Improve numerical stability of variance computation in MVNLayer #3162
Conversation
This file contains 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
| temp_.mutable_cpu_data()); | ||
| // subtract mean | ||
| caffe_cpu_gemv<Dtype>(CblasNoTrans, num, dim, 1. / dim, bottom_data, | ||
| sum_multiplier_.cpu_data(), 0., mean_.mutable_cpu_data()); // EX |
There was a problem hiding this comment.
Please indent 4 spaces in continuing line, to be consistent with our code style.
|
@kkhoot Thanks for this PR! Please fix the style issue commented above. Everything else looks good to me :) |
|
@ronghanghu Thanks for reviewing this PR. :) The style issue has been fixed. |
|
Let's just leave the rest code as-is, to provide a more readable diff in this PR :) Thanks @kkhoot ! |
ronghanghu
added a commit
that referenced
this pull request
Oct 11, 2015
Improve numerical stability of variance computation in MVNLayer
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
This PR is to avoid numerical precision error in MVNLayer when the input image has large intensity values and relatively small variance, which possibly causes negative values in calculating var(X) = E(X^2) - (EX)^2.
I found this error when dealing with some real data having almost white color intensities resulting in NaN loss or loss divergence.
Solution to this problem is to compute the mean intensity and subtract it first, and compute variance using var(X) = E((X-EX)^2) - E(X-EX)^2, which simply is E((X-EX)^2).
Updated: removed redundant code (subtracting mean) in the forward functions.