Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

non-deterministic result when using cuDNN and upconv_7 #132

Open
nagadomi opened this issue Jul 24, 2016 · 0 comments
Open

non-deterministic result when using cuDNN and upconv_7 #132

nagadomi opened this issue Jul 24, 2016 · 0 comments

Comments

@nagadomi
Copy link
Owner

This issue is related to: lltcggie/waifu2x-caffe#41

Sometimes, upscaled image by waifu2x is unreproducible.
upconv_7 (default upscaling model of latest waifu2x) has Deconvolution layer(aka FullConvolution,Upconvolution). Deconvolution layer on cudnn.torch is implemented by the backward method of Convolution layer (swapping Forward/Backward method). And, backward method of Convolution layer on cuDNN is non-deterministic method. So the forward path of upconv_7 is non-deterministic.
The pixel-wise differences of result images that are caused by this issue is a very little. I think this issue is a small matter. but I was not intended this behavior.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant