two single-modal tasks, and utilizes state-of-the-art object detection and sound-tagging models to reduce the annotators’ effort to process each frame and the number of manually-annotated frames needed. A within-subject user study with 20 participants found that Peanut can significantly accelerate the audio-visual data annotation process while maintaining high annotation accuracy.

We propose the "runtime learning" hypothesis which states that people quickly learn to perform unfamiliar tasks as the tasks arise by using task-relevant instances of concepts stored in memory during mental training. To make learning rapid, the hypothesis claims that only a few class instances are used, but these instances are especially valuable for training. The paper motivates the hypothesis by describing related ideas from the cognitive science and machine learning literatures. Using computer simulation, we show that deep neural networks (DNNs) can learn effectively from small, curated training sets, and that valuable training items tend to lie toward the centers of data item clusters in an abstract feature space. In a series of three behavioral experiments, we show that people can also learn effectively from small, curated training sets. Critically, we find that participant reaction times and fitted drift rates are best accounted for by the confidences of DNNs trained on small datasets of highly valuable items. We conclude that the runtime learning hypothesis is a novel conjecture about the relationship between learning and memory with the potential for explaining a wide variety of cognitive phenomena.

Recently, large pretrained models (e.g., BERT, Style-GAN, CLIP) show great knowledge transfer and generalization capability on various downstream tasks within their domains. Inspired by these efforts, in this paper we propose a unified model for open-domain image editing focusing on color and tone adjustment of open-domain images while keeping their original content and structure. Our model learns a unified editing space that is more semantic, intu-itive, and easy to manipulate than the operation space (e.g., contrast, brightness, color curve) used in many existing photo editing softwares. Our model belongs to the image-to-image translation framework which consists of an image encoder and decoder, and is trained on pairs of before-and-after edited images to produce multimodal outputs. We show that by inverting image pairs into latent codes of the learned editing space, our model can be leveraged for vari-ous downstream editing tasks such as language-guided image editing, personalized editing, editing-style clustering, retrieval, etc. We extensively study the unique properties of the editing space in experiments and demonstrate superior performance on the aforementioned tasks11Code and supplementary material can be found at the project page https://jshi31.github.io/SpaceEdit.

Magnetic resonance imaging (MRI) can present multicontrast images of the same anatomical structures, enabling multi-contrast super-resolution (SR) techniques. Compared with SR reconstruction using a single-contrast, multicontrast SR reconstruction is promising to yield SR images with higher quality by leveraging diverse yet complementary information embedded in different imaging modalities. However, existing methods still have two shortcomings: (1) they neglect that the multi-contrast features at different scales contain different anatomical details and hence lack effective mechanisms to match and fuse these features for better reconstruction; and (2) they are still deficient in capturing long-range dependencies, which are essential for the regions with complicated anatomical structures. We propose a novel network to comprehensively address these problems by developing a set of innovative Transformer-empowered multi-scale contextual matching and aggregation techniques; we call it McMRSR. Firstly, we tame transformers to model long-range dependencies in both reference and target images. Then, a new multi-scale contextual matching method is proposed to capture corresponding contexts from reference features at different scales. Furthermore, we introduce a multi-scale aggregation mechanism to gradually and interactively aggregate multi-scale matched features for reconstructing the target SR MR image. Extensive experiments demonstrate that our network outperforms state-of-the-art approaches and has great potential to be applied in clinical practice. Codes are available at https://github.com/XAIMI-Lab/McMRSR.

Although progress has been made for text-to-image synthesis, previous methods fall short of generalizing to unseen or underrepresented attribute compositions in the input text. Lacking compositionality could have severe implications for robustness and fairness, e.g., inability to synthesize the face images of underrepresented demographic groups. In this paper, we introduce a new framework, StyleT2I, to improve the compositionality of text-to-image synthesis. Specifically, we propose a CLIP-guided Contrastive Loss to better distinguish different compositions among different sentences. To further improve the compositionality, we design a novel Semantic Matching Loss and a Spatial Constraint to identify attributes' latent directions for intended spatial region manipulations, leading to better disentangled latent representations of attributes. Based on the identified latent directions of attributes, we propose Compositional Attribute Adjustment to adjust the latent code, resulting in better compositionality of image synthesis. In addition, we leverage the$l_{2}$-norm regularization of identified latent directions (norm penalty) to strike a nice balance between image-text alignment and image fidelity. In the experiments, we devise a new dataset split and an evaluation metric to evaluate the compositionality of text-to-image synthesis models. The results show that StyleT2I outperforms previous approaches in terms of the consistency between the input text and synthesized images and achieves higher fidelity.

In this paper, we focus on the Audio-Visual Question Answering (AVQA) task, which aims to answer questions regarding different visual objects, sounds, and their associations in videos. The problem requires comprehensive multimodal understanding and spatio-temporal reasoning over audio-visual scenes. To benchmark this task and facilitate our study, we introduce a large-scale MUSIC-AVQA dataset, which contains more than 45K question-answer pairs covering 33 different question templates spanning over different modalities and question types. We develop several baselines and introduce a spatio-temporal grounded audio-visual network for the AVQA problem. Our results demonstrate that AVQA benefits from multisensory perception and our model outperforms recent A-, V-, and AVQA approaches. We believe that our built dataset has the potential to serve as testbed for evaluating and promoting progress in audio-visual scene understanding and spatio-temporal reasoning. Code and dataset: http://gewu-lab.github.io/MUSIC-AVQA/

In this study, we aim to predict the plausible future action steps given an observation of the past and study the task of instructional activity anticipation. Unlike previous anticipation tasks that aim at action label prediction, our work targets at generating natural language outputs that provide interpretable and accurate descriptions of future action steps. It is a challenging task due to the lack of semantic information extracted from the instructional videos. To overcome this challenge, we propose a novel knowledge distillation framework to exploit the related external textual knowledge to assist the visual anticipation task. However, previous knowledge distillation techniques generally transfer information within the same modality. To bridge the gap between the visual and text modalities during the distillation process, we devise a novel cross-modal contrastive distillation (CCD) scheme, which facilitates knowledge distillation between teacher and student in heterogeneous modalities with the proposed cross-modal distillation loss. We evaluate our method on the Tasty Videos dataset. CCD improves the anticipation performance of the visual-alone student model by a large margin of 40.2% relatively in BLEU4. Our approach also outperforms the state-of-the-art approaches by a large margin.

The deep learning-based face forgery detection is a novel yet challenging task. Despite impressive results have been achieved, there are still some limitations in the existing methods. For example, the previous methods are hard to maintain consistent predictions for consecutive frames, even if all of those frames are actually forged. We propose a symmetric transformer for channel and spatial feature extraction, which is because the channel and spatial features of a robust forgery detector should be consistent in the temporal domain. The symmetric transformer adopt the newly-designed attention-based strategies for channel variance and spatial gradients as the vital features, which greatly improves the robustness of deepfake video detection. Moreover, this symmetric structure acts on temporal and spatial features respectively, which ensures the robustness of detection from two different aspects. Our symmetric transformer is an end-to-end optimized network. Experiments are conducted on various settings, the proposed methods achieve significantly improvement on prediction robustness and perform better than state-of-the-art methods on different datasets.

Recently, language-guided global image editing draws increasing attention with growing application potentials. However, previous GAN-based methods are not only confined to domain-specific, low-resolution data but also lacking in interpretability. To overcome the collective difficulties, we develop a text-to-operation model to map the vague editing language request into a series of editing operations, e.g., change contrast, brightness, and saturation. Each operation is interpretable and differentiable. Furthermore, the only supervision in the task is the target image, which is insufficient for a stable training of sequential decisions. Hence, we propose a novel operation planning algorithm to generate possible editing sequences from the target image as pseudo ground truth. Comparison experiments on the newly collected MA5k-Req dataset and GIER dataset show the advantages of our methods. Code is available at https://github.com/jshi31/T2ONet.

3D video avatars can empower virtual communications by providing compression, privacy, entertainment, and a sense of presence in AR/VR. Best 3D photo-realistic AR/VR avatars driven by video, that can minimize uncanny effects, rely on person-specific models. However, existing person-specific photo-realistic 3D models are not robust to lighting, hence their results typically miss subtle facial behaviors and cause artifacts in the avatar. This is a major drawback for the scalability of these models in communication systems (e.g., Messenger, Skype, FaceTime) and AR/VR. This paper addresses previous limitations by learning a deep learning lighting model, that in combination with a high-quality 3D face tracking algorithm, provides a method for subtle and robust facial motion transfer from a regular video to a 3D photo-realistic avatar. Extensive experimental validation and comparisons to other state-of-the-art methods demonstrate the effectiveness of the proposed framework in real-world scenarios with variability in pose, expression, and illumination. Our project page can be found at this website.

There are rich synchronized audio and visual events in our daily life. Inside the events, audio scenes are associated with the corresponding visual objects; meanwhile, sounding objects can indicate and help to separate their individual sounds in the audio track. Based on this observation, in this paper, we propose a cyclic co-learning (CCoL) paradigm that can jointly learn sounding object visual grounding and audio-visual sound separation in a unified framework. Concretely, we can leverage grounded object-sound relations to improve the results of sound separation. Meanwhile, benefiting from discriminative information from separated sounds, we improve training example sampling for sounding object grounding, which builds a co-learning cycle for the two tasks and makes them mutually beneficial. Extensive experiments show that the proposed framework outperforms the compared recent approaches on both tasks, and they can benefit from each other with our cyclic co-learning. The source code and pre-trained models are released in https://github.com/YapengTian/CCOL-CVPR21.

In this paper, we propose to make a systematic study on machines’ multisensory perception under attacks. We use the audio-visual event recognition task against multi-modal adversarial attacks as a proxy to investigate the robustness of audio-visual learning. We attack audio, visual, and both modalities to explore whether audio-visual integration still strengthens perception and how different fusion mechanisms affect the robustness of audio-visual models. For interpreting the multimodal interactions under attacks, we learn a weakly-supervised sound source visual localization model to localize sounding regions in videos. To mitigate multimodal attacks, we propose an audio-visual defense approach based on an audio-visual dissimilarity constraint and external feature memory banks. Extensive experiments demonstrate that audio-visual models are susceptible to multimodal adversarial attacks; audio-visual integration could decrease the model robustness rather than strengthen under multimodal attacks; even a weakly-supervised sound source visual localization model can be successfully fooled; our defense method can improve the invulnerability of audio-visual networks without significantly sacrificing clean model performance. The source code and pre-trained models are released in https://github.com/YapengTian/AV-Robustness-CVPR21.

We investigate the weakly-supervised scene graph generation, which is a challenging task since no correspondence of label and object is provided. The previous work regards such correspondence as a latent variable which is iteratively updated via nested optimization of the scene graph generation objective. However, we further reduce the complexity by decoupling it into an efficient first-order graph matching module optimized via contrastive learning to obtain such correspondence, which is used to train a standard scene graph generation model. The extensive experiments show that such a simple pipeline can significantly surpass the previous state-of-the-art by more than 30% on the Visual Genome dataset, both in terms of graph matching accuracy and scene graph quality. We believe this work serves as a strong baseline for future research. Code is available at https://github.com/jshi31/WS-SGG.

Learning new skills by observing humans’ behaviors is an essential capability of AI. In this work, we leverage instructional videos to study humans’ decision-making processes, focusing on learning a model to plan goal-directed actions in real-life videos. In contrast to conventional action recognition, goal-directed actions are based on expectations of their outcomes requiring causal knowledge of potential consequences of actions. Thus, integrating the environment structure with goals is critical for solving this task. Previous works learn a single world model will fail to distinguish various tasks, resulting in an ambiguous la-tent space; planning through it will gradually neglect the desired outcomes since the global information of the future goal degrades quickly as the procedure evolves. We address these limitations with a new formulation of procedure planning and propose novel algorithms to model human behaviors through Bayesian Inference and model-based Imitation Learning. Experiments conducted on real-world instructional videos show that our method can achieve state-of-the-art performance in reaching the indicated goals. Furthermore, the learned contextual information presents interesting features for planning in a latent space.

We present a simple yet highly generalizable method for explaining interacting parts within a neural network’s reasoning process. First, we design an algorithm based on cross derivatives for computing statistical interaction effects between individual features, which is generalized to both 2-way and higher-order (3-way or more) interactions. We present results side by side with a weight-based attribution technique, corroborating that cross derivatives are a superior metric for both 2-way and higher-order interaction detection. Moreover, we extend the use of cross derivatives as an explanatory device in neural networks to the computer vision setting by expanding Grad-CAM, a popular gradient-based explanatory tool for CNNs, to the higher order. While Grad-CAM can only explain the importance of individual objects in images, our method, which we call Taylor-CAM, can explain a neural network’s relational reasoning across multiple objects. We show the success of our explanations both qualitatively and quantitatively, including with a user study. We will release all code as a tool package to facilitate explainable deep learning.

Recent works find that AI algorithms learn biases from data. Therefore, it is urgent and vital to identify biases in AI algorithms. However, the previous bias identification pipeline overly relies on human experts to conjecture potential biases (e.g., gender), which may neglect other underlying biases not realized by humans. To help human experts better find the AI algorithms’ biases, we study a new problem in this work – for a classifier that predicts a target attribute of the input image, discover its unknown biased attribute.To solve this challenging problem, we use a hyperplane in the generative model’s latent space to represent an image attribute; thus, the original problem is transformed to optimizing the hyperplane’s normal vector and offset. We propose a novel total-variation loss within this framework as the objective function and a new orthogonalization penalty as a constraint. The latter prevents trivial solutions in which the discovered biased attribute is identical with the target or one of the known-biased attributes. Extensive experiments on both disentanglement datasets and real-world datasets show that our method can discover biased attributes and achieve better disentanglement w.r.t. target attributes. Furthermore, the qualitative results show that our method can discover unnoticeable biased attributes for various object and scene classifiers, proving our method’s generalizability for detecting biased attributes in diverse domains of images.

Learning from image-text data has demonstrated recent success for many recognition tasks, yet is currently limited to visual features or individual visual concepts such as objects. In this paper, we propose one of the first methods that learn from image-sentence pairs to extract a graphical representation of localized objects and their relationships within an image, known as scene graph. To bridge the gap between images and texts, we leverage an off-the-shelf object detector to identify and localize object instances, match labels of detected regions to concepts parsed from captions, and thus create "pseudo" labels for learning scene graph. Further, we design a Transformer-based model to predict these "pseudo" labels via a masked token prediction task. Learning from only image-sentence pairs, our model achieves 30% relative gain over a latest method trained with human-annotated unlocalized scene graphs. Our model also shows strong results for weakly and fully supervised scene graph generation. In addition, we explore an open-vocabulary setting for detecting scene graphs, and present the first result for open-set scene graph generation.

Weakly Supervised Object Localization (WSOL) methods generate both classification and localization results by learning from only image category labels. Previous methods usually utilize class activation map (CAM) to obtain target object regions. However, most of them only focus on improving foreground object parts in CAM, but ignore the important effect of its background contents. In this paper, we propose a confidence segmentation (ConfSeg) module that builds confidence score for each pixel in CAM without introducing additional hyper-parameters. The generated sample-specific confidence mask is able to indicate the extent of determination for each pixel in CAM, and further supervises additional CAM extended from internal feature maps. Besides, we introduce Co-supervised Augmentation (CoAug) module to capture feature-level representation for foreground and background parts in CAM separately. Then a metric loss is applied at batch sample level to augment distinguish ability of our model, which helps a lot to localize more related object parts. Our final model, CSoA, combines the two modules and achieves superior performance, e.g. 37.69% and 48.81% Top-1 localization error on CUB-200 and ILSVRC datasets, respectively, which outperforms all previous methods and becomes the new state-of-the-art.

Understanding web instructional videos is an essential branch of video understanding in two aspects. First, most existing video methods focus on short-term actions for a-few-second-long video clips; these methods are not directly applicable to long videos. Second, unlike unconstrained long videos, e.g., movies, instructional videos are more structured in that they have step-by-step procedures constraining the understanding task. In this work, we study problem-solving on instructional videos via Visual Question Answering (VQA). Surprisingly, it has not been an emphasis for the video community despite its rich applications. We thereby introduce YouCookQA, an annotated QA dataset for instructional videos based on YouCook2 [27]. The questions in YouCookQA are not limited to cues on a single frame but relations among multiple frames in the temporal dimension. Observing the lack of effective representations for modeling long videos, we propose a set of carefully designed models including a Recurrent Graph Convolutional Network (RGCN) that captures both temporal order and relational information. Furthermore, we study multiple modalities including descriptions and transcripts for the purpose of boosting video understanding. Extensive experiments on YouCookQA suggest that RGCN performs the best in terms of QA accuracy and better performance is gained by introducing human-annotated descriptions. YouCookQA dataset is available at https://github.com/Jossome/YoucookQA.

Some multi-task correlation filter trackers achieve the top-ranked performance in terms of accuracy and robustness. However, they directly fuse multiple types of features into a single kernel space. This operation fails to fully explore the discriminative strength and diversity of different features, and also ignores the structured correspondence of different tasks. To solve these issues, we propose a structured multi-kernel subtask correlation filter tracker with temporal-spatial consistency, which enjoys the merits of both layered multi-kernel subtask learning and structured correlation filter. Specifically, we firstly assign one kernel space to each channel feature. Multi-channel features correspond to multi-kernel spaces to boost their powerful discriminability. And then, we divide the target into multi-layer patches with different sizes, and regard the correlation filter trace of each patch with one channel feature as a subtask. In the following, we incorporate globally and locally structured correlation filters into a unified multi-kernel subtask particle tracking framework. The global and local subtasks complement and enhance each other with similar motion model. The proposed tracker not only exploits the cooperation and complementarity of layered multi-kernel subtask correlation filters, but also mines the underlying geometric structure of global subtasks, and the inner spatial locality correspondences of local subtasks inside the target. This operation is achieved by dual group sparsity regularized terms with mixed-norm lp,q, which decomposes the multi-kernel subtask filter matrix into two collaborative components. They correspond to the adaptive filter feature selection and outlier subtask detection, respectively. Besides, the developed tracking model maintains the temporal coherence and spatial consistency of multi-layer subtask filters via the smooth regularizer. Finally, the tracking formulation is optimized by the accelerated proximal gradient approach (APG). Encouraging analyses on six benchmark datasets, verify the favorable effectiveness and robustness of our method against state-of-the-art trackers.

Pose guided synthesis aims to generate a new image in an arbitrary target pose while preserving the appearance details from the source image. Existing approaches rely on either hard-coded spatial transformations or 3D body modeling. They often overlook complex non-rigid pose deformation or unmatched occluded regions, thus fail to effectively preserve appearance information. In this article, we propose a pose flow learning scheme that learns to transfer the appearance details from the source image without resorting to annotated correspondences. Based on such learned pose flow, we proposed GarmentNet and SynthesisNet, both of which use multi-scale feature-domain alignment for coarse-to-fine synthesis. Experiments on the DeepFashion, MVC dataset and additional real-world datasets demonstrate that our approach compares favorably with the state-of-the-art methods and generalizes to unseen poses and clothing styles.

Learning from Demonstrations (LfD) via Behavior Cloning (BC) works well on multiple complex tasks. However, a limitation of the typical LfD approach is that it requires expert demonstrations for all scenarios, including those in which the algorithm is already well-trained. The recently proposed Learning from Interventions (LfI) overcomes this limitation by using an expert overseer. The expert overseer only intervenes when it suspects that an unsafe action is about to be taken. Although LfI significantly improves over LfD, the state-of-the-art LfI fails to account for delay caused by the expert's reaction time and only learns short-term behavior. We address these limitations by 1) interpolating the expert's interventions back in time, and 2) by splitting the policy into two hierarchical levels, one that generates sub-goals for the future and another that generates actions to reach those desired sub-goals. This sub-goal prediction forces the algorithm to learn long-term behavior while also being robust to the expert's reaction time. Our experiments show that LfI using sub-goals in a hierarchical policy framework trains faster and achieves better asymptotic performance than typical LfD.

We address weakly-supervised video actor-action segmentation (VAAS), which extends general video object segmentation (VOS) to additionally consider action labels of the actors. The most successful methods on VOS synthesize a pool of pseudo-annotations (PAs) and then refine them iteratively. However, they face challenges as to how to select from a massive amount of PAs high-quality ones, how to set an appropriate stop condition for weakly-supervised training, and how to initialize PAs pertaining to VAAS. To overcome these challenges, we propose a general Weakly-Supervised framework with a Wise Selection of training samples and model evaluation criterion (WS2). Instead of blindly trusting quality-inconsistent PAs, WS2employs a learning-based selection to select effective PAs and a novel region integrity criterion as a stopping condition for weakly-supervised training. In addition, a 3D-Conv GCAM is devised to adapt to the VAAS task. Extensive experiments show that WS2achieves state-of-the-art performance on both weakly-supervised VOS and VAAS tasks and is on par with the best fully-supervised method on VAAS.

The perceptual-based grouping process produces a hierarchical and compositional image representation that helps both human and machine vision systems recognize heterogeneous visual concepts. Examples can be found in the classical hierarchical superpixel segmentation or image parsing works. However, the grouping process is largely overlooked in modern CNN-based image segmentation networks due to many challenges, including the inherent incompatibility between the grid-shaped CNN feature map and the irregular-shaped perceptual grouping hierarchy. Overcoming these challenges, we propose a deep grouping model (DGM) that tightly marries the two types of representations and defines a bottom-up and a top-down process for feature exchanging. When evaluating the model on the recent Broden+ dataset for the unified perceptual parsing task, it achieves state-of-the-art results while having a small computational overhead compared to other contextual-based segmentation models. Furthermore, the DGM has better interpretability compared with modern CNN methods.

Video super-resolution (VSR) aims to restore a photo-realistic high-resolution (HR) video frame from both its corresponding low-resolution (LR) frame (reference frame) and multiple neighboring frames (supporting frames). Due to varying motion of cameras or objects, the reference frame and each support frame are not aligned. Therefore, temporal alignment is a challenging yet important problem for VSR. Previous VSR methods usually utilize optical flow between the reference frame and each supporting frame to warp the supporting frame for temporal alignment. However, both inaccurate flow and the image-level warping strategy will lead to artifacts in the warped supporting frames. To overcome the limitation, we propose a temporally-deformable alignment network (TDAN) to adaptively align the reference frame and each supporting frame at the feature level without computing optical flow. The TDAN uses features from both the reference frame and each supporting frame to dynamically predict offsets of sampling convolution kernels. By using the corresponding kernels, TDAN transforms supporting frames to align with the reference frame. To predict the HR video frame, a reconstruction network taking aligned frames and the reference frame is utilized. Experimental results demonstrate that the TDAN is capable of alleviating occlusions and artifacts for temporal alignment and the TDAN-based VSR model outperforms several recent state-of-the-art VSR networks with a comparable or even much smaller model size. The source code and pre-trained models are released in https://github.com/YapengTian/TDAN-VSR.

In this paper, we explore the space-time video super-resolution task, which aims to generate a high-resolution (HR) slow-motion video from a low frame rate (LFR), low-resolution (LR) video. A simple solution is to split it into two sub-tasks: video frame interpolation (VFI) and video super-resolution (VSR). However, temporal interpolation and spatial super-resolution are intra-related in this task. Two-stage methods cannot fully take advantage of the natural property. In addition, state-of-the-art VFI or VSR networks require a large frame-synthesis or reconstruction module for predicting high-quality video frames, which makes the two-stage methods have large model sizes and thus be time-consuming. To overcome the problems, we propose a one-stage space-time video super-resolution framework, which directly synthesizes an HR slow-motion video from an LFR, LR video. Rather than synthesizing missing LR video frames as VFI networks do, we firstly temporally interpolate LR frame features in missing LR video frames capturing local temporal contexts by the proposed feature temporal interpolation network. Then, we propose a deformable ConvLSTM to align and aggregate temporal information simultaneously for better leveraging global temporal contexts. Finally, a deep reconstruction network is adopted to predict HR slow-motion video frames. Extensive experiments on benchmark datasets demonstrate that the proposed method not only achieves better quantitative and qualitative performance but also is more than three times faster than recent two-stage state-of-the-art methods, e.g., DAIN+EDVR and DAIN+RBPN.

Acoustic cues are not the only component in speech communication; if the visual counterpart is present, it is shown to benefit speech comprehension. In this work, we propose an end-to-end (no pre- or post-processing) system that can generate talking faces from arbitrarily long noisy speech. We propose a mouth region mask to encourage the network to focus on mouth movements rather than speech irrelevant movements. In addition, we use generative adversarial network (GAN) training to improve the image quality and mouth-speech synchronization. Furthermore, we employ noise-resilient training to make our network robust to unseen non-stationary noise. We evaluate our system with image quality and mouth shape (landmark) measures on noisy speech utterances with five types of unseen non-stationary noise between -10 dB and 30 dB signal-to-noise ratio (SNR) with increments of 1 dB SNR. Results show that our system outperforms a state-of-the-art baseline system significantly, and our noise-resilient training improves performance for noisy speech in a wide range of SNR.

Attribute editing has become an important and emerging topic of computer vision. In this paper, we consider a task: given a reference garment image A and another image B with target attribute (collar/sleeve), generate a photorealistic image which combines the texture from reference A and the new attribute from reference B. The highly convoluted attributes and the lack of paired data are the main challenges to the task. To overcome those limitations, we propose a novel self-supervised model to synthesize garment images with disentangled attributes (e.g., collar and sleeves) without paired data. Our method consists of a reconstruction learning step and an adversarial learning step. The model learns texture and location information through reconstruction learning. And, the model's capability is generalized to achieve single-attribute manipulation by adversarial learning. Meanwhile, we compose a new dataset, named GarmentSet, with annotation of landmarks of collars and sleeves on clean garment images. Extensive experiments on this dataset and real-world samples demonstrate that our method can synthesize much better results than the state-of-the-art methods in both quantitative and qualitative comparisons.

Visual cues such as lip movements, when available, play an important role in speech communication. They are especially helpful for the hearing impaired population or in noisy environments. When not available, having a system to automatically generate talking faces in sync with input speech would enhance speech communication and enable many novel applications. In this article, we present a new system that can generate 3D talking face landmarks from speech in an online fashion. We employ a neural network that accepts the raw waveform as an input. The network contains convolutional layers with 1D kernels and outputs the active shape model (ASM) coefficients of face landmarks. To promote smoother transitions between video frames, we present a variant of the model that has the same architecture but also accepts the previous frame's ASM coefficients as an additional input. To cope with background noise, we propose a new training method to incorporate speech enhancement ideas at the feature level. Objective evaluations on landmark prediction show that the proposed system yields statistically significantly smaller errors than two state-of-the-art baseline methods on both a single-speaker dataset and a multi-speaker dataset. Experiments on noisy speech input with five types of non-stationary unseen noise show statistically significant improvements of the system performance thanks to the noise-resilient training method. Finally, subjective evaluations show that the generated talking faces have a significantly more convincing match with the input audio, achieving a similarly convincing level of realism as the ground-truth landmarks.

We invest the problem of weakly-supervised video grounding, where only video-level sentences are provided. This is a challenging task, and previous Multi-Instance Learning (MIL) based image grounding methods turn to fail in the video domain. Recent work attempts to decompose the video-level MIL into frame-level MIL by applying weighted sentence-frame ranking loss over frames, but it is not robust and does not exploit the rich temporal information in videos. In this work, we address these issues by extending frame-level MIL with a false positive frame-bag constraint and modeling the visual feature consistency in the video. In specific, we design a contextual similarity between semantic and visual features to deal with sparse objects association across frames. Furthermore, we leverage temporal coherence by strengthening the clustering effect of similar features in the visual space. We conduct an extensive evaluation on YouCookII and RoboWatch datasets, and demonstrate our method significantly outperforms prior state-of-the-art methods.

We devise a cascade GAN approach to generate talking face video, which is robust to different face shapes, view angles, facial characteristics, and noisy audio conditions. Instead of learning a direct mapping from audio to video frames, we propose first to transfer audio to high-level structure, i.e., the facial landmarks, and then to generate video frames conditioned on the landmarks. Compared to a direct audio-to-image approach, our cascade approach avoids fitting spurious correlations between audiovisual signals that are irrelevant to the speech content. We, humans, are sensitive to temporal discontinuities and subtle artifacts in video. To avoid those pixel jittering problems and to enforce the network to focus on audiovisual-correlated regions, we propose a novel dynamically adjustable pixel-wise loss with an attention mechanism. Furthermore, to generate a sharper image with well-synchronized facial movements, we propose a novel regression-based discriminator structure, which considers sequence-level information along with frame-level information. Thoughtful experiments on several datasets and real-world samples demonstrate significantly better results obtained by our method than the state-of-the-art methods in both quantitative and qualitative comparisons.

Expectation maximization (EM) algorithm is to find maximum likelihood solution for models having latent variables. A typical example is Gaussian Mixture Model (GMM) which requires Gaussian assumption, however, natural images are highly non-Gaussian so that GMM cannot be applied to perform image clustering task on pixel space. To overcome such limitation, we propose a GAN based EM learning framework that can maximize the likelihood of images and estimate the latent variables. We call this model GAN-EM, which is a framework for image clustering, semi-supervised classification and dimensionality reduction. In M-step, we design a novel loss function for discriminator of GAN to perform maximum likelihood estimation (MLE) on data with soft class label assignments. Specifically, a conditional generator captures data distribution for K classes, and a discriminator tells whether a sample is real or fake for each class. Since our model is unsupervised, the class label of real data is regarded as latent variable, which is estimated by an additional network (E-net) in E-step. The proposed GAN-EM achieves state-of-the-art clustering and semi-supervised classification results on MNIST, SVHN and CelebA, as well as comparable quality of generated images to other recently developed generative models.

The potential for agents, whether embodied or software, to learn by observing other agents performing procedures involving objects and actions is rich. Current research on automatic procedure learning heavily relies on action labels or video subtitles, even during the evaluation phase, which makes them infeasible in real-world scenarios. This leads to our question: can the human-consensus structure of a procedure be learned from a large set of long, unconstrained videos (e.g., instructional videos from YouTube) with only visual evidence? To answer this question, we introduce the problem of procedure segmentation---to segment a video procedure into category-independent procedure segments. Given that no large-scale dataset is available for this problem, we collect a large-scale procedure segmentation dataset with procedure segments temporally localized and described; we use cooking videos and name the dataset YouCook2. We propose a segment-level recurrent network for generating procedure segments by modeling the dependencies across segments. The generated segments can be used as pre-processing for other tasks, such as dense video captioning and event parsing. We show in our experiments that the proposed model outperforms competitive baselines in procedure segmentation.

In this work, we address the task of weakly-supervised human action segmentation in long, untrimmed videos. Recent methods have relied on expensive learning models, such as Recurrent Neural Networks (RNN) and Hidden Markov Models (HMM). However, these methods suffer from expensive computational cost, thus are unable to be deployed in large scale. To overcome the limitations, the keys to our design are efficiency and scalability. We propose a novel action modeling framework, which consists of a new temporal convolutional network, named Temporal Convolutional Feature Pyramid Network (TCFPN), for predicting frame-wise action labels, and a novel training strategy for weakly-supervised sequence modeling, named Iterative Soft Boundary Assignment (ISBA), to align action sequences and update the network in an iterative fashion. The proposed framework is evaluated on two benchmark datasets, Breakfast and Hollywood Extended, with four different evaluation metrics. Extensive experimental results show that our methods achieve competitive or superior performance to state-of-the-art methods.

As an important complement to person re-identification, text-based person search in large-scale database is concerned greatly for person search applications. Given language description of a person, existing frameworks search the images in the dataset that describe the same person, by computing the affinity score between the description and each image. In this paper, we first propose an efficient patch-word matching model, which can accurately capture the local matching details between image and text. In particular, it computes the affinity between an image and a word as the affinity of the best matching patch of the image toward the word. Compared with the state-of-the-art framework, it achieves competitive performance, but yields lowcomplexity structure. In addition, we put forward a significant limitation of affinity-based model, it is overly sensitive to the matching degree of a corresponding image-word pair. For this limitation, we feed a creative adaptive threshold mechanism into the model, it automatically learns an adaptive threshold for each word, and effectively “compress” the affinity score between a word and an image when the score exceeds the words threshold. Extensive experiments on the benchmark dataset demonstrate the effectiveness of the proposed framework, which outperforms other approaches for text-based person search. To provide a deeper insight into the proposed model, we visualize the matching details between spatial patches of images and words of texts on typical examples, and illustrate how adaptive threshold mechanism compresses the affinity score and benefits the final rank of different images toward a text description.

Fine-grained activity understanding in videos has attracted considerable recent attention with a shift from action classification to detailed actor and action understanding that provides compelling results for perceptual needs of cutting-edge autonomous systems. However, current methods for detailed understanding of actor and action have significant limitations: they require large amounts of finely labeled data, and they fail to capture any internal relationship among actors and actions. To address these issues, in this paper, we propose a novel, robust multi-task ranking model for weakly supervised actor-action segmentation where only video-level tags are given for training samples. Our model is able to share useful information among different actors and actions while learning a ranking matrix to select representative supervoxels for actors and actions respectively. Final segmentation results are generated by a conditional random field that considers various ranking scores for different video parts. Extensive experimental results on the Actor-Action Dataset (A2D) demonstrate that the proposed approach outperforms the state-of-the-art weakly supervised methods and performs as well as the top-performing fully supervised method.

Dance is a unique and meaningful type of human expression, composed of abundant and various action elements. However, existing methods based on associated texts and spatial visual features have difficulty capturing the highly articulated motion patterns. To overcome this limitation, we propose to take advantage of the intrinsic motion information in dance videos to solve the video recommendation problem. We present a novel system that recommends dance videos based on a mid-level action representation, termed Dancelets. The Dancelets are used to bridge the semantic gap between video content and high-level concept, dance style, which plays a significant role in characterizing different types of dances. The proposed method executes automatic mining of dancelets with a concatenation of normalized cut clustering and linear discriminant analysis. This ensures that the discovered dancelets are both representative and discriminative. Additionally, to exploit the motion cues in videos, we employ motion boundaries as saliency priors to generate volumes of interest and extract C3D features to capture spatiotemporal information from the mid-level patches. Extensive experiments validated on our proposed large dance dataset, HIT Dances dataset, demonstrate the effectiveness of the proposed methods for dance style-based video recommendation.

Actor-action semantic segmentation made an important step toward advanced video understanding: what action is happening, who is performing the action, and where is the action happening in space-time. Current methods based on layered CRFs for this problem are local and unable to capture the long-ranging interactions of video parts. We propose a new model that combines the labeling CRF with a supervoxel hierarchy, where supervoxels at various scales provide cues for possible groupings of nodes in the CRF to encourage adaptive and long-ranging interactions. The new model defines a dynamic and continuous process of information exchange: the CRF influences what supervoxels in the hierarchy are active, and these active supervoxels, in turn, affect the connectivities in the CRF, we hence call it a grouping process model. By further incorporating the video-level recognition, the proposed method achieves a large margin of 60% relative improvement over the state of the art on the recent A2D large-scale video labeling dataset, which demonstrates the effectiveness of our modeling.

Can humans fly? Emphatically no. Can cars eat? Again, absolutely not. Yet, these absurd inferences result from the current disregard for particular types of actors in action understanding. There is no work we know of on simultaneously inferring actors and actions in the video, not to mention a dataset to experiment with. Our paper hence marks the first effort in the computer vision community to jointly consider various types of actors undergoing various actions. To start with the problem, we collect a dataset of 3782 videos from YouTube and label both pixel-level actors and actions in each video. We formulate the general actor-action understanding problem and instantiate it at various granularities: both video-level single- and multiple-label actor-action recognition and pixel-level actor-action semantic segmentation. Our experiments demonstrate that inference jointly over actors and actions outperforms inference independently over them, and hence concludes our argument of the value of explicit consideration of various actors in comprehensive action understanding.

The problem of describing images through natural language has gained importance in the computer vision community. Solutions to image description have either focused on a top-down approach of generating language through combinations of object detections and language models or bottom-up propagation of keyword tags from training images to test images through probabilistic or nearest neighbor techniques. In contrast, describing videos with natural language is a less studied problem. In this paper, we combine ideas from the bottom-up and top-down approaches to image description and propose a method for video description that captures the most relevant contents of a video in a natural language description. We propose a hybrid system consisting of a low level multimodal latent topic model for initial keyword annotation, a middle level of concept detectors and a high level module to produce final lingual descriptions. We compare the results of our system to human descriptions in both short and long forms on two datasets, and demonstrate that final system output has greater agreement with the human descriptions than any single level.

Supervoxel hierarchies provide a rich multiscale decomposition of a given video suitable for subsequent processing in video analysis. The hierarchies are typically computed by an unsupervised process that is susceptible to under-segmentation at coarse levels and over-segmentation at fine levels, which make it a challenge to adopt the hierarchies for later use. In this paper, we propose the first method to overcome this limitation and flatten the hierarchy into a single segmentation. Our method, called the uniform entropy slice, seeks a selection of supervoxels that balances the relative level of information in the selected supervoxels based on some post hoc feature criterion such as object-ness. For example, with this criterion, in regions nearby objects, our method prefers finer supervoxels to capture the local details, but in regions away from any objects we prefer coarser supervoxels. We formulate the uniform entropy slice as a binary quadratic program and implement four different feature criteria, both unsupervised and supervised, to drive the flattening. Although we apply it only to supervoxel hierarchies in this paper, our method is generally applicable to segmentation tree hierarchies. Our experiments demonstrate both strong qualitative performance and superior quantitative performance to state of the art baselines on benchmark internet videos.

Supervoxel segmentation has strong potential to be incorporated into early video analysis as superpixel segmentation has in image analysis. However, there are many plausible supervoxel methods and little understanding as to when and where each is most appropriate. Indeed, we are not aware of a single comparative study on supervoxel segmentation. To that end, we study five supervoxel algorithms in the context of what we consider to be a good supervoxel: namely, spatiotemporal uniformity, object/region boundary detection, region compression and parsimony. For the evaluation we propose a comprehensive suite of 3D volumetric quality metrics to measure these desirable supervoxel characteristics. We use three benchmark video data sets with a variety of content-types and varying amounts of human annotations. Our findings have led us to conclusive evidence that the hierarchical graph-based and segmentation by weighted aggregation methods perform best and almost equally-well on nearly all the metrics and are the methods of choice given our proposed assumptions.