Research map / Video / Video segmentation and tracking
PA-VAD: Diffusion-Based Pseudo-Only Video Anomaly Detection via Domain-Aligned Memory Updates
SATOSHI HASHIMOTO, Yanan Wang, Hitoshi Nishimura, Mori Kurokawa
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CLIP-based initial image selection (ablation) | PA-VAD · ShanghaiTech | AUC: 86.7 -> 94.9 | Improved |
| add VLM prompt refinement after initial selection (ablation) | PA-VAD · ShanghaiTech | AUC: 94.9 -> 96.0 | Improved |
| add usage-aware memory update (ablation) | PA-VAD · ShanghaiTech | AUC: 94.9 -> 97.6 | Improved |
| add domain alignment alone (ablation) | PA-VAD · ShanghaiTech | AUC: 94.9 -> 96.0 (initial+DA), 96.0 -> 96.8 with refine | No clear effect |
| combine domain alignment and usage-aware update (full DARM) (ablation) | PA-VAD · ShanghaiTech | AUC: 96.0 -> 98.2 | Improved |
| replace UR-DMU with DARM | PA-VAD · UCF-Crime | AUC: 80.2 -> 82.5 | Improved |
| replace UR-DMU with DARM | PA-VAD · XD-Violence | AUC: 92.5 -> 95.1 | Improved |
| apply DARM to pseudo anomalies (ablation) | PA-VAD · SHT feature norm | mean L2 norm: 199.48 -> 8.39 (real 9.13) | Improved |
| add VLM prompt refinement to generation (ablation) | CA-PAG · SHT pseudo-clip quality | FVD: 701 -> 604 | Improved |
| increase pseudo clips from 140 to 175 (ablation) | PA-VAD · ShanghaiTech | AUC: 98.2 -> 97.7 | Hurt |
| increase pseudo clips from 14 to 140 (ablation) | PA-VAD · ShanghaiTech | AUC: 85.4 -> 98.2 | Improved |
| train with pseudo-only anomalies vs real-abnormal baseline on UCF-Crime | PA-VAD · UCF-Crime | AUC: 82.5 vs 87.0 (UR-DMU) | Hurt |
Closest papers by meaning
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.88)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · similarity 0.88)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.87)
- MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection (NeurIPS 2025 · similarity 0.87)
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
