| title: PatchCore Defect Detection | |
| emoji: 🔍 | |
| colorFrom: gray | |
| colorTo: red | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| license: cc-by-nc-sa-4.0 | |
| short_description: Find manufacturing defects without ever seeing one | |
| models: | |
| - yandyassine/patchcore-mvtec-ad | |
| # PatchCore defect detection, in your browser | |
| From-scratch reimplementation of PatchCore (CVPR 2022), exported to ONNX and run client-side with ONNX Runtime Web | |
| (WebGPU when available, WebAssembly otherwise). No server: uploaded images never leave the visitor's device. | |
| Image AUROC on MVTec AD: **1.000 / 0.999 / 0.977** (bottle / metal nut / grid), measured with the same ONNX files. | |
| - Model files and card: [yandyassine/patchcore-mvtec-ad](https://huggingface.co/yandyassine/patchcore-mvtec-ad) | |
| - Code & write-up: [COMPUTER-VISION-PROJECT-LIBRARY / 13](https://github.com/yandouziyassine/COMPUTER-VISION-PROJECT-LIBRARY/tree/main/13-anomaly-detection-patchcore) | |
| - Data: MVTec AD, CC BY-NC-SA 4.0 (non-commercial). | |