Mutual-Channel-Loss icon indicating copy to clipboard operation
Mutual-Channel-Loss copied to clipboard

Published 20 hours ago • verified publisher icon Kurumi233

Linear and Parameter Settings

Open zhangchuanyi96 opened this issue 4 years ago • 1 comments

Hi, thank you for your codes. I have managed to reproduce the results of vgg16 and resnet50 with dense layers. However, when I tried to use linear layer, the result is bad. Can you please provide the parameter settings of using linear metric? Thanks.

zhangchuanyi96 avatar Sep 10 '20 12:09 zhangchuanyi96

Hi, thank you for your codes. I have managed to reproduce the results of vgg16 and resnet50 with dense layers. However, when I tried to use linear layer, the result is bad. Can you please provide the parameter settings of using linear metric? Thanks.

The set of dense layers is following the oldest official code, and linear layer is just like a general classification task. I had got a similar problem when using a linear layer, but I didn't go into it. Maybe the max pool operation with one linear layer causes some overfitting. A possible answer in https://github.com/PRIS-CV/Mutual-Channel-Loss/issues/7.

Kurumi233 avatar Sep 10 '20 12:09 Kurumi233