search

xrli-U / MuSc

This is an official PyTorch implementation for "MuSc : Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images" (MuSc ICLR2024).
MIT License
207 stars 12 forks source link

Results question #9

Open yuhbai opened 2 months ago

yuhbai commented 2 months ago

when I use your default setting in configs/musc.yaml (e.g. vit-l-14-336, batch_size: 4, feature_layers: [5, 11, 17, 23], r_list: [1, 3, 5]), I only get the results: 96.8 96.6 98.8 97.1 62.2 62.3 93.5 (auroc_sp f1_sp ap_sp auroc_px f1_px ap_px aupro)

xrli-U commented 2 months ago

The segmentation results you reported are correct, but the classification results are slightly lower. I don't know what happened. I cloned the program and rerun it just now. The result is exactly the same as in the README.

issue_9

yuhbai commented 2 months ago

sorry, I changed [1, 2, 3] of multi-window setting in RsCIN to [2 ,3]. Now my results are similar to your reported results. However, theoretically the results should be the same as yours? Maybe you know how to explain this subtle difference? image

xrli-U commented 2 months ago

The difference is only 0.0X%, which I suspect may have something to do with your hardware. I just tried the experiment on three servers, but all with the same results.

yuhbai commented 2 months ago

thanks a lot!