How far are deep models from real-world video anomaly understanding (VAU)? Current works typically emphasize detecting unexpected occurrences deviating from normal patterns or comprehending anomalous events with interpretable descriptions. However, they exhibit only a superficial comprehension of real-world anomalies, with limited breadth in complex principles and subtle contexts that distinguish the anomalies from normalities, e.g., climbing cliffs with safety gear vs. without it. To this end, we introduce CueBench, the first of its kind Benchmark, devoted to Context-aware video anomalies within a Unified Evaluation framework. We comprehensively establish an event-centric hierarchical taxonomy that anchors two core event types: 14 conditional and 18 absolute anomaly events, defined by their refined semantics from diverse contexts across 174 scenes and 198 attributes. Based on this, we propose to unify and benchmark context-aware VAU with various challenging tasks across recognition, temporal grounding, detection, and anticipation. It also serves as a rigorous and fair probing evaluation suite for generalized and specialized vision-language models (VLMs) across both generative and discriminative paradigms. To address the challenges underlying CueBench, we further develop Cue-R1 based on R1-style reinforcement fine-tuning with verifiable, task-aligned, and hierarchy-refined rewards in a unified generative manner. Extensive results on CueBench reveal that, existing VLMs are still far from satisfactory real-world anomaly understanding, while our Cue-R1 surpasses these state-of-the-art approaches by over 24% on average.

Zero-shot action recognition (ZSAR) requires collaborative multi-modal spatiotemporal understanding. However, finetuning CLIP directly for ZSAR yields suboptimal performance, given its inherent constraints in capturing essential temporal dynamics from both vision and text perspectives, especially when encountering novel actions with fine-grained spatiotemporal discrepancies. In this work, we propose Spatiotemporal Dynamic Duo (STDD), a novel CLIP-based framework to comprehend multi-modal spatiotemporal dynamics synergistically. For the vision side, we propose an efficient Space-time Cross Attention, which captures spatiotemporal dynamics flexibly with simple yet effective operations applied before and after spatial attention, without adding additional parameters or increasing computational complexity. For the semantic side, we conduct spatiotemporal text augmentation by comprehensively constructing an Action Semantic Knowledge Graph (ASKG) to derive nuanced text prompts. The ASKG elaborates on static and dynamic concepts and their interrelations, based on the idea of decomposing actions into spatial appearances and temporal motions. During the training phase, the frame-level video representations are meticulously aligned with prompt-level nuanced text representations, which are concurrently regulated by the video representations from the frozen CLIP to enhance generalizability. Extensive experiments validate the effectiveness of our approach, which consistently surpasses state-of-the-art approaches on popular video benchmarks (i.e., Kinetics-600, UCF101, and HMDB51) under challenging ZSAR settings.

Leveraging the effective visual-text alignment and static generalizability from CLIP, recent video learners adopt CLIP initialization with further regularization or recombination for generalization in open-vocabulary action recognition in-context. However, due to the static bias of CLIP, such video learners tend to overfit on shortcut static features, thereby compromising their generalizability, especially to novel out-of-context actions. To address this issue, we introduce Open-MeDe, a novel Meta-optimization framework with static Debiasing for Open-vocabulary action recognition. From a fresh perspective of generalization, Open-MeDe adopts a meta-learning approach to improve known-to-open generalizing and image-to-video debiasing in a cost-effective manner. Specifically, Open-MeDe introduces a cross-batch meta-optimization scheme that explicitly encourages video learners to quickly generalize to arbitrary subsequent data via virtual evaluation, steering a smoother optimization landscape. In effect, the free of CLIP regularization during optimization implicitly mitigates the inherent static bias of the video meta-learner. We further apply self-ensemble over the optimization trajectory to obtain generic optimal parameters that can achieve robust generalization to both in-context and out-of-context novel data. Extensive evaluations show that Open-MeDe not only surpasses state-of-the-art regularization methods tailored for in-context open-vocabulary action recognition but also substantially excels in out-of-context scenarios.Code is released at https://github.com/Mia-YatingYu/Open-MeDe.

Video anomaly detection (VAD) is an important computer vision problem. Thanks to the mode coverage capabilities of generative models, the likelihood-based paradigm is catching growing interest, as it can model normal distribution and detect out-of-distribution anomalies. However, these likelihood-based methods are blind to the anomalies located in local modes near the learned distribution. To handle these ``unseen" anomalies, we dive into three gaps uniquely existing in VAD regarding scene, motion and appearance. Specifically, we first build a noise-conditioned score transformer for denoising score matching. Then, we introduce a scene-dependent and motion-aware score function by embedding the scene condition of input sequences into our model and assigning motion weights based on the difference between key frames of input sequences. Next, to solve the problem of blindness in principle, we integrate unaffected visual information via a novel autoregressive denoising score matching mechanism for inference. Through autoregressively injecting intensifying Gaussian noise into the denoised data and estimating the corresponding score function, we compare the denoised data with the original data to get a difference and aggregate it with the score function for an enhanced appearance perception and accumulate the abnormal context. With all three gaps considered, we can compute a more comprehensive anomaly indicator. Experiments on three popular VAD benchmarks demonstrate the state-of-the-art performance of our method.

—Weakly supervised video anomaly detection (WS-VAD) is challenging because it relies on video-level binary annotations to make frame-level predictions. Existing methods often convert WSVAD into a multiple instance learning (MIL) task, focusing on isolated segments that contribute most to the classification while neglecting the temporal context and detailed feature distinctions. In this paper, we propose a contrastive clustering strategy that enhances the representation of normal and abnormal features. Specifically, we treat the clustering center features and their corresponding categories as positive sample pairs, while features from different categories are treated as negative samples. This approach enables the network to better explore the distinction between normal and abnormal features. Furthermore, we address the bias in pre-trained models, where I3D pre-training features tend to overfit to normal videos and CLIP features exhibit a bias towards abnormal videos. To mitigate this, we introduce a simple early fusion method that combines pre-trained features to eliminate bias and obtain more comprehensive spatio-temporal representations. Extensive experiments on the UCF-Crime and XD-Violence datasets demonstrate the effectiveness of our approach, achieving state-of-the-art performance.

Video anomaly detection (VAD) plays a crucial role in intelligent surveillance. However, an essential type of anomaly named scene-dependent anomaly is overlooked. Moreover, the task of video anomaly anticipation (VAA) also deserves attention. To fill these gaps, we build a comprehensive dataset named NWPU Campus, which is the largest semi-supervised VAD dataset and the first dataset for scene-dependent VAD and VAA. Meanwhile, we introduce a novel forward-backward framework for scene-dependent VAD and VAA, in which the forward network individually solves the VAD and jointly solves the VAA with the backward network. Particularly, we propose a scene-dependent generative model in latent space for the forward and backward networks. First, we propose a hierarchical variational auto-encoder to extract scene-generic features. Next, we design a score-based diffusion model in latent space to refine these features more compact for the task and generate scene-dependent features with a scene information auto-encoder, modeling the relationships between video events and scenes. Finally, we develop a temporal loss from key frames to constrain the motion consistency of video clips. Extensive experiments demonstrate that our method can handle both scene-dependent anomaly detection and anticipation well, achieving state-of-the-art performance on ShanghaiTech, CUHK Avenue, and the proposed NWPU Campus datasets.

Egocentric action anticipation is a challenging task that aims to make advanced predictions of future actions from current and historical observations in the first-person view. Most existing methods focus on improving the model architecture and loss function based on the visual input and recurrent neural network to boost the anticipation performance. However, these methods, which merely consider visual information and rely on a single network architecture, gradually reach a performance plateau. In order to fully understand what has been observed and capture the dependencies between current observations and future actions well enough, we propose a novel visual-semantic fusion enhanced and Transformer-GRU-based action anticipation framework in this paper. Firstly, high-level semantic information is introduced to improve the performance of action anticipation for the first time. We propose to use the semantic features generated based on the class labels or directly from the visual observations to augment the original visual features. Secondly, to take advantage of both the parallel and autoregressive models, we design a Transformer-based encoder for long-term sequential modeling and a GRU-based decoder for flexible iteration decoding. This hybrid architecture allows for better performance with fewer parameters and computations. Thirdly, an effective visual-semantic fusion module is proposed to make up for the semantic gap and fully utilize the complementarity of different modalities. Extensive experiments on two large-scale first-person view datasets and two third-person datasets validate the effectiveness of our proposed method, which achieves new state-of-the-art performance, outperforming previous approaches by a large margin. The code will be released after acceptance athttps://github.com/sunze992/VS-TransGRU.

Temporal action localization (TAL) is a prevailing task due to its great application potential. Existing works in this field mainly suffer from two weaknesses: (1) They often neglect the multi-label case and only focus on temporal modeling. (2) They ignore the semantic information in class labels and only use the visual information. To solve these problems, we propose a novel Co-Occurrence Relation Module (CORM) that explicitly models the co-occurrence relationship between actions. Besides the visual information, it further utilizes the semantic embeddings of class labels to model the co-occurrence relationship. The CORM works in a plug-and-play manner and can be easily incorporated with the existing sequence models. By considering both visual and semantic co-occurrence, our method achieves high multi-label relationship modeling capacity. Meanwhile, existing datasets in TAL always focus on low-semantic atomic actions. Thus we construct a challenging multi-label dataset UCF-Crime-TAL that focuses on high-semantic actions by annotating the UCF-Crime dataset at frame level and considering the semantic overlap of different events. Extensive experiments on two commonly used TAL datasets, i.e., MultiTHUMOS and TSU, and our newly proposed UCF-Crime-TAL demenstrate the effectiveness of the proposed CORM, which achieves state-of-the-art performance on these datasets.

Video anomaly detection aims to find the events in a video that do not conform to the expected behavior. The prevalent methods mainly detect anomalies by snippet reconstruction or future frame prediction error. However, the error is highly dependent on the local context of the current snippet and lacks the understanding of normality. To address this issue, we propose to detect anomalous events not only by the local context, but also according to the consistency between the testing event and the knowledge about normality from the training data. Concretely, we propose a novel two-stream framework based on context recovery and knowledge retrieval, where the two streams can complement each other. For the context recovery stream, we propose a spatiotemporal U-Net which can fully utilize the motion information to predict the future frame. Furthermore, we propose a maximum local error mechanism to alleviate the problem of large recovery errors caused by complex foreground objects. For the knowledge retrieval stream, we propose an improved learnable locality-sensitive hashing, which optimizes hash functions via a Siamese network and a mutual difference loss. The knowledge about normality is encoded and stored in hash tables, and the distance between the testing event and the knowledge representation is used to reveal the probability of anomaly. Finally, we fuse the anomaly scores from the two streams to detect anomalies. Extensive experiments demonstrate the effectiveness and complementarity of the two streams, whereby the proposed two-stream framework achieves state-of-the-art performance on ShanghaiTech, Avenue and Corridor datasets among the methods without object detection. Even if compared with the methods using object detection, our method reaches competitive or better performance on the ShanghaiTech, Avenue, and Ped2 datasets.

Semi-supervised video anomaly detection (VAD) is a critical task in the intelligent surveillance system. However, an essential type of anomaly in VAD named scene-dependent anomaly has not received the attention of researchers. Moreover, there is no research investigating anomaly anticipation, a more significant task for preventing the occurrence of anomalous events. To this end, we propose a new comprehensive dataset, NWPU Campus, containing 43 scenes, 28 classes of abnormal events, and 16 hours of videos. At present, it is the largest semi-supervised VAD dataset with the largest number of scenes and classes of anomalies, the longest duration, and the only one considering the scene-dependent anomaly. Meanwhile, it is also the first dataset proposed for video anomaly anticipation. We further propose a novel model capable of detecting and anticipating anomalous events simultaneously. Compared with 7 outstanding VAD algorithms in recent years, our method can cope with scene-dependent anomaly detection and anomaly anticipation both well, achieving state-of-the-art performance on ShanghaiTech, CUHK Avenue, IITB Corridor and the newly proposed NWPU Campus datasets consistently. Our dataset and code is available at: https://campusvad.github.io.

Weakly supervised video anomaly detection (WSVAD) is a challenging task since only video-level labels are available for training. In previous studies, the discriminative power of the learned features is not strong enough, and the data imbalance resulting from the mini-batch training strategy is ignored. To address these two issues, we propose a novel WSVAD method based on cross-batch clustering guidance. To enhance the discriminative power of features, we propose a batch clustering based loss to encourage a clustering branch to generate distinct normal and abnormal clusters based on a batch of data. Meanwhile, we design a cross-batch learning strategy by introducing clustering results from previous minibatches to reduce the impact of data imbalance. In addition, we propose to generate more accurate segment-level anomaly scores based on batch clustering guidance to further improve the performance of WSVAD. Extensive experiments on two public datasets demonstrate the effectiveness of our approach.

Video anomaly detection (VAD) mainly refers to identifying anomalous events that have not occurred in the training set where only normal samples are available. Existing works usually formulate VAD as a reconstruction or prediction problem. However, the adaptability and scalability of these methods are limited. In this paper, we propose a novel distance-based VAD method to take advantage of all the available normal data efficiently and flexibly. In our method, the smaller the distance between a testing sample and normal samples, the higher the probability that the testing sample is normal. Specifically, we propose to use locality-sensitive hashing (LSH) to map the samples whose similarity exceeds a certain threshold into the same bucket in advance. To utilize multiple hashes and further alleviate the computation and memory usage, we propose to use the hash codes rather than the features as the representations of the samples. In this manner, the complexity of near neighbor search is cut down significantly. To make the samples that are semantically similar get closer and those not similar get further apart, we propose a novel learnable version of LSH that embeds LSH into a neural network and optimizes the hash functions with contrastive learning strategy. The proposed method is robust to data imbalance and can handle the large intra-class variations in normal data flexibly. Besides, it has a good ability of scalability. Extensive experiments demonstrate the superiority of our method, which achieves new state-of-the-art results on VAD benchmarks.

For weakly supervised anomaly detection, most existing work is limited to the problem of inadequate video representation due to the inability of modeling long-term contextual information. To solve this, we propose a novel weakly supervised adaptive graph convolutional network (WAGCN) to model the complex contextual relationship among video segments. By which, we fully consider the influence of other video segments on the current one when generating the anomaly probability score for each segment. Firstly, we combine the temporal consistency as well as feature similarity of video segments to construct a global graph, which makes full use of the association information among spatial-temporal features of anomalous events in videos. Secondly, we propose a graph learning layer in order to break the limitation of setting topology manually, which can extract graph adjacency matrix based on data adaptively and effectively. Extensive experiments on two public datasets (i.e., UCF-Crime dataset and ShanghaiTech dataset) demonstrate the effectiveness of our approach which achieves state-of-the-art performance.

Few-shot learning is a fundamental and challenging problem since it requires recognizing novel categories from only a few examples. The objects for recognition have multiple variants and can locate anywhere in images. Directly comparing query images with example images can not handle content misalignment. The representation and metric for comparison are critical but challenging to learn due to the scarcity and wide variation of the samples in few-shot learning. In this paper, we present a novel semantic alignment model to compare relations, which is robust to content misalignment. We propose to add two key ingredients to existing few-shot learning frameworks for better feature and metric learning ability. First, we introduce a semantic alignment loss to align the relation statistics of the features from samples that belong to the same category. And second, local and global mutual information maximization is introduced, allowing for representations that contain locally-consistent and intra-class shared information across structural locations in an image. Furthermore, we introduce a principled approach to weigh multiple loss functions by considering the homoscedastic uncertainty of each stream. We conduct extensive experiments on several few-shot learning datasets. Experimental results show that the proposed method is capable of comparing relations with semantic alignment strategies, and achieves state-of-the-art performance.

Recognizing the actions performed in a video is challenging for an intelligent system since there are wide variations and enormous information in the video. Attention mechanism pays attention to key target areas, ignores irrelevant information and extracts more discriminant features. In recent years, attention mechanism has been introduced into video recognition. Although a rich literature has been spawned, most of the research on attention aims to aggregate local features by attention. Instead of feature aggregation, we propose to aggregate decisions based on local spatio-temporal attention regions for action recognition, which is inspired by ensemble learning. The proposed decision fusion module is easy to interpret and architecture-independent. In this article, the regions around the body joints are regarded as the key regions. We use the corresponding regions of the body joints in the 3-D feature maps as the basic local features for local classification. Finally, all the local classification results are combined to make a global decision. Furthermore, when training the network, we can selectively add supervision to the local and global decisions. We experimentally show that the proposed mechanism can improve the recognition performance on multiple datasets which demonstrates its effectiveness.

For skeleton-based action recognition, most of the existing works used recurrent neural networks. Using convolutional neural networks (CNNs) is another attractive solution considering their advantages in parallelization, effectiveness in feature learning, and model base sufficiency. Besides these, skeleton data are low-dimensional features. It is natural to arrange a sequence of skeleton features chronologically into an image, which retains the original information. Therefore, we solve the sequence learning problem as an image classification task using CNNs. For better learning ability, we build a classification network with stacked residual blocks and having a special design called linear skip gated connection which can benefit information propagation across multiple residual blocks. When arranging the coordinates of body joints in one frame into a skeleton feature, we systematically investigate the performance of part-based, chain-based, and traversal-based orders. Furthermore, a fully convolutional permutation network is designed to learn an optimized order for data rearrangement. Without any bells and whistles, our proposed model achieves state-of-the-art performance on two challenging benchmark datasets, outperforming existing methods significantly.

3-D convolutional neural networks (3-D CNNs) have been established as a powerful tool to simultaneously learn features from both spatial and temporal dimensions, which is suitable to be applied to video-based action recognition. In this paper, we propose not to directly use the activations of fully connected layers of a 3-D CNN as the video feature, but to use selective convolutional layer activations to form a discriminative descriptor for video. It pools the feature on the convolutional layers under the guidance of body joint positions. Two schemes of mapping body joints into convolutional feature maps for pooling are discussed. The body joint positions can be obtained from any off-the-shelf skeleton estimation algorithm. The helpfulness of the body joint guided feature pooling with inaccurate skeleton estimation is systematically evaluated. To make it end-to-end and do not rely on any sophisticated body joint detection algorithm, we further propose a two-stream bilinear model which can learn the guidance from the body joints and capture the spatio-temporal features simultaneously. In this model, the body joint guided feature pooling is conveniently formulated as a bilinear product operation. Experimental results on three real-world datasets demonstrate the effectiveness of body joint guided pooling which achieves promising performance.

Gesture is a natural interface in human-computer interaction, especially interacting with wearable devices, such as VR/AR helmet and glasses. However, in the gesture recognition community, it lacks of suitable datasets for developing egocentric (first-person view) gesture recognition methods, in particular in the deep learning era. In this paper, we introduce a new benchmark dataset named EgoGesture with sufficient size, variation, and reality to be able to train deep neural networks. This dataset contains more than 24 000 gesture samples and 3 000 000 frames for both color and depth modalities from 50 distinct subjects. We design 83 different static and dynamic gestures focused on interaction with wearable devices and collect them from six diverse indoor and outdoor scenes, respectively, with variation in background and illumination. We also consider the scenario when people perform gestures while they are walking. The performances of several representative approaches are systematically evaluated on two tasks: gesture classification in segmented data and gesture spotting and recognition in continuous data. Our empirical study also provides an in-depth analysis on input modality selection and domain adaptation between different scenes.

Gesture is a natural interface in interacting with wearable devices such as VR/AR helmet and glasses. The main challenge of gesture recognition in egocentric vision arises from the global camera motion caused by the spontaneous head movement of the device wearer. In this paper, we address the problem by a novel recurrent 3D convolutional neural network for end-to-end learning. We specially design a spatiotemporal transformer module with recurrent connections between neighboring time slices which can actively transform a 3D feature map into a canonical view in both spatial and temporal dimensions. To validate our method, we introduce a new dataset with sufficient size, variation and reality, which contains 83 gestures designed for interaction with wearable devices, and more than 24,000 RGB-D gesture samples from 50 subjects captured in 6 scenes. On this dataset, we show that the proposed network outperforms competing state-of-the-art algorithms. Moreover, our method can achieve state-of-the-art performance on the challenging GTEA egocentric action dataset.

With the development of sensing equipments, data from different modalities is available for gesture recognition. In this paper, we propose a novel multi-modal learning framework. A coupled hidden Markov model (CHMM) is employed to discover the correlation and complementary information across different modalities. In this framework, we use two configurations: one is multi-modal learning and multi-modal testing, where all the modalities used during learning are still available during testing; the other is multi-modal learning and single-modal testing, where only one modality is available during testing. Experiments on two real-world gesture recognition data sets have demonstrated the effectiveness of our multi-modal learning framework. Improvements on both of the multi-modal and single-modal testing have been observed.

Understanding human activities in video is a fundamental problem in computer vision. In real life, human activities are composed of temporal and spatial arrangement of actions. Understanding such complex activities requires recognizing not only each individual action, but more importantly, capturing their spatio-temporal relationships. This paper addresses the problem of complex activity recognition with a unified hierarchical model. We expand triangular-chain CRFs (TriCRFs) to the spatial dimension. The proposed architecture can be perceived as a spatio-temporal version of the TriCRFs, in which the labels of actions and activity are modeled jointly and their complex dependencies are exploited. Experiments show that our model generates promising results, outperforming competing methods significantly. The framework also can be applied to model other structured sequential data.