Research map / Label-efficient and robust learning / OOD and anomaly detection
Training-Free Industrial Defect Generation with Diffusion Models
Ruyi Xu, Yen-Tzu Chiu, Tai-I Chen, Oscar Chew, Yung-Yu Chuang, Wen-Huang Cheng
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 Sinkhorn-distance feature alignment guidance (ablation) | Anydoor baseline · MVTec AD (few-shot) | classification accuracy (%): 65.41 -> 78.33 | Improved |
| add Sinkhorn feature alignment (ablation) | Anydoor baseline · MVTec AD (few-shot) | Local IS: 3.10 -> 3.27 | Improved |
| add Adaptive Anomaly Mask guidance (ablation) | TF-IDG with Sinkhorn · MVTec AD (few-shot) | pixel AUROC: 98.2 -> 99.0 | Improved |
| add Texture Preservation module (ablation) | TF-IDG with Sinkhorn + AAM · MVTec AD (few-shot) | classification accuracy (%): 79.15 -> 79.84 | Improved |
| use few-shot references instead of one-shot (ablation) | TF-IDG · MVTec AD classification | accuracy (%): 71.92 -> 79.84 | Improved |
| increase seen anomalies from 1 to 5 (ablation) | TF-IDG · MVTec AD | classification accuracy (%): 71.9 -> 78.7 | Improved |
| add generated defects to GLASS in place of hand-crafted synthesis | GLASS · VisA | Pro: 92.2 -> 95.3 | Improved |
| add generated defects to UniAD | UniAD · VisA | image AUROC: 88.8 -> 94.3 | Improved |
| add generated defects to UniAD | UniAD · MVTec AD | image AUROC: 96.5 -> 98.9 | Improved |
| replace Crop-Paste with TF-IDG for training detector | U-Net detector · VisA | image AUROC: 83.5 -> 97.4 | Improved |
| compare with AnoDiff on diversity | TF-IDG · MVTec AD | IC-LPIPS: 0.32 vs 0.32 | No clear effect |
| use gradient guidance at every timestep rather than a limited range (ablation) | TF-IDG · qualitative | artifacts | Hurt |
Closest papers by meaning
- Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · similarity 0.91)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.91)
- SeaS: Few-shot Industrial Anomaly Image Generation with Separation and Sharing Fine-tuning (ICCV 2025 · similarity 0.89)
- DefectFill: Realistic Defect Generation with Inpainting Diffusion Model for Visual Inspection (CVPR 2025 · similarity 0.88)
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.87)
- Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection (CVPR 2025 · similarity 0.86)
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.
