How to get the official experimental data

caoyunkang / CDO

[TII 2023] Collaborative Discrepancy Optimization for Reliable Image Anomaly Localization

MIT License

61 stars 7 forks source link

How to get the official experimental data #2

Closed dt9640214 closed 1 year ago

dt9640214 commented 1 year ago

I used the parameters you provided to train MVTec2D and the i_roc I got is enough high, but the p_roc and p_pro are about 80 and 60 (according to different categories, some are even only 40 and 20) and p_pro represents the AU-PRO you announced ? Or can AU-PRO get in other ways?

caoyunkang commented 1 year ago

Hi! Could you share your experimental categories and your training parameters? I could help to check it.

dt9640214 commented 1 year ago

抱歉，我发现可能是我在做其他测试的时候不小心改到程式码我用重新下载的程式码跑之后数据正常了不过我想确认程式中的： i_roc对应期刊中的AU-ROC p_roc没有对应 p_pro对应期刊中的AU-PRO 这样理解对吗？

谢谢

caoyunkang commented 1 year ago

I would clarify that p_roc denotes to pixel-level au-roc and p_pro denotes to pixel-level au-pro. Hope this helps!