Research map / Video / Video segmentation and tracking
Weakly Supervised Temporal Anomaly Segmentation With Dynamic Time Warping
Dongha Lee, Sehun Yu, Hyunjun Ju, Hwanjo Yu
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 DTW-based segmentation (ablation) | WETAS (DiCNN) · SMD | F1: 0.1265 -> 0.4358 | Improved |
| add alignment loss (with DTW segmentation) (ablation) | WETAS (DiCNN) · SMD | F1: 0.3384 -> 0.4358 | Improved |
| add alignment loss (without DTW segmentation) (ablation) | WETAS (DiCNN) · SMD | F1: 0.1225 -> 0.1265 | No clear effect |
| use DiCNN instead of LSTM (ablation) | WETAS · SMD | F1: 0.3685 -> 0.4358 | Improved |
| use global max pooling instead of average pooling (ablation) | WETAS · SMD | F1: 0.3046 -> 0.4358 | Improved |
| use corruption instead of random sampling for negative pseudo-label (ablation) | WETAS · SMD | F1: 0.4272 -> 0.4358 | Improved |
| use WETAS weak supervision instead of best DeepMIL | DeepMIL-4/8/16 · EMG | F1: 0.4699 -> 0.5803 | Improved |
| use WETAS instead of best DeepMIL | DeepMIL-8 · Subway | F1: 0.6471 -> 0.7414 | Improved |
| use instance-level labels via weak supervision instead of unsupervised point-level detectors | OmniAnomaly++ · GHL | F1: 0.1209 -> 0.2295 | Improved |
| train only on normal instances (semi-supervised ++ variant) | OmniAnomaly · GHL | F1: 0.0611 -> 0.1209 | Improved |
| train only on normal instances (semi-supervised ++ variant) | OmniAnomaly · SMD | F1-best: 0.1223 -> 0.1750 | Improved |
| train only on normal instances (semi-supervised ++ variant) | Donut · SMD | F1-best: 0.3206 -> 0.2875 | Hurt |
Closest papers by meaning
- Weakly-Supervised Temporal Localization via Occurrence Count Learning (ICML 2019 · similarity 0.84)
- Learning Time in Static Classifiers (AAAI 2026 · similarity 0.84)
- VETime: Vision Enhanced Zero-Shot Time Series Anomaly Detection (NeurIPS 2026 · similarity 0.83)
- Learning Precise Temporal Point Event Detection with Misaligned Labels (AAAI 2021 · similarity 0.83)
- Fortifying Time Series: DTW-Certified Robust Anomaly Detection (NeurIPS 2025 · similarity 0.82)
- TAIL-MIL: Time-Aware and Instance-Learnable Multiple Instance Learning for Multivariate Time Series Anomaly Detection (AAAI 2025 · similarity 0.81)
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.
