The key to video anomaly detection is understanding the appearance and motion differences between normal abnormal events. However, previous works either considered characteristics of or in isolation treated them without distinction, making model fail exploit unique both. In this brief, we propose an appearance-motion united auto-encoder (AMAE) framework jointly learn prototypical spatial temporal patterns AMAE includes a normality, channel attention-based spatial-temporal decoder fuse...

Video anomaly detection (VAD) under weak supervision aims to temporally locate abnormal clips using the easy-to-obtain video-level labels. In this brief, we introduce underlying thought of unsupervised VAD weakly supervised and propose a collaborative normality learning framework obtain more discriminative deep representations. Specifically, auto-encoder is first trained in an manner learn prototypical spatial-temporal patterns normal videos. Then, both videos are used train regression...

As essential tools for industry safety protection, automatic video anomaly detection systems (AVADS) are designed to detect anomalous events of concern in surveillance videos. Existing VAD methods lack effective exploration the prototypical appearance and motion features leading poor performance realistic scenarios. Specifically, they either misreport regular as anomalies due insufficient representation power, or lead missed detections with over-power generalization. In this regard, we...

Video anomaly detection is a challenging task in the Computer vision community. Most single task-based methods do not consider independence of unique spatial and temporal patterns, while two-stream structures lack exploration correlations. In this paper, we propose spatial-temporal memories augmented auto-encoder framework, which learns appearance normality motion normal-ity independently explores correlations via adversar-ial learning. Specifically, first design two proxy tasks to train...

The automatic detection of abnormal events in surveillance videos with weak supervision has been formulated as a multiple instance learning task, which aims to localize the clips containing temporally video-level labels. However, most existing methods rely on features extracted by pre-trained action recognition models, are not discriminative enough for video anomaly detection. In this work, we propose spatial-temporal attention mechanism learn inter- and intra-correlations clips, boosted...

Video anomaly detection is an essential yet challenging task in the multimedia community, with promising applications smart cities and secure communities. Existing methods attempt to learn abstract representations of regular events statistical dependence model endogenous normality, which discriminates anomalies by measuring deviations learned distribution. However, conventional representation learning only a crude description video normality lacks exploration its underlying causality. The...

The complementarity of multimodal signal is essential for video anomaly detection. However, existing methods either lack exploration to data or ignore the implicit alignment features. In our work, we address this problem using a novel fusion method and propose Multimodal Supervise-Attention enhanced Fusion (MSAF) framework under weak supervision. Our can be divided into two parts: 1) labels refinement part refines video-level ground truth pseudo clip-level subsequent training, 2)...

Video Anomaly Detection (VAD) is an essential yet challenging task in the signal processing community, which aims to understand spatial and temporal contextual interactions between objects surrounding scenes detect unexpected events surveillance videos. However, existing unsupervised methods either use a single network learn global prototype patterns without making unique distinction foreground background or try strip from frames, ignoring that essence of anomalies lies unusual object-scene...

Diffusion models (DMs) have emerged as a powerful class of generative AI models, showing remarkable potential in anomaly detection (AD) tasks across various domains, such cybersecurity, fraud detection, healthcare, and manufacturing. The intersection these two fields, termed diffusion for (DMAD), offers promising solutions identifying deviations increasingly complex high-dimensional data. In this survey, we systematically review recent advances DMAD research investigate their capabilities....

In recent years, generating 3D human models from images has gained significant attention in reconstruction. However, deploying large neural network practical applications remains challenging, particularly on resource-constrained edge devices. This problem is primarily because require significantly higher computational power, which imposes greater demands hardware capabilities and inference time. To address this issue, we can optimize the architecture to reduce number of model parameters,...

Video Anomaly Detection (VAD) remains a fundamental yet formidable task in the video understanding community, with promising applications areas such as information forensics and public safety protection. Due to rarity diversity of anomalies, existing methods only use easily collected regular events model inherent normality normal spatial-temporal patterns an unsupervised manner. Although have made significant progress benefiting from development deep learning, they attempt statistical...

Violence detection is an essential and challenging problem in the computer vision community. Most existing works focus on single modal data analysis, which not effective when multi-modality available. Therefore, we propose a two-stage multi-modal information fusion method for violence detection: 1) first stage adopts multiple instance learning strategies to refine video-level hard labels into clip-level soft labels, 2) next uses fused attention module achieve fusion, supervised carried out...

Unsupervised Domain Adaptation (UDA) aims to free models from labeled information of target domain by minimizing the discrepancy distributions between different domains. Most existing methods are designed learn domain-invariant features either discrimination or matching lower-order moments. However, these not robust due limited representation statistical characteristics for non-Gaussian and thus fail in matching. In addition, they often focus on while considering class decision boundaries To...

Automatic facial expression recognition (FER) based on face images is essential for affective robots, which are designed interactive companions and intelligent healthcare. Although existing DL-based FERs have made significant progress, an accurate FER model in robots challenging due to the subtle differences expressions across various scenarios. To address this issue, we propose a multigranularity region relation representation network (MGR <sup xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Existing unsupervised video anomaly detection methods often suffer from performance degradation due to the overgeneralization of deep models. In this paper, we propose a simple yet effective Multi-Scale Normality network (MSN-net) that uses hierarchical memories learn multi-level prototypical spatial-temporal patterns normal events. Specifically, memory module interacts with encoder through reading and writing operations during training phase, preserving multi-scale normality in three...

Video anomaly detection (VAD) remains a challenging task in the pattern recognition community due to ambiguity and diversity of abnormal events. Existing deep learning-based VAD methods usually leverage proxy tasks learn normal patterns discriminate instances that deviate from such as abnormal. However, most them do not take full advantage spatial-temporal correlations among video frames, which is critical for understanding patterns. In this paper, we address unsupervised by learning...

Video Anomaly Detection (VAD) serves as a pivotal technology in the intelligent surveillance systems, enabling temporal or spatial identification of anomalous events within videos. While existing reviews predominantly concentrate on conventional unsupervised methods, they often overlook emergence weakly-supervised and fully-unsupervised approaches. To address this gap, survey extends scope VAD beyond encompassing broader spectrum termed Generalized Event (GVAED). By skillfully incorporating...