Hi @zhiCHEN96 , this is a fantastic work! I already ran the evaluation part and the result looked good.
However, I have a question for the evaluation part.
In my experiment, I set only one CW_layer for one experiment. For example, set layer 2 as CW layer and next time set layer 16 as CW layer.
However, I saw the code that I can convert multi-layer as CW layers. When I did that, the top1, top5 Acc will be really low. Could you explain why this happened? Is there a good way to train multi-CW-layer model?
Thank you so much!
Hi @zhiCHEN96 , this is a fantastic work! I already ran the evaluation part and the result looked good. However, I have a question for the evaluation part. In my experiment, I set only one CW_layer for one experiment. For example, set layer 2 as CW layer and next time set layer 16 as CW layer. However, I saw the code that I can convert multi-layer as CW layers. When I did that, the top1, top5 Acc will be really low. Could you explain why this happened? Is there a good way to train multi-CW-layer model? Thank you so much!
Best, Harold.