Research map / Label-efficient and robust learning / OOD and anomaly detection
Few-Shot Anomaly-Driven Generation for Anomaly Classification and Segmentation
Guan Gui, Bin-Bin Gao, Jun Liu, Chengjie Wang, Yunsheng Wu
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 generated anomalies with weak-supervision to DRAEM training | DRAEM · MVTec pixel-level | AU-PR: 67.4 -> 73.2 | Improved |
| add generated anomalies to DeSTSeg training | DeSTSeg · MVTec pixel-level | AU-PR: 76.6 -> 78.1 | Improved |
| add generated anomalies to DRAEM | DRAEM · MVTec image-level | AU-ROC: 97.1 -> 98.7 | Improved |
| use mask-guided loss when learning the anomaly embedding (ablation) | AnoGen embedding · MVTec pixel-level | AU-PR: 70.4 -> 73.2 | Improved |
| use 3 support anomalies instead of 1 (ablation) | AnoGen · MVTec pixel-level | AU-PR: 70.5 -> 73.2 | Improved |
| use 5 support anomalies instead of 3 (ablation) | AnoGen · MVTec pixel-level | AU-PR: 73.2 -> 73.0 | No clear effect |
| set confidence threshold tau to 0.9 versus 1.0 or 0.8 (ablation) | weakly-supervised loss · MVTec pixel-level | AU-PR: 68.9 (tau 1.0) and 65.4 (tau 0.8) -> 73.2 (tau 0.9) | Improved |
| combine DRAEM synthetic and generated anomalies (ablation) | DRAEM · MVTec pixel-level | AU-PR: 67.4 / 68.3 -> 73.2 | Improved |
| use generated anomalies only versus DRAEM synthetic only (ablation) | DRAEM · MVTec pixel-level | AU-PR: 67.4 -> 68.3 | No clear effect |
| use a different support set of the same anomaly type (ablation) | AnoGen · MVTec pixel-level | AU-PR: 73.2 -> 71.4 | No clear effect |
Closest papers by meaning
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.93)
- Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · similarity 0.91)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.90)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.90)
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.89)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.89)
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.
