facebookresearch / facebookresearch/SlowFast
How to handle class imbalance?
Open
- Dominant language
- Python
- Stars
- 7.4k
- Forks
- 1.3k
- PR merge metrics
- No merged PRs in 30d
Description
Hello Authors,
I am training i3d for a custom dataset whose task is very similar to AVA i.e., predict the action performed by each object given video frames and the objects' bounding boxes in every frame as input.
However there is imbalance in the dataset with one 1 of the 15 classes is present for 20 times more than the other classes.
Is there any parameter or any way to weigh loss corresponding to different classes diffierently?
ThankYou
Contributor guide
Assessment
This issue has not been assessed yet.