takuoko

@hhaAndroid Yes.

> May I ask why the performance drops and can we fix them? I do not know the cause of the drop in performance at this time. I was wondering...

I found two differences. 1. drop out rate in MultiScaleDeformableAttention. official = 0.0 mine = 0.1(default of mmcv) 2. use layer_scale or no. offical = use layer_scale mine = not...

In my experiments, layer scale made mAP worse. I recorded box AP= 45.6 and mask AP= 40.9 with the latest config, which is close to the official performance. After refactor...

@ZwwWayne I understand. However, the mmcls PRs have not been merged yet, so we may have to wait for that.

Why do so many cpu tests fail, including some that have not been changed? https://github.com/open-mmlab/mmagic/actions/runs/6258709202/job/16993265761?pr=2040#step:10:4026

I am fixing the test error on mmengine PR. https://github.com/open-mmlab/mmengine/pull/1365

@hhaAndroid This PR is ready!

@hhaAndroid This PR is ready.