Research map / Detection and segmentation / Open-set and few-shot detection
Defect Spectrum: A Granular Look of Large-scale Defect Datasets with Rich Semantics
Shuai Yang, ZhiFei Chen, Pengguang Chen, Xi Fang, Yixun Liang, Shu Liu, Yingcong Chen
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 Defect-Gen synthetic training data | MiT-B0 · DS-MVTec | mIoU: 46.45 -> 56.21 | Improved |
| add Defect-Gen synthetic training data | DeepLabV3+ · DS-Cotton | mIoU: 48.73 -> 58.58 | Improved |
| add Defect-Gen synthetic training data | DeepLabV3+ · DS-VISION | mIoU: 52.33 -> 53.46 | Improved |
| add Defect-Gen synthetic data to Mask2Former | Mask2Former · DS-Cotton | mIoU: 64.09 -> 65.39 | Improved |
| use synthetic data equal to 100% of the training set rather than more (ablation) | DeepLabV3+ / MiT-B0 · Defect Spectrum (MVTec) | mIoU | Mixed |
| train on Defect Spectrum refined annotations instead of original annotations | segmentation model · MVTec simulated production | false positive rate (%): 49.60 -> 16.50 | Improved |
| train on Defect Spectrum refined annotations instead of original annotations | segmentation model · MVTec simulated production | recall (%): 85.33 -> 96.07 | Improved |
| augment with sinDiffusion-generated images | segmentation model · Original MVTec (mean) | mIoU: 64.01 -> 63.51 | Hurt |
| augment with DDPM-generated images | segmentation model · Original MVTec (mean) | mIoU: 64.01 -> 65.07 | Improved |
| augment with Defect-Gen images | segmentation model · Original MVTec (mean) | mIoU: 64.01 -> 67.76 | Improved |
| increase the DDPM-style switch timestep u from 25 to 700 (ablation) | Defect-Gen (small receptive field) · capsule class (generation) | FID: 115.2754 -> 78.4115 | Improved |
| use a larger receptive field in the small model (ablation) | Defect-Gen · capsule class (generation) | LPIPS: 0.3666 -> 0.2870 at u=50 (small to large); FID 93.2839 -> 56.6246 | Mixed |
Closest papers by meaning
- Automated Defect Report Generation for Enhanced Industrial Quality Control (AAAI 2024 · similarity 0.90)
- Towards Open-Vocabulary Industrial Defect Understanding with a Large-Scale Multimodal Dataset (CVPR 2026 · similarity 0.89)
- DefectFill: Realistic Defect Generation with Inpainting Diffusion Model for Visual Inspection (CVPR 2025 · similarity 0.88)
- MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context (AAAI 2025 · similarity 0.87)
- Wavelet and Prototype Augmented Query-based Transformer for Pixel-level Surface Defect Detection (CVPR 2025 · similarity 0.86)
- 3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly Detection (AAAI 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.
