As the scale of video surveillance data outpaces manual annotation capacities, weakly supervised video anomaly detection (WSVAD) has emerged as a critical research frontier. Most existing approaches formulate WSVAD within a Multiple Instance Learning (MIL) framework that relies on rigid, hand-crafted temporal priors to supervise anomaly scoring. However, such formulations exhibit limited adaptability to the wide variation in anomaly durations and temporal dynamics observed in real-world videos, often leading to unstable or unreliable snippet-level predictions. To address this limitation, we propose an adaptive temporal modeling framework for WSVAD that explicitly accounts for variations in video dynamics across multiple temporal granularities. First, we introduce a Temporal Refinement Module (TRM) that leverages dynamic positional encoding and a learnable class token to model long-range temporal dependencies while distilling a stable global video-level representation. Second, to capture anomalous events with varying frequency and duration, we develop an adaptive Event Segmentation Module (ESM) that identifies event boundaries through temporal discontinuity analysis and aggregates snippet features into discriminative event-level representations. Finally, for snippet-level and event-level predictions, we propose an adaptive similarity-based fusion strategy that dynamically integrates anomaly scores into video-level predictions, replacing fixed top-k aggregation heuristics with global semantic relevance. Extensive experiments on two benchmarks demonstrate that the proposed framework consistently outperforms state-of-the-art methods.
Figures & tables
Figure 1: Overview of the proposed framework. Snippet-level features are first extracted from input videos using backbone networks. These features are then refined by the TRM to capture long-range temporal dependencies and produce a global video-level representation enriched with contextual information. Subsequently, the ESM is adaptively applied to identify event boundaries and aggregate snippet features into event-level representations. Finally, the model is jointly optimized through a multi-scale prediction strategy that integrates snippet-, event-, and video-level outputs.
Supervision
Methods
Feature Encoder
Ground Truth
AUC(%)
FAR(%)
Semi
Object-Centric [ 7 ]
-
N
61.60
-
Conv-AE [ 5 ]
AE
N
50.60
27.2
Weak
HL-Net [ 19 ]
I3D
B
82.44
-
RTFM [ 16 ]
I3D
B
84.30
-
DDL [ 12 ]
I3D
B
85.12
-
MGFN [ 2 ]
I3D
B
86.98
-
Table 1: Performance comparison of SOTA methods on the UCF-Crime dataset.
Supervision
Methods
Feature Encoder
Ground Truth
AP(%)
FAR(%)
Semi
OCSVM [ 13 ]
I+V
N
27.25
-
Conv-AE [ 5 ]
I+V
N
30.77
-
Weak
HL-Net [ 19 ]
I3D
B
75.41
-
RTFM [ 16 ]
I3D
B
77.81
-
DDL [ 12 ]
I3D
B
80.72
-
MGFN [ 2 ]
I3D
B
79.19
-
Table 2: Performance comparison of SOTA methods on the XD-Violence dataset.
No
I3D
TRM
ESM
class token
AUC(%)
FAR(%)
1
✓
-
-
-
82.50
1.84
2
✓
✓
-
-
84.78
0.85
3
✓
✓
-
✓
85.52
0.53
4
✓
✓
✓
✓
87.24
0.34
Table 3: Ablation study of different modules in our method on the UCF-Crime dataset.
Figure 2: Visualized anomaly detection results of our proposed method. Examples in the first row are from the UCF-Crime dataset, while those in the second row are from the XD-Violence dataset. In each plot, the blue curve indicates the anomaly score at each time step, with green backgrounds representing normal segments and red backgrounds denoting anomalous segments.
Weakly supervised video anomaly detection (WSVAD) has predominantly focused on temporal localization, identifying when anomalies occur while largely neglecting their spatial extent within frames. Yet, spatial localization is essential for interpretability and practical deployment in real-world settings. We introduce a patch-based spatiotemporal framework for weakly supervised anomaly localization that jointly models where and when anomalies occur. Our approach operates on grid-level patch features and learns region-level anomaly scores under a multiple instance learning paradigm. We further propose a Proximity-Aware Top-k spatiotemporal selection strategy that enables the model to generate fine-grained spatial anomaly maps without requiring bounding-box supervision during training. Our method surpasses existing state-of-the-art approaches across multiple benchmarks, yielding substantial gains in spatiotemporal localization accuracy. In addition, we release frame-level bounding-box annotations for the test sets of two widely used datasets, along with our code and pretrained models, providing new resources to facilitate future research in spatially grounded WSVAD.
Weakly supervised video anomaly detection (WS-VAD) involves identifying the temporal intervals that contain anomalous events in untrimmed videos, where only video-level annotations are provided as supervisory signals. However, a key limitation persists in WS-VAD, as dense frame-level annotations are absent, which often leaves existing methods struggling to learn anomaly semantics effectively. To address this issue, we propose a novel framework named LAS-VAD, short for Learning Anomaly Semantics for WS-VAD, which integrates anomaly-connected component mechanism and intention awareness mechanism. The former is designed to assign video frames into distinct semantic groups within a video, and frame segments within the same group are deemed to share identical semantic information. The latter leverages an intention-aware strategy to distinguish between similar normal and abnormal behaviors (e.g., taking items and stealing). To further model the semantic information of anomalies, as anomaly occurrence is accompanied by distinct characteristic attributes (i.e., explosions are characterized by flames and thick smoke), we additionally incorporate anomaly attribute information to guide accurate detection. Extensive experiments on two benchmark datasets, XD-Violence and UCF-Crime, demonstrate that our LAS-VAD outperforms current state-of-the-art methods with remarkable gains.
Yu Wang, Shengjie Zhao
School of Computer Science and Technology, Tongji University, Shanghai, China
Weakly supervised video anomaly detection relies solely on video-level labels for training, making it difficult to accurately localize anomalous events in complex scenes. In real-world videos, anomalous behaviors exhibit large variations in appearance and temporal duration, while scene appearance and action dynamics are often tightly entangled. Consequently, existing models tend to rely on scene-related statistical cues rather than true behavioral deviations, resulting in unstable detection performance. To address this challenge, we propose a Structured Evidence Selection framework (SESAD) that reformulates anomaly detection as a structured reasoning process over clip-level visual evidence. Instead of directly mapping aggregated features to anomaly scores, SESAD reorganizes clip representations into semantically structured candidate evidence and performs context-conditioned selection under scene and action constraints. This mechanism adaptively emphasizes anomaly-relevant semantics while suppressing scene interference, thereby alleviating semantic entanglement under weak supervision. Furthermore, we introduce a lightweight geometric discrimination module that constructs a dual-prototype structure in the embedding space, enabling anomaly decisions through relative geometric relations. Extensive experiments on UBnormal, ShanghaiTech, and UCF-Crime show that SESAD achieves 67.92, 97.99, and 88.46 AUC, respectively, while maintaining high computational efficiency and overall consistently stable anomaly discrimination.
Chenglizhao Chen, Tianxiang Nan, Wen Li +4
China University of Petroleum (East China) Qingdao, China