Unsupervised video anomaly detection (UVAD) has gained significant attention due to its label-free nature. Typically, UVAD methods can be categorized into two branches, i.e. the one-class classification (OCC) methods and fully UVAD ones. However, the former may suffer from data imbalance and high false alarm rates, while the latter relies heavily on feature representation and pseudo-labels. In this paper, a novel feature reconstruction and disruption model (FRD-UVAD) is proposed for effective feature refinement and better pseudo-label generation in fully UVAD, based on cascade cross-attention transformers, a latent anomaly memory bank and an auxiliary scorer. The clip features are reconstructed using the space-time intra-clip information, as well as cross-inter-clip knowledge. Moreover, instead of blindly reconstructing all training features as OCC methods, a new disruption process is proposed to cooperate with the feature reconstruction simultaneously. Using the collected pseudo anomaly samples, it is able to emphasize the feature differences between normal and abnormal events. Additionally, a pre-trained UVAD scorer is utilized as a different criteria for anomaly prediction, which further refines the pseudo-labels. To demonstrate its effectiveness, comprehensive experiments and detailed ablation studies are conducted on three video benchmarks, namely CUHK Avenue, ShanghaiTech and UCF-Crime. Our proposed model (FRD-UVAD) achieves the best AUC performance (91.23%, 80.14%, and 82.12%) on all three datasets, surpassing other state-of-the-art OCC and fully UVAD methods. Furthermore, it obtains the lowest false alarm rate with a lower scene dependency, compared with other OCC methods. The code is available at https://github.com/tcc-power/FRD-unsupervised-video-anomaly-detection .

Recent semi-supervised learning (SSL) algorithms such as FixMatch achieve state-of-the-art performance by exploiting consistency regularization and entropy minimization techniques. However, many consistency-based SSL algorithms extract pseudo-labels from unlabeled data through a fixed threshold and ignore the different learning progress of each category, which makes the easy-to-learn categories have more examples contributing to the loss, resulting in a class-imbalance problem and affecting training efficiency. In order to improve the training reliability, we propose adaptive weighted losses (AWL). Through the evaluation of the class-wise learning progress, the loss contribution of the pseudo-labeled data of each category is continuously and dynamically adjusted during the learning process, and the pseudo-label discrimination ability of the model can be steadily improved. Moreover, to improve the training efficiency, we propose a bidirectional distribution approximation (DA) method, which introduces the consistency information of the predictions under the threshold into the loss calculation, and significantly improves the model convergence speed. Through the combination of AWL and DA, our method surpasses the performance of other algorithms on multiple benchmarks with a faster convergence efficiency, especially in the case of labeled data extremely limited. For example, AWL&DA achieves 95.29% test accuracy on the CIFAR-10-40-labels experiment and 92.56% accuracy on a faster experiment setting with only 2¹⁸ iterations.

In the edge intelligence environment, multiple sensing devices perceive and recognize the current scene in real time to provide specific user services. However, the generalizability of the fixed recognition model will gradually weaken due to the time-varying perception scene. To ensure the stability of the perception and recognition service, each edge model/agent needs to continuously learn from the new perception data unassisted to adapt to the perception environment changes and jointly build the online evolutive learning (OEL) system. The generalization degradation problem can be addressed by deploying the semi-supervised learning (SSL) method on multi-view agents and continuously tuning each discriminative model by collaborative perception. This paper proposes a multi-view agent’s collaborative perception (MACP) semi-supervised online evolutive learning method. First, each view model will be initialized based on self-supervised learning methods, and each initialized model can learn differentiated feature-extraction patterns with certain discriminative independence. Then, through the discriminative information fusion of multi-view model predictions on the unlabeled perceptual data, reliable pseudo-labels are obtained for the consistency regularization process of SSL. Moreover, we introduce additional critical parameter constraints to continuously improve the discriminative independence of each view model during training. We compare our method with multiple representative multi-model and single-model SSL methods on various benchmarks. Experimental results show the superiority of the MACP in terms of convergence efficiency and performance. Meanwhile, we construct an ideal multi-view experiment to demonstrate the application potential of MACP in practical perception scenarios.

Video anomaly detection (VAD) has been intensively studied for years because of its potential applications in intelligent video systems. Existing unsupervised VAD methods tend to learn normality from training sets consisting of only normal videos and regard instances deviating from such normality as anomalies. However, they often consider only local or global normality in the temporal dimension. Some of them focus on learning local spatiotemporal representations from consecutive frames to enhance the representation for normal events. But powerful representation allows these methods to represent some anomalies and causes miss detection. In contrast, the other methods are devoted to memorizing prototypical normal patterns of whole training videos to weaken the generalization for anomalies, which also restricts them from representing diverse normal patterns and causes false alarm. To this end, we propose a two-branch model, Local-Global Normality Network (LGN-Net), to simultaneously learn local and global normality. Specifically, one branch learns the evolution regularities of appearance and motion from consecutive frames as local normality utilizing a spatiotemporal prediction network, while the other branch memorizes prototype features of the whole videos as global normality by a memory module. LGN-Net achieves a balance of representing normal and abnormal instances by fusing local and global normality. In addition, the fused normality enables LGN-Net to generalize to various scenes more than exploiting single normality. Experiments demonstrate the effectiveness and superior performance of our method. The code is available online: https://github.com/Myzhao1999/LGN-Net.