Closed sainatarajan closed 3 years ago
Oh yeah sorry about that, that must have been a last minute mistake. Simply replacing it by replace_conv(ch)
should work. The fix should be updated soon. Note that the batchnorm layers aren't being removed as of now, so the replace_conv isn't fully functional.
This commit https://github.com/vballoli/nfnets-pytorch/commit/24d9398d30f0c5f04838667cfc176cab1a5f136f fixes the issue and replaces the BatchNorm2d with identity. Feel free to re-open if you find more issues.
Describe the bug Hello, thank you for the repo. I'm having an issue with the
replace_conv()
.To Reproduce
Screenshots Here is a screenshot of the stacktrace