Research map / Video / Video segmentation and tracking
STEP: Score-Based Temporal Energy for Human Pose Video Anomaly Detection
Jakub Micorek, Mateusz Kozinski, Horst Possegger
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 |
|---|---|---|---|
| project pose sequences into whitened PCA space (K=48) instead of raw coordinates at long window (ablation) | STEP (no confidence) · UBnormal, T=32 | AUROC: 70.6 -> 81.3 | Improved |
| add sequence-level confidence weighting to raw-coordinate DSM (ablation) | MULDE raw coords · UBnormal, T=2 | AUROC: 80.7 -> 87.2 | Improved |
| add confidence weighting on top of PCA K=48 (ablation) | STEP · UBnormal, T=12 | AUROC: 84.0 -> 90.4 | Improved |
| increase temporal window with raw coordinates (ablation) | MULDE raw coords · UBnormal | AUROC: 80.7 -> 70.6 (T=2 to T=32) | Hurt |
| add sigma-modulated Residual MLP over vanilla MLP (ablation) | STEP · UBnormal | AUROC: 89.3 -> 90.4 | Improved |
| add whitening of PC-space (ablation) | STEP · UBnormal | AUROC: 88.2 -> 90.4 | Improved |
| hard-drop poses with confidence <0.4 instead of soft weighting (ablation) | STEP · UBnormal | AUROC: 90.4 -> 85.9 | Hurt |
| hard-drop poses with confidence <0.1 (ablation) | STEP · UBnormal | AUROC: 90.4 -> 90.5 | No clear effect |
| raise sigma_low to 0.5 (ablation) | STEP · ShanghaiTech | AUROC: 86.3 -> 82.7 | Hurt |
| lower sigma_low to 1e-4 (ablation) | STEP · UBnormal | AUROC: 90.4 -> 89.6 | Hurt |
| lower sigma_high to 0.2 (ablation) | STEP · UBnormal vs ShanghaiTech | AUROC: UBnormal 90.4 -> 89.9; ShanghaiTech 86.3 -> 86.4 | Mixed |
| lower sigma_high to 0.5 (ablation) | STEP · UBnormal | AUROC: 90.4 -> 90.6 | No clear effect |
Closest papers by meaning
- Sequential keypoint density estimator: an overlooked baseline of skeleton-based video anomaly detection (ICCV 2025 · similarity 0.90)
- Dual Conditioned Motion Diffusion for Pose-Based Video Anomaly Detection (AAAI 2025 · similarity 0.89)
- PAD: A Dataset and Benchmark for Pose-agnostic Anomaly Detection (NeurIPS 2023 · similarity 0.88)
- Multimodal Motion Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection (ICCV 2023 · similarity 0.88)
- Learning Regularity in Skeleton Trajectories for Anomaly Detection in Videos (CVPR 2019 · similarity 0.87)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 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.
