Abstract Reasoning over sports videos for question answering is an important task with numerous applications, such as player training and information retrieval. However, this task has not been explored due to the lack of relevant datasets and the challenging nature it presents. Most datasets for video question answering (VideoQA) focus mainly on general and coarse-grained understanding of daily-life videos, which is not applicable to sports scenarios requiring professional action understanding and fine-grained motion analysis. In this paper, we introduce the first dataset, named Sports-QA, specifically designed for the sports VideoQA task. The Sports-QA dataset includes various types of questions, such as descriptions, chronologies, causalities, and counterfactual conditions, covering multiple sports. Furthermore, to address the characteristics of the sports VideoQA task, we propose a new Auto-Focus Transformer (AFT) capable of automatically focusing on particular scales of temporal information for question answering. We conduct extensive experiments on Sports-QA, including baseline studies and the evaluation of different methods. The results demonstrate that our AFT achieves state-of-the-art performance.

Category-Agnostic Pose Estimation (CAPE) aims to detect keypoints of unseen object categories in a few-shot setting, where the scarcity of labeled data poses significant challenges to generalization. In this work, we propose Prompt Pose Matching (PPM), a novel framework that unleashes the power of off-the-shelf text-to-image diffusion models for CAPE. PPM learns pseudo prompts from few-shot examples via the text-to-image diffusion model. These learned pseudo prompts capture semantic information of keypoints, which can then be used to locate the same type of keypoints from images. To provide prompts with representative initialization, we introduce a category-agnostic pre-training strategy to capture the foreground prior shared across categories and keypoints. To support the reliable prompt pre-training, we propose a Foreground-Aware Region Aggregation (FARA) module to provide robust and consistent supervision signal. Based on the foreground prior, a Foreground-Guided Attention Refinement (FGAR) module is further proposed to reinforce cross-attention responses for accurate keypoint localization. For efficiency, a Prompt Ensemble Inference (PEI) scheme enables joint keypoint prediction. Unlike previous methods that highly rely on base-category annotated data, our PPM framework can operate in a base-category-free setting while retaining strong performance. Code will be available at: https://github.com/DuoPeng-CVer/Prompt-Pose-Matching.

Zero-shot skeleton-based action recognition (ZS-SAR) is fundamentally constrained by prevailing approaches that rely on aligning skeleton features with static, class-level semantics. This coarse-grained alignment fails to bridge the domain shift between seen, unseen classes, thereby impeding the effective transfer of fine-grained visual knowledge. To address these limitations, we introduce DynaPURLS, a unified framework that establishes robust, multi-scale visual-semantic correspondences, dynamically refines them at inference time to enhance generalization. Our framework leverages a large language model to generate hierarchical textual descriptions that encompass both global movements, local body-part dynamics. Concurrently, an adaptive partitioning module produces fine-grained visual representations by semantically grouping skeleton joints. To fortify this fine-grained alignment against the train-test domain shift, DynaPURLS incorporates a dynamic refinement module. During inference, this module adapts textual features to the incoming visual stream via a lightweight learnable projection. This refinement process is stabilized by a confidence-aware, class-balanced memory bank, which mitigates error propagation from noisy pseudo-labels. Extensive experiments on three large-scale benchmark datasets, including NTU RGB+D 60/120, PKU-MMD, demonstrate that DynaPURLS significantly outperforms prior art, setting new state-of-the-art records.

Significant progress has been made in the field of video question answering (VideoQA) thanks to deep learning and large-scale pretraining. Despite the presence of sophisticated model structures and powerful video-text foundation models, most existing methods focus solely on maximizing the correlation between answers and video-question pairs during training. We argue that these models often establish shortcuts, resulting in spurious correlations between questions and answers, especially when the alignment between video and text data is suboptimal. To address these spurious correlations, we propose a novel training framework in which the model is compelled to acknowledge its ignorance when presented with an intervened question, rather than making guesses solely based on superficial question-answer correlations. We introduce methodologies for intervening in questions, utilizing techniques such as displacement and perturbation, and design frameworks for the model to admit its lack of knowledge in both multi-choice VideoQA and open-ended settings. In practice, we integrate a state-of-the-art model into our framework to validate its effectiveness. The results clearly demonstrate that our framework can significantly enhance the performance of VideoQA models with minimal structural modifications.

This paper proposes a general solution to enable point cloud recognition models to handle distribution shifts at test time. Unlike prior methods, which rely heavily on training data (often inaccessible during online inference) and are limited to recognizing a fixed set of point cloud classes predefined during training, we explore a more practical and challenging scenario: adapting the model solely based on online test data to recognize both previously seen classes and novel, unseen classes at test time. To this end, we develop Point-Cache, a hierarchical cache model that captures essential clues of online test samples, particularly focusing on the global structure of point clouds and their local-part details. Point-Cache, which serves as a rich 3D knowledge base, is dynamically managed to prioritize the inclusion of high-quality samples. Designed as a plug-and- play module, our method can be flexibly integrated into large multimodal 3D models to support open-vocabulary point cloud recognition. Notably, our solution operates with efficiency comparable to zero-shot inference, as it is entirely training-free. Point-Cache demonstrates substantial gains across 8 challenging benchmarks and 4 representative large 3D models, highlighting its effectiveness. Code is available at https://github.com/auniquesun/Point-Cache.

Understanding and reconstructing occluded objects is a challenging problem, especially in open-world scenarios where categories and contexts are diverse and unpredictable. Traditional methods, however, are typically restricted to closed sets of object categories, limiting their use in complex, open-world scenes. We introduce Open-World Amodal Appearance Completion, a training-free framework that expands amodal completion capabilities by accepting flexible text queries as input. Our approach generalizes to arbitrary objects specified by both direct terms and abstract queries. We term this capability reasoning amodal completion, where the system reconstructs the full appearance of the queried object based on the provided image and language query. Our framework unifies segmentation, occlusion analysis, and inpainting to handle complex occlusions and generates completed objects as RGBA elements, enabling seamless integration into applications such as 3D reconstruction and image editing. Extensive evaluations demonstrate the effectiveness of our approach in generalizing to novel objects and occlusions, establishing a new benchmark for amodal completion in open-world settings. Code and datasets available: https://github.com/saraao/amodal.

Video diffusion models have recently achieved remarkable results in video generation. Despite their encouraging performance, most of these models are mainly designed and trained for short video generation, leading to challenges in maintaining temporal consistency and visual details in long video generation. In this paper, we propose LongDiff, a novel training-free method consisting of carefully designed components – Position Mapping (PM) and Informative Frame Selection (IFS) – to tackle two key challenges that hinder short-to-long video generation generalization: temporal position ambiguity and information dilution. Our LongDiff unlocks the potential of off-the-shelf video diffusion models to achieve high-quality long video generation in one go. Extensive experiments demonstrate the efficacy of our method.

Skeleton-based human action recognition is promising due to its privacy preservation, robustness to visual challenges, and computational efficiency. Especially, the practical necessity to recognize unseen actions has led to increased interest in zero-shot skeleton-based action recognition (ZSSAR). Existing ZSSAR approaches often rely on manually crafted action descriptions or visual assumptions to enhance knowledge transfer, which is limited in flexibility and prone to inaccuracies and noise. To overcome this, we introduce Semantic-guided Cross-Modal Prompt Learning (SCoPLe), a novel framework that replaces manual guidance with data-driven prompt learning for refinement and alignment of skeletal and textual features. Specifically, we introduce a dual-stream language prompting module that preserves the original semantic context from the pre-trained text encoder while still effectively tuning its ouput for ZSSAR task adaptation. We also introduce a joint-shaped prompting module that learns tuning for skeleton features and incorporate an adaptive visual representation sampler that leverages text semantics to strengthen the cross-modal prompting interactions during skeleton-to-text embedding projection. Experimental results on the NTU-RGB+D and PKU-MMD datasets demonstrate the state-of-the-art performance of our method in both ZS-SAR and generalized ZSSAR scenarios.

Text-to-motion generation has recently garnered significant research interest, primarily focusing on generating human motion sequences in blank backgrounds. However, human motions commonly occur within diverse 3D scenes, which has prompted exploration into scene-aware text-to-motion generation methods. Yet, existing scene-aware methods often rely on large-scale ground-truth motion sequences in diverse 3D scenes, which poses practical challenges due to the expensive cost. To mitigate this challenge, we are the first to propose a Training-free Scene-aware Text-to-Motion framework, dubbed as TSTMotion, that efficiently empowers pre-trained blank-background motion generators with the scene-aware capability. Specifically, conditioned on the given 3D scene and text description, we adopt foundation models together to reason, predict and validate a scene-aware motion guidance. Then, the motion guidance is incorporated into the blank-background motion generators with two modifications, resulting in scene-aware text-driven motion sequences. Extensive experiments demonstrate the efficacy and generalizability of our proposed framework. We release our code in Project Page.

Score-based generative models are essential in various machine learning applications, with strong capabilities in generation quality. In particular, high-order derivatives (scores) of data density offer deep insights into data distributions, building on the proven effectiveness of first-order scores for modeling and generating synthetic data, unlocking new possibilities for applications. However, learning them typically requires complete data, which is often unavailable in domains such as healthcare and finance due to data corruption, acquisition constraints, or incomplete records. To tackle this challenge, we introduce MissScore, a novel framework for estimating high-order scores in the presence of missing data. We derive objective functions for estimating high-order scores under different missing data mechanisms and propose a new algorithm specifically designed to handle missing data effectively. Our empirical results demonstrate that MissScore accurately and efficiently learns the high-order scores from incomplete data and generates high-quality samples, resulting in strong performance across a range of downstream tasks.

Humans naturally understand moments in a video by integrating visual and auditory cues. For example, localizing a scene in the video like “A scientist passionately speaks on wildlife conservation as dramatic orchestral music plays, with the audience nodding and applauding” requires simultaneous processing of visual, audio, and speech signals. However, existing models often struggle to effectively fuse and interpret audio information, limiting their capacity for comprehensive video temporal understanding. To address this, we present TriSense, a triple-modality large language model designed for holistic video temporal understanding through the integration of visual, audio, and speech modalities. Central to TriSense is a Query-Based Connector that adaptively reweights modality contributions based on the input query, enabling robust performance under modality dropout and allowing flexible combinations of available inputs. To support TriSense's multimodal capabilities, we introduce TriSense-2M, a high-quality dataset of over 2 million curated samples generated via an automated pipeline powered by fine-tuned LLMs. TriSense-2M includes long-form videos and diverse modality combinations, facilitating broad generalization. Extensive experiments across multiple benchmarks demonstrate the effectiveness of TriSense and its potential to advance multimodal video analysis.

We introduce Skeleton-Cache, the first training-free test-time adaptation framework for skeleton-based zero-shot action recognition (SZAR), aimed at improving model generalization to unseen actions during inference. Skeleton-Cache reformulates inference as a lightweight retrieval process over a non-parametric cache that stores structured skeleton representations, combining both global and fine-grained local descriptors. To guide the fusion of descriptor-wise predictions, we leverage the semantic reasoning capabilities of large language models (LLMs) to assign class-specific importance weights. By integrating these structured descriptors with LLM-guided semantic priors, Skeleton-Cache dynamically adapts to unseen actions without any additional training or access to training data. Extensive experiments on NTU RGB+D 60/120 and PKU-MMD II demonstrate that Skeleton-Cache consistently boosts the performance of various SZAR backbones under both zero-shot and generalized zero-shot settings. The code is publicly available at https://github.com/Alchemist0754/Skeleton-Cache.

Text-to-Image (T2I) models have advanced significantly, but their growing popularity raises security concerns due to their potential to generate harmful images. To address these issues, we propose UPAM, a novel framework to evaluate the robustness of T2I models from an attack perspective. Unlike prior methods that focus solely on textual defenses, UPAM unifies the attack on both textual and visual defenses. Additionally, it enables gradient-based optimization, overcoming reliance on enumeration for improved efficiency and effectiveness. To handle cases where T2I models block image outputs due to defenses, we introduce Sphere-Probing Learning (SPL) to enable optimization even without image results. Following SPL, our model bypasses defenses, inducing the generation of harmful content. To ensure semantic alignment with attacker intent, we propose Semantic-Enhancing Learning (SEL) for precise semantic control. UPAM also prioritizes the naturalness of adversarial prompts using In-context Naturalness Enhancement (INE), making them harder for human examiners to detect. Additionally, we address the issue of iterative queries-common in prior methods and easily detectable by API defenders-by introducing Transferable Attack Learning (TAL), allowing effective attacks with minimal queries. Extensive experiments validate UPAM's superiority in effectiveness, efficiency, naturalness, and low query detection rates.

While remarkable progress has been made on supervised skeleton-based action recognition, the challenge of zero-shot recognition remains relatively unexplored. In this pa-per, we argue that relying solely on aligning label-level se-mantics and global skeleton features is insufficient to effectively transfer locally consistent visual knowledge from seen to unseen classes. To address this limitation, we intro-duce Part-aware Unified Representation between Language and Skeleton (PURLS) to explore visual-semantic alignment at both local and global scales. PURLS introduces a new prompting module and a novel partitioning module to gen-erate aligned textual and visual representations across dif-ferent levels. The former leverages a pre-trained GPT-3 to infer refined descriptions of the global and local (body-part-based and temporal-interval-based) movements from the original action labels. The latter employs an adaptive sampling strategy to group visual features from all body joint movements that are semantically relevant to a given description. Our approach is evaluated on various skele-ton/language backbones and three large-scale datasets, i.e., NTU-RGB+D 60, NTU-RGB+D 120, and a newly curated dataset Kinetics-skeleton 200. The results showcase the universality and superior performance of PURLS, surpassing prior skeleton-based solutions and standard baselines from other domains. The source codes can be accessed at https://github.com/azzhl/PURLS.

Text-to-Image (T2I) models have raised security concerns due to their potential to generate inappropriate or harmful images. In this paper, we propose UPAM, a novel framework that investigates the robustness of T2I models from the attack perspective. Unlike most existing attack methods that focus on deceiving textual defenses, UPAM aims to deceive both textual and visual defenses in T2I models. UPAM enables gradient-based optimization, offering greater effectiveness and efficiency than previous methods. Given that T2I models might not return results due to defense mechanisms, we introduce a Sphere-Probing Learning (SPL) scheme to support gradient optimization even when no results are returned. Additionally, we devise a Semantic-Enhancing Learning (SEL) scheme to finetune UPAM for generating target-aligned images. Our framework also ensures attack stealthiness. Extensive experiments demonstrate UPAM's effectiveness and efficiency.

This paper investigates the 3D domain generalization (3DDG) ability of large 3D models based on prevalent prompt learning. Recent works demonstrate the performances of 3D point cloud recognition can be boosted remarkably by parameter-efficient prompt tuning. However, we observe that the improvement on downstream tasks comes at the expense of a severe drop in 3D domain generalization. To resolve this challenge, we present a comprehensive regulation framework that allows the learnable prompts to actively interact with the well-learned general knowledge in large 3D models to maintain good generalization. Specifically, the proposed framework imposes multiple explicit constraints on the prompt learning trajectory by maximizing the mutual agreement between task-specific predictions and task-agnostic knowledge. We design the regulation framework as a plug-and-play module to embed into existing representative large 3D models. Surprisingly, our method not only realizes consistently increasing generalization ability but also enhances task-specific 3D recognition performances across various 3DDG benchmarks by a clear margin. Considering the lack of study and evaluation on 3DDG, we also create three new benchmarks, namely base-to-new, cross-dataset and few-shot generalization benchmarks, to enrich the field and inspire future research. Code and benchmarks are available at \url{https://github.com/auniquesun/Point-PRC}.

Images of realistic scenes often contain intra-class objects that are heavily occluded from each other, making the amodal perception task that requires parsing the occluded parts of the objects challenging. Although important for downstream tasks such as robotic grasping systems, the lack of large-scale amodal datasets with detailed annotations makes it difficult to model intra-class occlusions explicitly. This paper introduces two new amodal datasets for image amodal completion tasks, which contain a total of over 267K images of intra-class occlusion scenarios, annotated with multiple masks, amodal bounding boxes, dual order relations and full appearance for instances and background. We also present a point-supervised scheme with layer priors for amodal instance segmentation specifically designed for intra-class occlusion scenarios1. Experiments show that our weakly supervised approach outperforms the SOTA fully supervised methods, while our layer priors design exhibits remarkable performance improvements in the case of intra-class occlusion in both synthetic and real images.

Image anomaly localization (IAL) is widely applied in fault detection and industrial inspection domains to discover anomalous patterns in images at the pixel level. The unique challenge of this task is the lack of comprehensive anomaly samples for model training. The state-of-the-art methods train end-to-end models that leverage outlier exposure to simulate pseudo anomalies, but they show poor transferability to new datasets due to the inherent bias to the synthesized outliers during training. Recently, two-stage instance-level self-supervised learning (SSL) has shown potential in learning generic representations for IAL. However, we hypothesize that dense-level SSL is more compatible as IAL requires pixel-level prediction. In this paper, we bridge these gaps by proposing a two-stage, dense pretraining model tailored for the IAL task. More specifically, our model utilizes dual positive-pair selection criteria and dual feature scales to learn more effective representations. Through extensive experiments, we show that our learned representations achieve significantly better anomaly localization performance among two-stage models, while requiring almost half the convergence time. Moreover, our learned representations have better transferability to unseen datasets. Code is available at https://github.com/terrlo/DS2.

Unsupervised Domain Adaptation (UDA) is quite challenging due to the large distribution discrepancy between the source domain and the target domain. Inspired by diffusion models which have strong capability to gradually convert data distributions across a large gap, we consider to explore the diffusion technique to handle the challenging UDA task. However, using diffusion models to convert data distribution across different domains is a non-trivial problem as the standard diffusion models generally perform conversion from the Gaussian distribution instead of from a specific domain distribution. Besides, during the conversion, the semantics of the source-domain data needs to be preserved to classify correctly in the target domain. To tackle these problems, we propose a novel Domain-Adaptive Diffusion (DAD) module accompanied by a Mutual Learning Strategy (MLS), which can gradually convert data distribution from the source domain to the target domain while enabling the classification model to learn along the domain transition process. Consequently, our method successfully eases the challenge of UDA by decomposing the large domain gap into small ones and gradually enhancing the capacity of classification model to finally adapt to the target domain. Our method outperforms the current state-of-the-arts by a large margin on three widely used UDA datasets.

Currently, salience-based channel pruning makes continuous breakthroughs in network compression. In the realization, the salience mechanism is used as a metric of channel salience to guide pruning. Therefore, salience-based channel pruning can dynamically adjust the channel width at run-time, which provides a flexible pruning scheme. However, there are two problems emerging: a gating function is often needed to truncate the specific salience entries to zero, which destabilizes the forward propagation; dynamic architecture brings more cost for indexing in inference which bottlenecks the inference speed. In this article, we propose a Progressive Channel-Shrinking (PCS) method to compress the selected salience entries at run-time instead of roughly approximating them to zero. We also propose a Running Shrinking Policy to provide a testing-static pruning scheme that can reduce the memory access cost for filter indexing. We evaluate our method on ImageNet and CIFAR10 datasets over two prevalent networks: ResNet and VGG, and demonstrate that our PCS outperforms all baselines and achieves state-of-the-art in terms of compression-performance tradeoff. Moreover, we observe a significant and practical acceleration of inference. The code is available athttps://github.com/JianhongPan-VLG/Progressive.Channel-Shrinking.Network.

We propose a Unified Pose Sequence Modeling approach to unify heterogeneous human behavior understanding tasks based on pose data, e.g., action recognition, 3D pose estimation and 3D early action prediction. A major obstacle is that different pose-based tasks require different output data formats. Specifically, the action recognition and prediction tasks require class predictions as outputs, while 3D pose estimation requires a human pose output, which limits existing methods to leverage task-specific network architectures for each task. Hence, in this paper, we propose a novel Unified Pose Sequence (UPS) model to unify heterogeneous output formats for the aforementioned tasks by considering text-based action labels and coordinate-based human poses as language sequences. Then, by optimizing a single auto-regressive transformer, we can obtain a unified output sequence that can handle all the aforementioned tasks. Moreover, to avoid the interference brought by the heterogeneity between different tasks, a dynamic routing mechanism is also proposed to empower our UPS with the ability to learn which subsets of parameters should be shared among different tasks. To evaluate the efficacy of the proposed UPS, extensive experiments are conducted on four different tasks with four popular behavior understanding benchmarks.

Monocular 3D human pose estimation is quite challenging due to the inherent ambiguity and occlusion, which often lead to high uncertainty and indeterminacy. On the other hand, diffusion models have recently emerged as an effective tool for generating high-quality images from noise. In-spired by their capability, we explore a novel pose estimation framework (DiffPose) that formulates 3D pose estimation as a reverse diffusion process. We incorporate novel designs into our DiffPose to facilitate the diffusion process for 3D pose estimation: a pose-specific initialization of pose uncertainty distributions, a Gaussian Mixture Model-based forward diffusion process, and a context-conditioned re-verse diffusion process. Our proposed DiffPose significantly outperforms existing methods on the widely used pose estimation benchmarks Human3.6M and MPI-INF-3DHP. Project page: https://gongjia0208.github.io/Diffpose/.

Translating images from a source domain to a target domain for learning target models is one of the most common strategies in domain adaptive semantic segmentation (DASS). However, existing methods still struggle to preserve semantically-consistent local details between the original and translated images. In this work, we present an innovative approach that addresses this challenge by using sourcedomain labels as explicit guidance during image translation. Concretely, we formulate cross-domain image translation as a denoising diffusion process and utilize a novel Semantic Gradient Guidance (SGG) method to constrain the translation process, conditioning it on the pixel-wise source labels. Additionally, a Progressive Translation Learning (PTL) strategy is devised to enable the SGG method to work reliably across domains with large gaps. Extensive experiments demonstrate the superiority of our approach over state-of-the-art methods.

Privacy-Preserving Action Recognition (PPAR) aims to transform raw videos into anonymous ones to prevent privacy leakage while maintaining action clues, which is an increasingly important problem in intelligent vision applications. Despite recent efforts in this task, it is still challenging to deal with novel privacy attributes and novel privacy attack models that are unavailable during the training phase. In this paper, from the perspective of meta-learning (learning to learn), we propose a novel Meta Privacy-Preserving Action Recognition (MPPAR) framework to improve both generalization abilities above (i.e., generalize to *novel privacy attributes* and *novel privacy attack models*) in a unified manner. Concretely, we simulate train/test task shifts by constructing disjoint support/query sets w.r.t. privacy attributes or attack models. Then, a virtual training and testing scheme is applied based on support/query sets to provide feedback to optimize the model's learning toward better generalization. Extensive experiments demonstrate the effectiveness and generalization of the proposed framework compared to state-of-the-arts.

Modern video summarization methods are based on deep neural networks that require a large amount of annotated data for training. However, existing datasets for video summarization are small-scale, easily leading to over-fitting of the deep models. Considering that the annotation of large-scale datasets is time-consuming, we propose a multimodal self-supervised learning framework to obtain semantic representations of videos, which benefits the video summarization task. Specifically, the self-supervised learning is conducted by exploring the semantic consistency between the videos and text in both coarse-grained and fine-grained fashions, as well as recovering masked frames in the videos. The multimodal framework is trained on a newly-collected dataset that consists of video-text pairs. Additionally, we introduce a progressive video summarization method, where the important content in a video is pinpointed progressively to generate better summaries. Extensive experiments have proved the effectiveness and superiority of our method in rank correlation coefficients and F-score1.

Skeleton-based action recognition receives increasing attention because skeleton sequences reduce training complexity by eliminating visual information irrelevant to actions. To further improve sample efficiency, meta-learning-based one-shot learning solutions were developed for skeleton-based action recognition. These methods predict by finding the nearest neighbors according to the similarity between instance-level global embedding. However, such measurement holds unstable representativity due to inadequate generalized learning on the averaged local invariant and noisy features, while intuitively, steady and fine-grained recognition relies on determining key local body movements. To address this limitation, we present the Adaptive Local-Component-aware Graph Convolutional Network, which replaces the comparison metric with a focused sum of similarity measurements on aligned local embedding of action-critical spatial/temporal segments. Comprehensive one-shot experiments on the public benchmark of NTURGB+D 120 indicate that our method provides a stronger representation than the global embedding and helps our model reach state-of-the-art.

Abstract In recent years, one of the most popular techniques in the computer vision community has been the deep learning technique. As a data-driven technique, deep model requires enormous amounts of accurately labelled training data, which is often inaccessible in many real-world applications. A data-space solution is Data Augmentation (DA), that can artificially generate new images out of original samples. Image augmentation strategies can vary by dataset, as different data types might require different augmentations to facilitate model training. However, the design of DA policies has been largely decided by the human experts with domain knowledge, which is considered to be highly subjective and error-prone. To mitigate such problem, a novel direction is to automatically learn the image augmentation policies from the given dataset using Automated Data Augmentation (AutoDA) techniques. The goal of AutoDA models is to find the optimal DA policies that can maximize the model performance gains. This survey discusses the underlying reasons of the emergence of AutoDA technology from the perspective of image classification. We identify three key components of a standard AutoDA model: a search space, a search algorithm and an evaluation function. Based on their architecture, we provide a systematic taxonomy of existing image AutoDA approaches. This paper presents the major works in AutoDA field, discussing their pros and cons, and proposing several potential directions for future improvements.

Video summarization aims to automatically generate a summary (storyboard or video skim) of a video, which can facilitate large-scale video retrieval and browsing. Most of the existing methods perform video summarization on individual videos, which neglects the correlations among similar videos. Such correlations, however, are also informative for video understanding and video summarization. To address this limitation, we propose Video Joint Modelling based on Hierarchical Transformer (VJMHT) for co-summarization, which takes into consideration the semantic dependencies across videos. Specifically, VJMHT consists of two layers of Transformer: the first layer extracts semantic representation from individual shots of similar videos, while the second layer performs shot-level video joint modelling to aggregate cross-video semantic information. By this means, complete cross-video high-level patterns are explicitly modelled and learned for the summarization of individual videos. Moreover, Transformer-based video representation reconstruction is introduced to maximize the high-level similarity between the summary and the original video. Extensive experiments are conducted to verify the effectiveness of the proposed modules and the superiority of VJMHT in terms of F-measure and rank-based evaluation.

Dynamic neural networks can greatly reduce computation redundancy without compromising accuracy by adapting their structures based on the input. In this paper, we explore the robustness of dynamic neural networks against energy-oriented attacks targeted at reducing their efficiency. Specifically, we attack dynamic models with our novel algorithm GradMDM. GradMDM is a technique that adjusts the direction and the magnitude of the gradients to effectively find a small perturbation for each input, that will activate more computational units of dynamic models during inference. We evaluate GradMDM on multiple datasets and dynamic models, where it outperforms previous energy-oriented attack techniques, significantly increasing computation complexity while reducing the perceptibility of the perturbations https://github.com/lingengfoo/GradMDM.

Human Action Recognition (HAR) aims to understand human behavior and assign a label to each action. It has a wide range of applications, and therefore has been attracting increasing attention in the field of computer vision. Human actions can be represented using various data modalities, such as RGB, skeleton, depth, infrared, point cloud, event stream, audio, acceleration, radar, and WiFi signal, which encode different sources of useful yet distinct information and have various advantages depending on the application scenarios. Consequently, lots of existing works have attempted to investigate different types of approaches for HAR using various modalities. In this article, we present a comprehensive survey of recent progress in deep learning methods for HAR based on the type of input data modality. Specifically, we review the current mainstream deep learning methods for single data modalities and multiple data modalities, including the fusion-based and the co-learning-based frameworks. We also present comparative results on several benchmark datasets for HAR, together with insightful observations and inspiring future research directions.

Early actionrecognition, i.e., recognizing an action before it is fully performed, is a challenging and important task. Existing works mainly focus on deterministic early action recognition outputting only a single class, and ignore the uncertainty and diversity that essentially exist in this task. Intuitively, when only the early portion of the action is observed, there could be multiple possibilities of the full action, as diversified actions can share almost identical early segments in many scenarios. Thus taking uncertainties and diversities into account, and outputting multiple plausible predictions, instead of a single one, can be important for the sake of authenticity and requirement of many practical applications. To this end, we propose a novel Diversified Early Action Recognition Network (Dear-Net) that is capable of outputting multiple reasonable action classes for each partial sequence by utilizing mode conversion. Specifically, we introduce an effective action diversity learning strategy to drive our network towards predicting diverse and reasonable results, in which each learnable action class is matched with the most suitable mode. Meanwhile, the collapsed modes which fail to receive any action class, are also considered in this strategy in order to ensure diversity. Moreover, we design a sequence decoder within our network to capture latent global information for better early action recognition. It provides a feasible scheme for weakly-supervised setting in which the Dear-Net leverages unlabelled data to improve performance. Experimental results on three challenging datasets clearly show the effectiveness of our approach.

The existing pose estimation approaches often require a large number of annotated images to attain good estimation performance, which are laborious to acquire. To reduce the human efforts on pose annotations, we propose a novel Meta Agent Teaming Active Learning (MATAL) framework to actively select and label informative images for effective learning. Our MATAL formulates the image selection procedure as a Markov Decision Process and learns an optimal sampling policy that directly maximizes the performance of the pose estimator based on the reward. Our framework consists of a novel state-action representation as well as a multi-agent team to enable batch sampling in the active learning procedure. The framework could be effectively optimized via Meta-Optimization to accelerate the adaptation to the gradually expanded labeled data during deployment. Finally, we show experimental results on both human hand and body pose estimation benchmark datasets and demonstrate that our method significantly outperforms all baselines continuously under the same amount of annotation budget. Moreover, to obtain similar pose estimation accuracy, our MATAL framework can save around 40% labeling efforts on average compared to state-of-the-art active learning frameworks.

Currently, single image inpainting has achieved promising results based on deep convolutional neural networks. However, inpainting on stereo images with missing regions has not been explored thoroughly, which is also a significant but different problem. One crucial requirement for stereo image inpainting is stereo consistency. To achieve it, we propose an Iterative Geometry-Aware Cross Guidance Network (IGGNet). The IGGNet contains two key ingredients, i.e., a Geometry-Aware Attention(GAA) module and an Iterative Cross Guidance(ICG) strategy. The GAA module relies on the epipolar geometry cues and learns the geometry-aware guidance from one view to another, which is beneficial to make the corresponding regions in two views consistent. However, learning guidance from co-existing missing regions is challenging. To address this issue, the ICG strategy is proposed, which can alternately narrow down the missing regions of the two views in an iterative manner. Experimental results demonstrate that our proposed network outperforms the latest stereo image inpainting model and state-of-the-art single image inpainting models.

Effective learning of spatial-temporal information within a point cloud sequence is highly important for many down-stream tasks such as 4D semantic segmentation and 3D action recognition. In this paper, we propose a novel frame-work named Point Spatial-Temporal Transformer (PST2) to learn spatial-temporal representations from dynamic 3D point cloud sequences. Our PST2consists of two major modules: a Spatio-Temporal Self-Attention (STSA) module and a Resolution Embedding (RE) module. Our STSA module is introduced to capture the spatial-temporal context in-formation across adjacent frames, while the RE module is proposed to aggregate features across neighbors to enhance the resolution of feature maps. We test the effectiveness our PST2with two different tasks on point cloud sequences, i.e., 4D semantic segmentation and 3D action recognition. Extensive experiments on three benchmarks show that our PST2outperforms existing methods on all datasets. The effectiveness of our STSA and RE modules have also been justified with ablation experiments.

Pedestrian crossing intention prediction is crucial to traffic safety, which is a challenging task in real traffic scenarios. Traditional methods infer the intention of pedestrians to cross by predicting their future movements based on the observed trajectories in history. The performance of those methods is limited due to insufficient features and sources of information. To address those limitations, we propose a ViT-based model which incorporates multi-modal data to predict the pedestrian crossing intention. Specifically, the proposed model takes into consideration the visual information, poses, bounding box coordinates and action annotations, and gradually fuses those features for the final prediction. Besides, different data processing methods are designed based on the corresponding characteristics of different modalities to make full use of each type of data. Extensive ablation studies are conducted to show the performance of temporal modelling and feature fusion.

Crowd counting is required for many situations and has historically been undertaken using approximate (manual) estimations and measures. Deep learning allows to improve this situation. Modern crowd counting models are commonly based on pixel-wise density maps using deep convolutional neural networks (CNNs) comprising tens of millions of parameters. These models require high-performance GPUs for training and subsequent usage and inference. As such, these models are difficult to deliver to edge devices that have limited computing resources such as surveillance cameras, mobile phones and Internet of Things (IoT)-type devices. This paper proposes a new method to tackle this issue based on three key components: feature fusion, Bayesian Loss [19] and datasets utilising bounding-box annotations to increase the efficiency of the crowd counting task. Experiments show that the proposed method can not only provide accuracy close to the latest state-of-the-art deep learning models, but support real-time inference in edge devices offering limited computational capacity.

Computer vision opportunities based on deep learning have seen an increasing trend in recent years through the proliferation of convolutional neural networks (CNN) and related deep learning models. However, limited training data often constrains the performance and accuracy of such models - this is a common situation for many detection and classification tasks especially in the ecological field. In this paper, we present a case study of how one such model: You Only Look Once (YOLO) version 5 can be applied to individual feral cat identification using a small unbalanced data set. We describe the procedures for preparing the training and validation data set, training the model using data augmentation and transfer learning techniques, and testing the model on both in-distribution and out-of-distribution samples. We explore the effectiveness of data augmentation methods including basic image manipulations and more advanced augmentation techniques that are now available.

Most of the state-of-the-art action recognition methods focus on offline learning, where the samples of all types of actions need to be provided at once. Here, we address continual learning of action recognition, where various types of new actions are continuously learned over time. This task is quite challenging, owing to the catastrophic forgetting problem stemming from the discrepancies between the previously learned actions and current new actions to be learned. Therefore, we propose Else-Net, a novel Elastic Semantic Network with multiple learning blocks to learn diversified human actions over time. Specifically, our Else-Net is able to automatically search and update the most relevant learning blocks w.r.t. the current new action, or explore new blocks to store new knowledge, preserving the unmatched ones to retain the knowledge of previously learned actions and alleviates forgetting when learning new actions. Moreover, even though different human actions may vary to a large extent as a whole, their local body parts can still share many homogeneous features. Inspired by this, our proposed Else-Net mines the shared knowledge of the decomposed human body parts from different actions, which benefits continual learning of actions. Experiments show that the proposed approach enables effective continual action recognition and achieves promising performance on two large-scale action recognition datasets.

Deep image inpainting aims to restore damaged or missing regions in an image with realistic contents. While having a wide range of applications such as object removal and image recovery, deep inpainting techniques also have the risk of being manipulated for image forgery. A promising countermeasure against such forgeries is deep inpainting detection, which aims to locate the inpainted regions in an image. In this paper, we make the first attempt towards universal detection of deep inpainting, where the detection network can generalize well when detecting different deep inpainting methods. To this end, we first propose a novel data generation approach to generate a universal training dataset, which imitates the noise discrepancies exist in real versus inpainted image contents to train universal detectors. We then design a Noise-Image Cross-fusion Network (NIX-Net) to effectively exploit the discriminative information contained in both the images and their noise patterns. We empirically show, on multiple benchmark datasets, that our approach outperforms existing detection methods by a large margin and generalize well to unseen deep inpainting techniques. Our universal training dataset can also significantly boost the generalizability of existing detection methods.

In this paper, we aim to tackle the task of Assessing Future Moment of an Action of Interest (AFM-AI). The goal of this task is to assess if an action of interest will happen or not as well as the starting moment of the action. We aim to assess starting moments at any time-horizon of the future. To this end, we tackle the regression task of the starting moments as a generation task using a Deterministic Residual Guided Variational Regression Module (DR-VRM), which is built on a Variational Regression Module (VRM) and a deterministic residual network. The VRM takes the uncertainty into account and is capable of generating diverse predictions for the starting moment. The deterministic network encourages the VRM to learn from deterministic residual information in order to generate more precise predictions for moment assessment. Experimental results on three datasets clearly show that the proposed method is capable of generating both diverse and precise predictions of starting moments for query actions.

Street Scene Change Detection (SSCD) aims to locate the changed regions between a given street-view image pair captured at different times, which is an important yet challenging task in the computer vision community. The intuitive way to solve the SSCD task is to fuse the extracted image feature pairs, and then directly measure the dissimilarity parts for producing a change map. Therefore, the key for the SSCD task is to design an effective feature fusion method that can improve the accuracy of the corresponding change maps. To this end, we present a novel Hierarchical Paired Channel Fusion Network (HPCFNet), which utilizes the adaptive fusion of paired feature channels. Specifically, the features of a given image pair are jointly extracted by a Siamese Convolutional Neural Network (SCNN) and hierarchically combined by exploring the fusion of channel pairs at multiple feature levels. In addition, based on the observation that the distribution of scene changes is diverse, we further propose a Multi-Part Feature Learning (MPFL) strategy to detect diverse changes. Based on the MPFL strategy, our framework achieves a novel approach to adapt to the scale and location diversities of the scene change regions. Extensive experiments on three public datasets (i.e., PCD, VL-CMU-CD and CDnet2014) demonstrate that the proposed framework achieves superior performance which outperforms other state-of-the-art methods with a considerable margin.

Hexagonal CNN models have shown superior performance in applications such as IACT data analysis and aerial scene classification due to their better rotation symmetry and reduced anisotropy. In order to realize hexagonal processing, existing studies mainly use the ZeroOut method to imitate hexagonal processing, which causes substantial memory and computation overheads. We address this deficiency with a novel native hexagonal CNN framework named HexCNN. HexCNN takes hexagon-shaped input and performs forward and backward propagation on the original form of the input based on hexagon-shaped filters, hence avoiding computation and memory overheads caused by imitation. For applications with rectangle-shaped input but require hexagonal processing, HexCNN can be applied by padding the input into hexagon-shape as preprocessing. In this case, we show that the time and space efficiency of HexCNN still outperforms existing hexagonal CNN methods substantially. Experimental results show that compared with the state-of-the-art models, which imitate hexagonal processing but using rectangle-shaped filters, HexCNN reduces the training time by up to 42.2%. Meanwhile, HexCNN saves the memory space cost by up to 25% and 41.7% for loading the input and performing convolution, respectively.

Human actions represented with 3D skeleton sequences are robust to clustered backgrounds and illumination changes. In this paper, we investigate skeleton-based action prediction, which aims to recognize an action from a partial skeleton sequence that contains incomplete action information. We propose a new Latent Global Network based on adversarial learning for action prediction. We demonstrate that the proposed network provides latent long-term global information that is complementary to the local action information of the partial sequences and helps improve action prediction. We show that action prediction can be improved by combining the latent global information with the local action information. We test the proposed method on three challenging skeleton datasets and report state-of-the-art performance.

The goal of human action anticipation is to predict future actions. Ideally, in real-world applications such as video surveillance and self-driving systems, future actions should not only be predicted with high accuracy but also at arbitrary and variable time-horizons ranging from short- to long-term predictions. Current work mostly focuses on predicting the next action and thus long-term prediction is achieved by recursive prediction of each next action, which is both inefficient and accumulates errors. In this paper, we propose a novel time-conditioned method for efficient and effective long-term action anticipation. There are two key ingredients to our approach. First, by explicitly conditioning our anticipation network on time allows to efficiently anticipate also long-term actions. And second, we propose an attended temporal feature and a time-conditioned skip connection to extract relevant and useful information from observations for effective anticipation. We conduct extensive experiments on the large-scale Epic-Kitchen and the 50Salads Datasets. Experimental results show that the proposed method is capable of anticipating future actions at both short-term and long-term, and achieves state-of-the-art performance.

This paper presents a new representation of skeleton sequences for 3D action recognition. Existing methods based on hand-crafted features or recurrent neural networks cannot adequately capture the complex spatial structures and the long-term temporal dynamics of the skeleton sequences, which are very important to recognize the actions. In this paper, we propose to transform each channel of the 3D coordinates of a skeleton sequence into a clip. Each frame of the generated clip represents the temporal information of the entire skeleton sequence and one particular spatial relationship between the skeleton joints. The entire clip incorporates multiple frames with different spatial relationships, which provide useful spatial structural information of the human skeleton. We also propose a multitask convolutional neural network (MTCNN) to learn the generated clips for action recognition. The proposed MTCNN processes all the frames of the generated clips in parallel to explore the spatial and temporal information of the skeleton sequences. The proposed method has been extensively tested on six challenging benchmark datasets. Experimental results consistently demonstrate the superiority of the proposed clip representation and the feature learning method for 3D action recognition compared to the existing techniques.

Predicting an interaction before it is fully executed is very important in applications, such as human-robot interaction and video surveillance. In a two-human interaction scenario, there are often contextual dependency structures between the global interaction context of the two humans and the local context of the different body parts of each human. In this paper, we propose to learn the structure of the interaction contexts and combine it with the spatial and temporal information of a video sequence to better predict the interaction class. The structural models, including the spatial and the temporal models, are learned with long short term memory (LSTM) networks to capture the dependency of the global and local contexts of each RGB frame and each optical flow image, respectively. LSTM networks are also capable of detecting the key information from global and local interaction contexts. Moreover, to effectively combine the structural models with the spatial and temporal models for interaction prediction, a ranking score fusion method is introduced to automatically compute the optimal weight of each model for score fusion. Experimental results on the BIT-Interaction Dataset and the UT-Interaction Dataset clearly demonstrate the benefits of the proposed method.

This paper presents a new method for 3D action recognition with skeleton sequences (i.e., 3D trajectories of human skeleton joints). The proposed method first transforms each skeleton sequence into three clips each consisting of several frames for spatial temporal feature learning using deep neural networks. Each clip is generated from one channel of the cylindrical coordinates of the skeleton sequence. Each frame of the generated clips represents the temporal information of the entire skeleton sequence, and incorporates one particular spatial relationship between the joints. The entire clips include multiple frames with different spatial relationships, which provide useful spatial structural information of the human skeleton. We propose to use deep convolutional neural networks to learn long-term temporal information of the skeleton sequence from the frames of the generated clips, and then use a Multi-Task Learning Network (MTLN) to jointly process all frames of the clips in parallel to incorporate spatial structural information for action recognition. Experimental results clearly show the effectiveness of the proposed new representation and feature learning method for 3D action recognition.

This letter presents SkeletonNet, a deep learning framework for skeleton-based 3-D action recognition. Given a skeleton sequence, the spatial structure of the skeleton joints in each frame and the temporal information between multiple frames are two important factors for action recognition. We first extract body-part-based features from each frame of the skeleton sequence. Compared to the original coordinates of the skeleton joints, the proposed features are translation, rotation, and scale invariant. To learn robust temporal information, instead of treating the features of all frames as a time series, we transform the features into images and feed them to the proposed deep learning network, which contains two parts: one to extract general features from the input images, while the other to generate a discriminative and compact representation for action recognition. The proposed method is tested on the SBU kinect interaction dataset, the CMU dataset, and the large-scale NTU RGB+D dataset and achieves state-of-the-art performance.

Rotation in closed contour recognition is a puzzling nuisance in most algorithms. In this paper we address three fundamental issues brought by rotation in shapes: 1) is alignment among shapes necessary? If the answer is "no", 2) how to exploit information in different rotations? and 3) how to use rotation unaware local features for rotation aware shape recognition? We argue that the origin of these issues is the use of hand crafted rotation-unfriendly features and measurements. Therefore our goal is to learn a set of hierarchical features that describe all rotated versions of a shape as a class, with the capability of distinguishing different such classes. We propose to rotate shapes as many times as possible as training samples, and learn the hierarchical feature representation by effectively adopting a convolutional neural network. We further show that our method is very efficient because the network responses of all possible shifted versions of the same shape can be computed effectively by re-using information in the overlapping areas. We tested the algorithm on three real datasets: Swedish Leaves dataset, ETH-80 Shape, and a subset of the recently collected Leafsnap dataset. Our approach used the curvature scale space and outperformed the state of the art.