facebookresearch / facebookresearch/SlowFast
Why Do You Add -1e10 to Video Predictions for Multi-label case in Meters.py?
Open
- Dominant language
- Python
- Stars
- 7.4k
- Forks
- 1.3k
- PR merge metrics
- No merged PRs in 30d
Description
Hi,
Just wondering, why do you add -1e10 as the initial video prediction during testing: https://github.com/facebookresearch/SlowFast/blob/main/slowfast/utils/meters.py#L273
Seems like a mistake to me if I have already applied sigmoid to the final prediction outputs.
Thanks,
Junwei
Contributor guide
Assessment
This issue has not been assessed yet.