You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
After ResNet, Batch Normalization becomes quite popular and is widely used recently. However, for visualization, I found parameters bn_gamma, bn_beta, bn_var, bn_mean occupy a lot of space while providing little information. plot_complex.pdf And it also made visualization of some large networks totally unreadable plot_complex_resnet.pdf
During my experiments, BN is always treated as a normal layer (like a convolution). Designers usually does not focus on beta/gamma/mean/var, especially on visualization. I suggest to hide these parameters during visualization.
Hi all
After ResNet, Batch Normalization becomes quite popular and is widely used recently. However, for visualization, I found parameters
bn_gamma
,bn_beta
,bn_var
,bn_mean
occupy a lot of space while providing little information. plot_complex.pdf And it also made visualization of some large networks totally unreadable plot_complex_resnet.pdfDuring my experiments, BN is always treated as a normal layer (like a convolution). Designers usually does not focus on beta/gamma/mean/var, especially on visualization. I suggest to hide these parameters during visualization.
Examples after modification
plot_clean.pdf
plot_clean_resnet.pdf
I have submitted a pull request for this issue.
The text was updated successfully, but these errors were encountered: