Skip to content

Conversation

frgfm
Copy link
Contributor

@frgfm frgfm commented Oct 21, 2020

Hey there 👋

A few months back I suggested the PR #2190 that added the eps attribute to FrozenBN. At that time, we set the default to 0 for backward compatibility. As per #2599, the silent difference of eps is an issue. So here is what this PR does:

  • updates __repr__ method to adds eps to align the method with the one of torch.nn.BatchNorm2d
  • updates the unittest on __repr__

Any suggestion is welcome!

@frgfm frgfm changed the title Aligned eps values in FrozenBN and BatchNorm Added eps in the __repr__ of FrozenBN Oct 21, 2020
Copy link
Member

@fmassa fmassa left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks a lot!

@fmassa fmassa merged commit 005355b into pytorch:master Oct 21, 2020
@frgfm frgfm deleted the frozenbn-eps branch October 21, 2020 15:23
bryant1410 pushed a commit to bryant1410/vision-1 that referenced this pull request Nov 22, 2020
* feat: Updated FrozenBN eps to align with BatchNorm

* feat: Added eps to __repr__ of FrozenBN

* test: Updated unittest of __repr__ for FrozenBN

* test: Updated unittest for eps value in BN and FrozenBN

* fix: Revert FrozenBN eps value

* test: Revert test on eps alignment between FrozenBN and BN
vfdev-5 pushed a commit to Quansight/vision that referenced this pull request Dec 4, 2020
* feat: Updated FrozenBN eps to align with BatchNorm

* feat: Added eps to __repr__ of FrozenBN

* test: Updated unittest of __repr__ for FrozenBN

* test: Updated unittest for eps value in BN and FrozenBN

* fix: Revert FrozenBN eps value

* test: Revert test on eps alignment between FrozenBN and BN
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

Successfully merging this pull request may close these issues.

2 participants