Previous work on long-term video action recognition relies on deep 3D-convolutional models that have a large temporal receptive field (RF). We argue that these models are not always the best choice for temporal modeling in videos. A large temporal receptive field allows the model to encode the exact sub-action order of a video, which causes a performance decrease when testing videos have a different sub-action order. In this work, we investigate whether we can improve the model robustness to the sub-action order by shrinking the temporal receptive field of action recognition models. For this, we design Video BagNet, a variant of the 3D ResNet-50 model with the temporal receptive field size limited to 1, 9, 17 or 33 frames. We analyze Video BagNet on synthetic and real-world video datasets and experimentally compare models with varying temporal receptive fields. We find that short receptive fields are robust to sub-action order changes, while larger temporal receptive fields are sensitive to the sub-action order.
While large-scale image-text pretrained models such as CLIP have been used for multiple video-level tasks on trimmed videos, their use for temporal localization in untrimmed videos is still a relatively unexplored task. We design a new approach for this called UnLoc, which uses pretrained image and text towers, and feeds tokens to a video-text fusion model. The output of the fusion module are then used to construct a feature pyramid in which each level connects to a head to predict a per-frame relevancy score and start/end time displacements. Unlike previous works, our architecture enables Moment Retrieval, Temporal Localization, and Action Segmentation with a single stage model, without the need for action proposals, motion based pretrained features or representation masking. Unlike specialized models, we achieve state of the art results on all three different localization tasks with a unified approach. Code will be available at: \url{https://github.com/google-research/scenic}.
In 3D human action recognition, limited supervised data makes it challenging to fully tap into the modeling potential of powerful networks such as transformers. As a result, researchers have been actively investigating effective self-supervised pre-training strategies. In this work, we show that instead of following the prevalent pretext task to perform masked self-component reconstruction in human joints, explicit contextual motion modeling is key to the success of learning effective feature representation for 3D action recognition. Formally, we propose the Masked Motion Prediction (MAMP) framework. To be specific, the proposed MAMP takes as input the masked spatio-temporal skeleton sequence and predicts the corresponding temporal motion of the masked human joints. Considering the high temporal redundancy of the skeleton sequence, in our MAMP, the motion information also acts as an empirical semantic richness prior that guide the masking process, promoting better attention to semantically rich temporal regions. Extensive experiments on NTU-60, NTU-120, and PKU-MMD datasets show that the proposed MAMP pre-training substantially improves the performance of the adopted vanilla transformer, achieving state-of-the-art results without bells and whistles. The source code of our MAMP is available at https://github.com/maoyunyao/MAMP.
Recently, methods for skeleton-based human activity recognition have been shown to be vulnerable to adversarial attacks. However, these attack methods require either the full knowledge of the victim (i.e. white-box attacks), access to training data (i.e. transfer-based attacks) or frequent model queries (i.e. black-box attacks). All their requirements are highly restrictive, raising the question of how detrimental the vulnerability is. In this paper, we show that the vulnerability indeed exists. To this end, we consider a new attack task: the attacker has no access to the victim model or the training data or labels, where we coin the term hard no-box attack. Specifically, we first learn a motion manifold where we define an adversarial loss to compute a new gradient for the attack, named skeleton-motion-informed (SMI) gradient. Our gradient contains information of the motion dynamics, which is different from existing gradient-based attack methods that compute the loss gradient assuming each dimension in the data is independent. The SMI gradient can augment many gradient-based attack methods, leading to a new family of no-box attack methods. Extensive evaluation and comparison show that our method imposes a real threat to existing classifiers. They also show that the SMI gradient improves the transferability and imperceptibility of adversarial samples in both no-box and transfer-based black-box settings.
Video Action Recognition (VAR) is a challenging task due to its inherent complexities. Though different approaches have been explored in the literature, designing a unified framework to recognize a large number of human actions is still a challenging problem. Recently, Multi-Modal Learning (MML) has demonstrated promising results in this domain. In literature, 2D skeleton or pose modality has often been used for this task, either independently or in conjunction with the visual information (RGB modality) present in videos. However, the combination of pose, visual information, and text attributes has not been explored yet, though text and pose attributes independently have been proven to be effective in numerous computer vision tasks. In this paper, we present the first pose augmented Vision-language model (VLM) for VAR. Notably, our scheme achieves an accuracy of 92.81% and 73.02% on two popular human video action recognition benchmark datasets, UCF-101 and HMDB-51, respectively, even without any video data pre-training, and an accuracy of 96.11% and 75.75% after kinetics pre-training.
Computer vision was long a tool used for aiding visually impaired people to move around their environment and avoid obstacles and falls. Solutions are limited to either indoor or outdoor scenes, which limits the kind of places and scenes visually disabled people can be in, including entertainment places such as theatres. Furthermore, most of the proposed computer-vision-based methods rely on RGB benchmarks to train their models resulting in a limited performance due to the absence of the depth modality. In this paper, we propose a novel RGB-D dataset containing theatre scenes with ground truth human actions and dense captions annotations for image captioning and human action recognition: TS-RGBD dataset. It includes three types of data: RGB, depth, and skeleton sequences, captured by Microsoft Kinect. We test image captioning models on our dataset as well as some skeleton-based human action recognition models in order to extend the range of environment types where a visually disabled person can be, by detecting human actions and textually describing appearances of regions of interest in theatre scenes.
Weakly-supervised temporal action localization (WTAL) is a practical yet challenging task. Due to large-scale datasets, most existing methods use a network pretrained in other datasets to extract features, which are not suitable enough for WTAL. To address this problem, researchers design several modules for feature enhancement, which improve the performance of the localization module, especially modeling the temporal relationship between snippets. However, all of them neglect the adverse effects of ambiguous information, which would reduce the discriminability of others. Considering this phenomenon, we propose Discriminability-Driven Graph Network (DDG-Net), which explicitly models ambiguous snippets and discriminative snippets with well-designed connections, preventing the transmission of ambiguous information and enhancing the discriminability of snippet-level representations. Additionally, we propose feature consistency loss to prevent the assimilation of features and drive the graph convolution network to generate more discriminative representations. Extensive experiments on THUMOS14 and ActivityNet1.2 benchmarks demonstrate the effectiveness of DDG-Net, establishing new state-of-the-art results on both datasets. Source code is available at \url{https://github.com/XiaojunTang22/ICCV2023-DDGNet}.
Training an effective video action recognition model poses significant computational challenges, particularly under limited resource budgets. Current methods primarily aim to either reduce model size or utilize pre-trained models, limiting their adaptability to various backbone architectures. This paper investigates the issue of over-sampled frames, a prevalent problem in many approaches yet it has received relatively little attention. Despite the use of fewer frames being a potential solution, this approach often results in a substantial decline in performance. To address this issue, we propose a novel method to restore the intermediate features for two sparsely sampled and adjacent video frames. This feature restoration technique brings a negligible increase in computational requirements compared to resource-intensive image encoders, such as ViT. To evaluate the effectiveness of our method, we conduct extensive experiments on four public datasets, including Kinetics-400, ActivityNet, UCF-101, and HMDB-51. With the integration of our method, the efficiency of three commonly used baselines has been improved by over 50%, with a mere 0.5% reduction in recognition accuracy. In addition, our method also surprisingly helps improve the generalization ability of the models under zero-shot settings.
Considering the instance-level discriminative ability, contrastive learning methods, including MoCo and SimCLR, have been adapted from the original image representation learning task to solve the self-supervised skeleton-based action recognition task. These methods usually use multiple data streams (i.e., joint, motion, and bone) for ensemble learning, meanwhile, how to construct a discriminative feature space within a single stream and effectively aggregate the information from multiple streams remains an open problem. To this end, this paper first applies a new contrastive learning method called BYOL to learn from skeleton data, and then formulate SkeletonBYOL as a simple yet effective baseline for self-supervised skeleton-based action recognition. Inspired by SkeletonBYOL, this paper further presents a Cross-Model and Cross-Stream (CMCS) framework. This framework combines Cross-Model Adversarial Learning (CMAL) and Cross-Stream Collaborative Learning (CSCL). Specifically, CMAL learns single-stream representation by cross-model adversarial loss to obtain more discriminative features. To aggregate and interact with multi-stream information, CSCL is designed by generating similarity pseudo label of ensemble learning as supervision and guiding feature generation for individual streams. Extensive experiments on three datasets verify the complementary properties between CMAL and CSCL and also verify that the proposed method can achieve better results than state-of-the-art methods using various evaluation protocols.
Recent video recognition models utilize Transformer models for long-range spatio-temporal context modeling. Video transformer designs are based on self-attention that can model global context at a high computational cost. In comparison, convolutional designs for videos offer an efficient alternative but lack long-range dependency modeling. Towards achieving the best of both designs, this work proposes Video-FocalNet, an effective and efficient architecture for video recognition that models both local and global contexts. Video-FocalNet is based on a spatio-temporal focal modulation architecture that reverses the interaction and aggregation steps of self-attention for better efficiency. Further, the aggregation step and the interaction step are both implemented using efficient convolution and element-wise multiplication operations that are computationally less expensive than their self-attention counterparts on video representations. We extensively explore the design space of focal modulation-based spatio-temporal context modeling and demonstrate our parallel spatial and temporal encoding design to be the optimal choice. Video-FocalNets perform favorably well against the state-of-the-art transformer-based models for video recognition on five large-scale datasets (Kinetics-400, Kinetics-600, SS-v2, Diving-48, and ActivityNet-1.3) at a lower computational cost. Our code/models are released at https://github.com/TalalWasim/Video-FocalNets.
This report describes our submission to the Ego4D Moment Queries Challenge 2023. Our submission extends ActionFormer, a latest method for temporal action localization. Our extension combines an improved ground-truth assignment strategy during training and a refined version of SoftNMS at inference time. Our solution is ranked 2nd on the public leaderboard with 26.62% average mAP and 45.69% Recall@1x at tIoU=0.5 on the test set, significantly outperforming the strong baseline from 2023 challenge. Our code is available at https://github.com/happyharrycn/actionformer_release.
Recent technological advancements have significantly expanded the potential of human action recognition through harnessing the power of 3D data. This data provides a richer understanding of actions, including depth information that enables more accurate analysis of spatial and temporal characteristics. In this context, We study the challenge of 3D human action recognition.Unlike prior methods, that rely on sampling 2D depth images, skeleton points, or point clouds, often leading to substantial memory requirements and the ability to handle only short sequences, we introduce a novel approach for 3D human action recognition, denoted as SpATr (Spiral Auto-encoder and Transformer Network), specifically designed for fixed-topology mesh sequences. The SpATr model disentangles space and time in the mesh sequences. A lightweight auto-encoder, based on spiral convolutions, is employed to extract spatial geometrical features from each 3D mesh. These convolutions are lightweight and specifically designed for fix-topology mesh data. Subsequently, a temporal transformer, based on self-attention, captures the temporal context within the feature sequence. The self-attention mechanism enables long-range dependencies capturing and parallel processing, ensuring scalability for long sequences. The proposed method is evaluated on three prominent 3D human action datasets: Babel, MoVi, and BMLrub, from the Archive of Motion Capture As Surface Shapes (AMASS). Our results analysis demonstrates the competitive performance of our SpATr model in 3D human action recognition while maintaining efficient memory usage. The code and the training results will soon be made publicly available at https://github.com/h-bouzid/spatr.
Human perception of surroundings is often guided by the various poses present within the environment. Many computer vision tasks, such as human action recognition and robot imitation learning, rely on pose-based entities like human skeletons or robotic arms. However, conventional Vision Transformer (ViT) models uniformly process all patches, neglecting valuable pose priors in input videos. We argue that incorporating poses into RGB data is advantageous for learning fine-grained and viewpoint-agnostic representations. Consequently, we introduce two strategies for learning pose-aware representations in ViTs. The first method, called Pose-aware Attention Block (PAAB), is a plug-and-play ViT block that performs localized attention on pose regions within videos. The second method, dubbed Pose-Aware Auxiliary Task (PAAT), presents an auxiliary pose prediction task optimized jointly with the primary ViT task. Although their functionalities differ, both methods succeed in learning pose-aware representations, enhancing performance in multiple diverse downstream tasks. Our experiments, conducted across seven datasets, reveal the efficacy of both pose-aware methods on three video analysis tasks, with PAAT holding a slight edge over PAAB. Both PAAT and PAAB surpass their respective backbone Transformers by up to 9.8% in real-world action recognition and 21.8% in multi-view robotic video alignment. Code is available at https://github.com/dominickrei/PoseAwareVT.
Deep neural networks (DNNs) are widely used models for investigating biological visual representations. However, existing DNNs are mostly designed to analyze neural responses to static images, relying on feedforward structures and lacking physiological neuronal mechanisms. There is limited insight into how the visual cortex represents natural movie stimuli that contain context-rich information. To address these problems, this work proposes the long-range feedback spiking network (LoRaFB-SNet), which mimics top-down connections between cortical regions and incorporates spike information processing mechanisms inherent to biological neurons. Taking into account the temporal dependence of representations under movie stimuli, we present Time-Series Representational Similarity Analysis (TSRSA) to measure the similarity between model representations and visual cortical representations of mice. LoRaFB-SNet exhibits the highest level of representational similarity, outperforming other well-known and leading alternatives across various experimental paradigms, especially when representing long movie stimuli. We further conduct experiments to quantify how temporal structures (dynamic information) and static textures (static information) of the movie stimuli influence representational similarity, suggesting that our model benefits from long-range feedback to encode context-dependent representations just like the brain. Altogether, LoRaFB-SNet is highly competent in capturing both dynamic and static representations of the mouse visual cortex and contributes to the understanding of movie processing mechanisms of the visual system. Our codes are available at https://github.com/Grasshlw/SNN-Neural-Similarity-Movie.
Recently, the significant achievements have been made in skeleton-based human action recognition with the emergence of graph convolutional networks (GCNs). However, the state-of-the-art (SOTA) models used for this task focus on constructing more complex higher-order connections between joint nodes to describe skeleton information, which leads to complex inference processes and high computational costs. To address the slow inference speed caused by overly complex model structures, we introduce re-parameterization and over-parameterization techniques to GCNs and propose two novel high-performance inference GCNs, namely HPI-GCN-RP and HPI-GCN-OP. After the completion of model training, model parameters are fixed. HPI-GCN-RP adopts re-parameterization technique to transform high-performance training model into fast inference model through linear transformations, which achieves a higher inference speed with competitive model performance. HPI-GCN-OP further utilizes over-parameterization technique to achieve higher performance improvement by introducing additional inference parameters, albeit with slightly decreased inference speed. The experimental results on the two skeleton-based action recognition datasets demonstrate the effectiveness of our approach. Our HPI-GCN-OP achieves performance comparable to the current SOTA models, with inference speeds five times faster. Specifically, our HPI-GCN-OP achieves an accuracy of 93\% on the cross-subject split of the NTU-RGB+D 60 dataset, and 90.1\% on the cross-subject benchmark of the NTU-RGB+D 120 dataset. Code is available at github.com/lizaowo/HPI-GCN.
Weakly-supervised temporal action localization aims to localize and recognize actions in untrimmed videos with only video-level category labels during training. Without instance-level annotations, most existing methods follow the Segment-based Multiple Instance Learning (S-MIL) framework, where the predictions of segments are supervised by the labels of videos. However, the objective for acquiring segment-level scores during training is not consistent with the target for acquiring proposal-level scores during testing, leading to suboptimal results. To deal with this problem, we propose a novel Proposal-based Multiple Instance Learning (P-MIL) framework that directly classifies the candidate proposals in both the training and testing stages, which includes three key designs: 1) a surrounding contrastive feature extraction module to suppress the discriminative short proposals by considering the surrounding contrastive information, 2) a proposal completeness evaluation module to inhibit the low-quality proposals with the guidance of the completeness pseudo labels, and 3) an instance-level rank consistency loss to achieve robust detection by leveraging the complementarity of RGB and FLOW modalities. Extensive experimental results on two challenging benchmarks including THUMOS14 and ActivityNet demonstrate the superior performance of our method.
Deep neural networks for learning Symmetric Positive Definite (SPD) matrices are gaining increasing attention in machine learning. Despite the significant progress, most existing SPD networks use traditional Euclidean classifiers on an approximated space rather than intrinsic classifiers that accurately capture the geometry of SPD manifolds. Inspired by Hyperbolic Neural Networks (HNNs), we propose Riemannian Multinomial Logistics Regression (RMLR) for the classification layers in SPD networks. We introduce a unified framework for building Riemannian classifiers under the metrics pulled back from the Euclidean space, and showcase our framework under the parameterized Log-Euclidean Metric (LEM) and Log-Cholesky Metric (LCM). Besides, our framework offers a novel intrinsic explanation for the most popular LogEig classifier in existing SPD networks. The effectiveness of our method is demonstrated in three applications: radar recognition, human action recognition, and electroencephalography (EEG) classification. The code is available at https://github.com/GitZH-Chen/SPDMLR.git.
Due to the lack of temporal annotation, current Weakly-supervised Temporal Action Localization (WTAL) methods are generally stuck into over-complete or incomplete localization. In this paper, we aim to leverage the text information to boost WTAL from two aspects, i.e., (a) the discriminative objective to enlarge the inter-class difference, thus reducing the over-complete; (b) the generative objective to enhance the intra-class integrity, thus finding more complete temporal boundaries. For the discriminative objective, we propose a Text-Segment Mining (TSM) mechanism, which constructs a text description based on the action class label, and regards the text as the query to mine all class-related segments. Without the temporal annotation of actions, TSM compares the text query with the entire videos across the dataset to mine the best matching segments while ignoring irrelevant ones. Due to the shared sub-actions in different categories of videos, merely applying TSM is too strict to neglect the semantic-related segments, which results in incomplete localization. We further introduce a generative objective named Video-text Language Completion (VLC), which focuses on all semantic-related segments from videos to complete the text sentence. We achieve the state-of-the-art performance on THUMOS14 and ActivityNet1.3. Surprisingly, we also find our proposed method can be seamlessly applied to existing methods, and improve their performances with a clear margin. The code is available at https://github.com/lgzlIlIlI/Boosting-WTAL.
Weakly Supervised Temporal Action Localization (WTAL) aims to classify and localize temporal boundaries of actions for the video, given only video-level category labels in the training datasets. Due to the lack of boundary information during training, existing approaches formulate WTAL as a classificationproblem, i.e., generating the temporal class activation map (T-CAM) for localization. However, with only classification loss, the model would be sub-optimized, i.e., the action-related scenes are enough to distinguish different class labels. Regarding other actions in the action-related scene ( i.e., the scene same as positive actions) as co-scene actions, this sub-optimized model would misclassify the co-scene actions as positive actions. To address this misclassification, we propose a simple yet efficient method, named bidirectional semantic consistency constraint (Bi-SCC), to discriminate the positive actions from co-scene actions. The proposed Bi-SCC firstly adopts a temporal context augmentation to generate an augmented video that breaks the correlation between positive actions and their co-scene actions in the inter-video; Then, a semantic consistency constraint (SCC) is used to enforce the predictions of the original video and augmented video to be consistent, hence suppressing the co-scene actions. However, we find that this augmented video would destroy the original temporal context. Simply applying the consistency constraint would affect the completeness of localized positive actions. Hence, we boost the SCC in a bidirectional way to suppress co-scene actions while ensuring the integrity of positive actions, by cross-supervising the original and augmented videos. Finally, our proposed Bi-SCC can be applied to current WTAL approaches, and improve their performance. Experimental results show that our approach outperforms the state-of-the-art methods on THUMOS14 and ActivityNet.
Skeleton-based action recognition has achieved remarkable results in human action recognition with the development of graph convolutional networks (GCNs). However, the recent works tend to construct complex learning mechanisms with redundant training and exist a bottleneck for long time-series. To solve these problems, we propose the Temporal-Spatio Graph ConvNeXt (TSGCNeXt) to explore efficient learning mechanism of long temporal skeleton sequences. Firstly, a new graph learning mechanism with simple structure, Dynamic-Static Separate Multi-graph Convolution (DS-SMG) is proposed to aggregate features of multiple independent topological graphs and avoid the node information being ignored during dynamic convolution. Next, we construct a graph convolution training acceleration mechanism to optimize the back-propagation computing of dynamic graph learning with 55.08\% speed-up. Finally, the TSGCNeXt restructure the overall structure of GCN with three Spatio-temporal learning modules,efficiently modeling long temporal features. In comparison with existing previous methods on large-scale datasets NTU RGB+D 60 and 120, TSGCNeXt outperforms on single-stream networks. In addition, with the ema model introduced into the multi-stream fusion, TSGCNeXt achieves SOTA levels. On the cross-subject and cross-set of the NTU 120, accuracies reach 90.22% and 91.74%.