Research map / Video / Video segmentation and tracking
Feature Prediction Diffusion Model for Video Anomaly Detection
Cheng Yan, Shiyu Zhang, Yang Liu, Guansong Pang, Wenjun Wang
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 the two diffusion modules over a plain auto-encoder (ablation) | FPDM · UCF-Crime | frame-level AUC: 53.1% -> 74.7% | Improved |
| add refinement module to prediction module (ablation) | FPDM · ShanghaiTech | frame-level AUC: 77.3% -> 78.6% | Improved |
| add prediction module to refinement module (ablation) | FPDM · UCF-Crime | frame-level AUC: 72.5% -> 74.7% | Improved |
| pre-train auto-encoder before optimizing FPDM (ablation) | FPDM · ShanghaiTech | frame-level AUC: 78.6% -> 78.6% | No clear effect |
| decode features back to images before computing anomaly score (ablation) | FPDM · ShanghaiTech | frame-level AUC: 78.6% -> 78.3% | Hurt |
| train prediction and refinement modules separately instead of jointly (ablation) | FPDM · ShanghaiTech | frame-level AUC: 70.8% -> 78.6% | Improved |
| add joint fine-tuning after separate training (ablation) | FPDM · ShanghaiTech | frame-level AUC: 78.6% -> 78.2% | Hurt |
| increase neighbor frames in prediction module from 0 to 12 (ablation) | FPDM · ShanghaiTech scene 3 | AUC: 73.8 -> 81.7 | Improved |
| increase neighbor frames from 0 to 12 (ablation) | FPDM · ShanghaiTech scene 11 (appearance anomalies) | AUC: 96.3 -> 96.6 | No clear effect |
| use FPDM with image features versus object-detection or 3D feature methods | FPDM · ShanghaiTech | frame-level AUC: 78.6% vs 84.3% (SSL) | Hurt |
Closest papers by meaning
- Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model (AAAI 2025 · similarity 0.92)
- Dual Conditioned Motion Diffusion for Pose-Based Video Anomaly Detection (AAAI 2025 · similarity 0.89)
- Open-Vocabulary Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025 · similarity 0.88)
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.
