Compressed video action recognition classifies video samples by leveraging the different modalities in compressed videos, namely motion vectors, residuals, and intra-frames. For this purpose, three neural networks are deployed, each dedicated to processing one modality. Our observations indicate that the network processing intra-frames tend to converge to a flatter minimum than the network processing residuals, which in turn converges to a flatter minimum than the motion vector network. This hierarchy in convergence motivates our strategy for knowledge transfer among modalities to achieve flatter minima, which are generally associated with better generalization. With this insight, we propose Progressive Knowledge Distillation (PKD), a technique that incrementally transfers knowledge across the modalities. This method involves attaching early exits (Internal Classifiers - ICs) to the three networks. PKD distills knowledge starting from the motion vector network, followed by the residual, and finally, the intra-frame network, sequentially improving IC accuracy. Further, we propose the Weighted Inference with Scaled Ensemble (WISE), which combines outputs from the ICs using learned weights, boosting accuracy during inference. Our experiments demonstrate the effectiveness of training the ICs with PKD compared to standard cross-entropy-based training, showing IC accuracy improvements of up to 5.87% and 11.42% on the UCF-101 and HMDB-51 datasets, respectively. Additionally, WISE improves accuracy by up to 4.28% and 9.30% on UCF-101 and HMDB-51, respectively.
Recent proposed neural network-based Temporal Action Detection (TAD) models are inherently limited to extracting the discriminative representations and modeling action instances with various lengths from complex scenes by shared-weights detection heads. Inspired by the successes in dynamic neural networks, in this paper, we build a novel dynamic feature aggregation (DFA) module that can simultaneously adapt kernel weights and receptive fields at different timestamps. Based on DFA, the proposed dynamic encoder layer aggregates the temporal features within the action time ranges and guarantees the discriminability of the extracted representations. Moreover, using DFA helps to develop a Dynamic TAD head (DyHead), which adaptively aggregates the multi-scale features with adjusted parameters and learned receptive fields better to detect the action instances with diverse ranges from videos. With the proposed encoder layer and DyHead, a new dynamic TAD model, DyFADet, achieves promising performance on a series of challenging TAD benchmarks, including HACS-Segment, THUMOS14, ActivityNet-1.3, Epic-Kitchen 100, Ego4D-Moment QueriesV1.0, and FineAction. Code is released to https://github.com/yangle15/DyFADet-pytorch.
The use of skeletal data allows deep learning models to perform action recognition efficiently and effectively. Herein, we believe that exploring this problem within the context of Continual Learning is crucial. While numerous studies focus on skeleton-based action recognition from a traditional offline perspective, only a handful venture into online approaches. In this respect, we introduce CHARON (Continual Human Action Recognition On skeletoNs), which maintains consistent performance while operating within an efficient framework. Through techniques like uniform sampling, interpolation, and a memory-efficient training stage based on masking, we achieve improved recognition accuracy while minimizing computational overhead. Our experiments on Split NTU-60 and the proposed Split NTU-120 datasets demonstrate that CHARON sets a new benchmark in this domain. The code is available at https://github.com/Sperimental3/CHARON.
We introduce a new task called Referring Atomic Video Action Recognition (RAVAR), aimed at identifying atomic actions of a particular person based on a textual description and the video data of this person. This task differs from traditional action recognition and localization, where predictions are delivered for all present individuals. In contrast, we focus on recognizing the correct atomic action of a specific individual, guided by text. To explore this task, we present the RefAVA dataset, containing 36,630 instances with manually annotated textual descriptions of the individuals. To establish a strong initial benchmark, we implement and validate baselines from various domains, e.g., atomic action localization, video question answering, and text-video retrieval. Since these existing methods underperform on RAVAR, we introduce RefAtomNet -- a novel cross-stream attention-driven method specialized for the unique challenges of RAVAR: the need to interpret a textual referring expression for the targeted individual, utilize this reference to guide the spatial localization and harvest the prediction of the atomic actions for the referring person. The key ingredients are: (1) a multi-stream architecture that connects video, text, and a new location-semantic stream, and (2) cross-stream agent attention fusion and agent token fusion which amplify the most relevant information across these streams and consistently surpasses standard attention-based fusion on RAVAR. Extensive experiments demonstrate the effectiveness of RefAtomNet and its building blocks for recognizing the action of the described individual. The dataset and code will be made publicly available at https://github.com/KPeng9510/RAVAR.
Video understanding requires the extraction of rich spatio-temporal representations, which transformer models achieve through self-attention. Unfortunately, self-attention poses a computational burden. In NLP, Mamba has surfaced as an efficient alternative for transformers. However, Mamba's successes do not trivially extend to vision tasks, including those in video analysis. In this paper, we theoretically analyze the differences between self-attention and Mamba. We identify two limitations in Mamba's token processing: historical decay and element contradiction. We propose VideoMambaPro (VMP) that solves the identified limitations by adding masked backward computation and elemental residual connections to a VideoMamba backbone. Differently sized VideoMambaPro models surpass VideoMamba by 1.6-2.8% and 1.1-1.9% top-1 on Kinetics-400 and Something-Something V2, respectively. Even without extensive pre-training, our models present an increasingly attractive and efficient alternative to current transformer models. Moreover, our two solutions are orthogonal to recent advances in Vision Mamba models, and are likely to provide further improvements in future models.
Recent studies have shown promising results in utilizing multimodal large language models (MLLMs) for computer vision tasks such as object detection and semantic segmentation. However, many challenging video tasks remain under-explored. Video-language tasks necessitate spatial and temporal comprehension and require significant compute. Therefore, prior works have developed complex, highly specialized architectures or leveraged additional input signals such as video transcripts to best encode contextual and temporal information, which limits their generality and can be impractical. One particularly challenging task is video moment retrieval, which requires precise temporal and contextual grounding. This work demonstrates the surprising effectiveness of leveraging image-text pretrained MLLMs for moment retrieval. We introduce Mr. BLIP (Mr. as in Moment Retrieval), a multimodal, single-stage model that requires no expensive video-language pretraining, no additional input signal (e.g., no transcript or audio), and has a simpler and more versatile design than prior state-of-the-art methods. We achieve a new state-of-the-art in moment retrieval on the widely used benchmarks Charades-STA, QVHighlights, and ActivityNet Captions. Notably, we attain over 9% (absolute) higher Recall (at 0.5 and 0.7 IoU) on the challenging long-video multi-moment QVHighlights benchmark. Our code is publicly available.
In the realm of skeleton-based action recognition, the traditional methods which rely on coarse body keypoints fall short of capturing subtle human actions. In this work, we propose Expressive Keypoints that incorporates hand and foot details to form a fine-grained skeletal representation, improving the discriminative ability for existing models in discerning intricate actions. To efficiently model Expressive Keypoints, the Skeleton Transformation strategy is presented to gradually downsample the keypoints and prioritize prominent joints by allocating the importance weights. Additionally, a plug-and-play Instance Pooling module is exploited to extend our approach to multi-person scenarios without surging computation costs. Extensive experimental results over seven datasets present the superiority of our method compared to the state-of-the-art for skeleton-based human action recognition. Code is available at https://github.com/YijieYang23/SkeleT-GCN.
Keypoint data has received a considerable amount of attention in machine learning for tasks like action detection and recognition. However, human experts in movement such as doctors, physiotherapists, sports scientists and coaches use a notion of joint angles standardised by the International Society of Biomechanics to precisely and efficiently communicate static body poses and movements. In this paper, we introduce the basic biomechanical notions and show how they can be used to convert common keypoint data into joint angles that uniquely describe the given pose and have various desirable mathematical properties, such as independence of both the camera viewpoint and the person performing the action. We experimentally demonstrate that the joint angle representation of keypoint data is suitable for machine learning applications and can in some cases bring an immediate performance gain. The use of joint angles as a human meaningful representation of kinematic data is in particular promising for applications where interpretability and dialog with human experts is important, such as many sports and medical applications. To facilitate further research in this direction, we will release a python package to convert keypoint data into joint angles as outlined in this paper.
Open-Vocabulary Temporal Action Localization (OVTAL) enables a model to recognize any desired action category in videos without the need to explicitly curate training data for all categories. However, this flexibility poses significant challenges, as the model must recognize not only the action categories seen during training but also novel categories specified at inference. Unlike standard temporal action localization, where training and test categories are predetermined, OVTAL requires understanding contextual cues that reveal the semantics of novel categories. To address these challenges, we introduce OVFormer, a novel open-vocabulary framework extending ActionFormer with three key contributions. First, we employ task-specific prompts as input to a large language model to obtain rich class-specific descriptions for action categories. Second, we introduce a cross-attention mechanism to learn the alignment between class representations and frame-level video features, facilitating the multimodal guided features. Third, we propose a two-stage training strategy which includes training with a larger vocabulary dataset and finetuning to downstream data to generalize to novel categories. OVFormer extends existing TAL methods to open-vocabulary settings. Comprehensive evaluations on the THUMOS14 and ActivityNet-1.3 benchmarks demonstrate the effectiveness of our method. Code and pretrained models will be publicly released.
Video action recognition (VAR) plays crucial roles in various domains such as surveillance, healthcare, and industrial automation, making it highly significant for the society. Consequently, it has long been a research spot in the computer vision field. As artificial neural networks (ANNs) are flourishing, convolution neural networks (CNNs), including 2D-CNNs and 3D-CNNs, as well as variants of the vision transformer (ViT), have shown impressive performance on VAR. However, they usually demand huge computational cost due to the large data volume and heavy information redundancy introduced by the temporal dimension. To address this challenge, some researchers have turned to brain-inspired spiking neural networks (SNNs), such as recurrent SNNs and ANN-converted SNNs, leveraging their inherent temporal dynamics and energy efficiency. Yet, current SNNs for VAR also encounter limitations, such as nontrivial input preprocessing, intricate network construction/training, and the need for repetitive processing of the same video clip, hindering their practical deployment. In this study, we innovatively propose the directly trained SVFormer (Spiking Video transFormer) for VAR. SVFormer integrates local feature extraction, global self-attention, and the intrinsic dynamics, sparsity, and spike-driven nature of SNNs, to efficiently and effectively extract spatio-temporal features. We evaluate SVFormer on two RGB datasets (UCF101, NTU-RGBD60) and one neuromorphic dataset (DVS128-Gesture), demonstrating comparable performance to the mainstream models in a more efficient way. Notably, SVFormer achieves a top-1 accuracy of 84.03% with ultra-low power consumption (21 mJ/video) on UCF101, which is state-of-the-art among directly trained deep SNNs, showcasing significant advantages over prior models.
Patients with mental disorders often exhibit risky abnormal actions, such as climbing walls or hitting windows, necessitating intelligent video behavior monitoring for smart healthcare with the rising Internet of Things (IoT) technology. However, the development of vision-based Human Action Recognition (HAR) for these actions is hindered by the lack of specialized algorithms and datasets. In this paper, we innovatively propose to build a vision-based HAR dataset including abnormal actions often occurring in the mental disorder group and then introduce a novel Scene-Motion-aware Action Recognition Technology framework, named SMART, consisting of two technical modules. First, we propose a scene perception module to extract human motion trajectory and human-scene interaction features, which introduces additional scene information for a supplementary semantic representation of the above actions. Second, the multi-stage fusion module fuses the skeleton motion, motion trajectory, and human-scene interaction features, enhancing the semantic association between the skeleton motion and the above supplementary representation, thus generating a comprehensive representation with both human motion and scene information. The effectiveness of our proposed method has been validated on our self-collected HAR dataset (MentalHAD), achieving 94.9% and 93.1% accuracy in un-seen subjects and scenes and outperforming state-of-the-art approaches by 6.5% and 13.2%, respectively. The demonstrated subject- and scene- generalizability makes it possible for SMART's migration to practical deployment in smart healthcare systems for mental disorder patients in medical settings. The code and dataset will be released publicly for further research: https://github.com/Inowlzy/SMART.git.
Due to its widespread applications, human action recognition is one of the most widely studied research problems in Computer Vision. Recent studies have shown that addressing it using multimodal data leads to superior performance as compared to relying on a single data modality. During the adoption of deep learning for visual modelling in the last decade, action recognition approaches have mainly relied on Convolutional Neural Networks (CNNs). However, the recent rise of Transformers in visual modelling is now also causing a paradigm shift for the action recognition task. This survey captures this transition while focusing on Multimodal Human Action Recognition (MHAR). Unique to the induction of multimodal computational models is the process of "fusing" the features of the individual data modalities. Hence, we specifically focus on the fusion design aspects of the MHAR approaches. We analyze the classic and emerging techniques in this regard, while also highlighting the popular trends in the adaption of CNN and Transformer building blocks for the overall problem. In particular, we emphasize on recent design choices that have led to more efficient MHAR models. Unlike existing reviews, which discuss Human Action Recognition from a broad perspective, this survey is specifically aimed at pushing the boundaries of MHAR research by identifying promising architectural and fusion design choices to train practicable models. We also provide an outlook of the multimodal datasets from their scale and evaluation viewpoint. Finally, building on the reviewed literature, we discuss the challenges and future avenues for MHAR.
Human action recognition in dark videos is a challenging task for computer vision. Recent research focuses on applying dark enhancement methods to improve the visibility of the video. However, such video processing results in the loss of critical information in the original (un-enhanced) video. Conversely, traditional two-stream methods are capable of learning information from both original and processed videos, but it can lead to a significant increase in the computational cost during the inference phase in the task of video classification. To address these challenges, we propose a novel teacher-student video classification framework, named Dual-Light KnowleDge Distillation for Action Recognition in the Dark (DL-KDD). This framework enables the model to learn from both original and enhanced video without introducing additional computational cost during inference. Specifically, DL-KDD utilizes the strategy of knowledge distillation during training. The teacher model is trained with enhanced video, and the student model is trained with both the original video and the soft target generated by the teacher model. This teacher-student framework allows the student model to predict action using only the original input video during inference. In our experiments, the proposed DL-KDD framework outperforms state-of-the-art methods on the ARID, ARID V1.5, and Dark-48 datasets. We achieve the best performance on each dataset and up to a 4.18% improvement on Dark-48, using only original video inputs, thus avoiding the use of two-stream framework or enhancement modules for inference. We further validate the effectiveness of the distillation strategy in ablative experiments. The results highlight the advantages of our knowledge distillation framework in dark human action recognition.
Human action recognition is crucial in computer vision systems. However, in real-world scenarios, human actions often fall outside the distribution of training data, requiring a model to both recognize in-distribution (ID) actions and reject out-of-distribution (OOD) ones. Despite its importance, there has been limited research on OOD detection in human actions. Existing works on OOD detection mainly focus on image data with RGB structure, and many methods are post-hoc in nature. While these methods are convenient and computationally efficient, they often lack sufficient accuracy, fail to consider the exposure of OOD samples, and ignore the application in skeleton structure data. To address these challenges, we propose a novel end-to-end skeleton-based model called Skeleton-OOD, which is committed to improving the effectiveness of OOD tasks while ensuring the accuracy of ID recognition. Through extensive experiments conducted on NTU-RGB+D 60, NTU-RGB+D 120, and Kinetics-400 datasets, Skeleton-OOD demonstrates the superior performance of our proposed approach compared to state-of-the-art methods. Our findings underscore the effectiveness of classic OOD detection techniques in the context of skeleton-based action recognition tasks, offering promising avenues for future research in this field. Code is available at https://github.com/YilliaJing/Skeleton-OOD.git.
Skeleton-based action representation learning aims to interpret and understand human behaviors by encoding the skeleton sequences, which can be categorized into two primary training paradigms: supervised learning and self-supervised learning. However, the former one-hot classification requires labor-intensive predefined action categories annotations, while the latter involves skeleton transformations (e.g., cropping) in the pretext tasks that may impair the skeleton structure. To address these challenges, we introduce a novel skeleton-based training framework (C$^2$VL) based on Cross-modal Contrastive learning that uses the progressive distillation to learn task-agnostic human skeleton action representation from the Vision-Language knowledge prompts. Specifically, we establish the vision-language action concept space through vision-language knowledge prompts generated by pre-trained large multimodal models (LMMs), which enrich the fine-grained details that the skeleton action space lacks. Moreover, we propose the intra-modal self-similarity and inter-modal cross-consistency softened targets in the cross-modal representation learning process to progressively control and guide the degree of pulling vision-language knowledge prompts and corresponding skeletons closer. These soft instance discrimination and self-knowledge distillation strategies contribute to the learning of better skeleton-based action representations from the noisy skeleton-vision-language pairs. During the inference phase, our method requires only the skeleton data as the input for action recognition and no longer for vision-language prompts. Extensive experiments on NTU RGB+D 60, NTU RGB+D 120, and PKU-MMD datasets demonstrate that our method outperforms the previous methods and achieves state-of-the-art results. Code is available at: https://github.com/cseeyangchen/C2VL.
Deep neural networks (DNNs) on Riemannian manifolds have garnered increasing interest in various applied areas. For instance, DNNs on spherical and hyperbolic manifolds have been designed to solve a wide range of computer vision and nature language processing tasks. One of the key factors that contribute to the success of these networks is that spherical and hyperbolic manifolds have the rich algebraic structures of gyrogroups and gyrovector spaces. This enables principled and effective generalizations of the most successful DNNs to these manifolds. Recently, some works have shown that many concepts in the theory of gyrogroups and gyrovector spaces can also be generalized to matrix manifolds such as Symmetric Positive Definite (SPD) and Grassmann manifolds. As a result, some building blocks for SPD and Grassmann neural networks, e.g., isometric models and multinomial logistic regression (MLR) can be derived in a way that is fully analogous to their spherical and hyperbolic counterparts. Building upon these works, we design fully-connected (FC) and convolutional layers for SPD neural networks. We also develop MLR on Symmetric Positive Semi-definite (SPSD) manifolds, and propose a method for performing backpropagation with the Grassmann logarithmic map in the projector perspective. We demonstrate the effectiveness of the proposed approach in the human action recognition and node classification tasks.
Point cloud videos can faithfully capture real-world spatial geometries and temporal dynamics, which are essential for enabling intelligent agents to understand the dynamically changing world. However, designing an effective 4D backbone remains challenging, mainly due to the irregular and unordered distribution of points and temporal inconsistencies across frames. Also, recent transformer-based 4D backbones commonly suffer from large computational costs due to their quadratic complexity, particularly for long video sequences.To address these challenges, we propose a novel point cloud video understanding backbone purely based on the State Space Models (SSMs). Specifically, we first disentangle space and time in 4D video sequences and then establish the spatio-temporal correlation with our designed Mamba blocks. The Intra-frame Spatial Mamba module is developed to encode locally similar geometric structures within a certain temporal stride. Subsequently, locally correlated tokens are delivered to the Inter-frame Temporal Mamba module, which integrates long-term point features across the entire video with linear complexity. Our proposed Mamba4d achieves competitive performance on the MSR-Action3D action recognition (+10.4% accuracy), HOI4D action segmentation (+0.7 F1 Score), and Synthia4D semantic segmentation (+0.19 mIoU) datasets. Especially, for long video sequences, our method has a significant efficiency improvement with 87.5% GPU memory reduction and 5.36 times speed-up.
The widespread adoption of deep neural networks in machine learning calls for an objective quantification of esoteric trust. In this paper we propose GradTrust, a classification trust measure for large-scale neural networks at inference. The proposed method utilizes variance of counterfactual gradients, i.e. the required changes in the network parameters if the label were different. We show that GradTrust is superior to existing techniques for detecting misprediction rates on $50000$ images from ImageNet validation dataset. Depending on the network, GradTrust detects images where either the ground truth is incorrect or ambiguous, or the classes are co-occurring. We extend GradTrust to Video Action Recognition on Kinetics-400 dataset. We showcase results on $14$ architectures pretrained on ImageNet and $5$ architectures pretrained on Kinetics-400. We observe the following: (i) simple methodologies like negative log likelihood and margin classifiers outperform state-of-the-art uncertainty and out-of-distribution detection techniques for misprediction rates, and (ii) the proposed GradTrust is in the Top-2 performing methods on $37$ of the considered $38$ experimental modalities. The code is available at: https://github.com/olivesgatech/GradTrust
Spatiotemporal prediction plays an important role in solving natural problems and processing video frames, especially in weather forecasting and human action recognition. Recent advances attempt to incorporate prior physical knowledge into the deep learning framework to estimate the unknown governing partial differential equations (PDEs), which have shown promising results in spatiotemporal prediction tasks. However, previous approaches only restrict neural network architectures or loss functions to acquire physical or PDE features, which decreases the representative capacity of a neural network. Meanwhile, the updating process of the physical state cannot be effectively estimated. To solve the above mentioned problems, this paper proposes a physical-guided neural network, which utilizes the frequency-enhanced Fourier module and moment loss to strengthen the model's ability to estimate the spatiotemporal dynamics. Furthermore, we propose an adaptive second-order Runge-Kutta method with physical constraints to model the physical states more precisely. We evaluate our model on both spatiotemporal and video prediction tasks. The experimental results show that our model outperforms state-of-the-art methods and performs best in several datasets, with a much smaller parameter count.
Action recognition is a key technology in building interactive metaverses. With the rapid development of deep learning, methods in action recognition have also achieved great advancement. Researchers design and implement the backbones referring to multiple standpoints, which leads to the diversity of methods and encountering new challenges. This paper reviews several action recognition methods based on deep neural networks. We introduce these methods in three parts: 1) Two-Streams networks and their variants, which, specifically in this paper, use RGB video frame and optical flow modality as input; 2) 3D convolutional networks, which make efforts in taking advantage of RGB modality directly while extracting different motion information is no longer necessary; 3) Transformer-based methods, which introduce the model from natural language processing into computer vision and video understanding. We offer objective sights in this review and hopefully provide a reference for future research.