YuliangXiu/ICON

Checkpoints resume

Closed this issue · 3 comments

Why just load the MLP weights when resume the checkpoints for retraining?

image

This MLP is not named perfectly, it contains both MLP and image filter if any.

Normal network weights come from normal.ckpt

image
我发现代码在构建网络时,MLP和filter是分开构建的,所以MLP并不是filter请问是这样麻?如图所示

This MLP is not named perfectly, it contains both MLP and image filter if any.

Normal network weights come from normal.ckpt

I found that when the code builds the network, the MLP and the filter are built separately, so the MLP is not the filter please is this the case? As shown in the picture
https://user-images.githubusercontent.com/57577944/237590703-70c7959b-307c-47b6-a728-7d64bdbfa9ee.png