Add Flatten, and take batch norm's statistics per channel - #101
Merged
Conversation
On a 4-D activation the old reduction gave every pixel position its own mean and variance, normalizing each position against the batch rather than normalizing the channel.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #101 +/- ##
=======================================
Coverage 97.89% 97.89%
=======================================
Files 59 59
Lines 2850 2856 +6
=======================================
+ Hits 2790 2796 +6
Misses 60 60 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A conv stack had no way to reach a dense head, and batch norm placed after a convolution was computing the wrong thing.
Flattencollapses everything after the batch axis. Batch norm now reduces over every axis but the last, so a channel gets one mean and variance instead of every pixel position getting its own against the batch. It also rejects rank-1 input now, which has no batch axis to take statistics over and used to normalize across features instead.BatchNorm2Dis renamed toBatchNorm, since the reduction rule is identical at every rank. That's what flax and mlx each do with a single class; torch's1d/2d/3dsplit is three rank validators over one shared rule rather than three semantics.