facebookresearch / facebookresearch/SlowFast
Question regarding SubBatchNorm3d
Open
- Dominant language
- Python
- Stars
- 7.4k
- Forks
- 1.3k
- PR merge metrics
- No merged PRs in 30d
Description
Hello, thanks for sharing the code!
I have a question regarding the line:
https://github.com/facebookresearch/SlowFast/blob/9bcc68fed2a746973f0e155cc41ebb9c43561c6e/slowfast/models/batchnorm_helper.py#L78-L81
1. This is minor but "std" technically means variance right?
2. I was curious why you add an additional term after "stds.view(n, -1).sum(0) / n" when aggregating for std. Why it is not simply the average of sub batch statistics like you do for mean?
Thank you,
Jaejin Cho
Contributor guide
Assessment
This issue has not been assessed yet.