Research map / Video / Video segmentation and tracking
Video Event Restoration Based on Keyframes for Video Anomaly Detection
Zhiwei Yang, Jing Liu, Zhaoyang Wu, Peng Wu, Xiaotao Liu
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 dual skip connections (cross-attention + temporal upsampling residual) to USTN-DSC (ablation) | USTN-DSC · Ped2, Avenue, ShanghaiTech | AUC (%): +7.0 Ped2, +6.7 Avenue, +6.4 ShanghaiTech | Improved |
| add cross-attention skip connection only (ablation) | USTN-DSC · Ped2, Avenue, ShanghaiTech | AUC (%): +5.3 Ped2, +4.7 Avenue, +3.8 ShanghaiTech | Improved |
| add temporal upsampling residual skip connection only (ablation) | USTN-DSC · Ped2, Avenue, ShanghaiTech | AUC (%): +4.1 Ped2, +2.2 Avenue, +3.2 ShanghaiTech | Improved |
| add adjacent frame difference loss (AFD loss) (ablation) | USTN-DSC · Ped2, Avenue, ShanghaiTech | AUC (%): +0.9 Ped2, +1.4 Avenue, +2.3 ShanghaiTech | Improved |
| increase ST block depth N from 2 to 4 to 6 (ablation) | USTN-DSC · Ped2, Avenue, ShanghaiTech | AUC (%): N=2→6: +1.0 Ped2, +2.1 Avenue, +2.3 ShanghaiTech | Improved |
| USTN-DSC on ShanghaiTech versus best prediction baseline | USTN-DSC · ShanghaiTech | AUC (%): 73.8 vs ROADMAP 76.6 and DLAN-AC 74.7 | Hurt |
| USTN-DSC as a video-event restoration method for VAD compared with prior SOTA on Ped2 and Avenue | USTN-DSC · Ped2, Avenue | AUC (%): Ped2 98.1 vs 97.7 (best prior); Avenue 89.9 tied with DLAN-AC | Improved |
Closest papers by meaning
- Deep Anomaly Discovery From Unlabeled Videos via Normality Advantage and Self-Paced Refinement (CVPR 2022 · similarity 0.88)
- Dual Conditioned Motion Diffusion for Pose-Based Video Anomaly Detection (AAAI 2025 · similarity 0.86)
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.85)
- UCF-Crime-DVS: A Novel Event-Based Dataset for Video Anomaly Detection with Spiking Neural Networks (AAAI 2025 · similarity 0.85)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.85)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.85)
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.
