facebookresearch / facebookresearch/SlowFast
Duplicate train dataset filelist AVA
Open
- Dominant language
- Python
- Stars
- 7.4k
- Forks
- 1.3k
- PR merge metrics
- No merged PRs in 30d
Description
https://github.com/facebookresearch/SlowFast/blob/fd41618191d3c21c1ca21a61369ce9917646cf9c/configs/AVA/SLOW_8x8_R50_SHORT.yaml#L22-L25
In the configs above, `ava_train_v2.2.csv` is added into the `TRAIN_PREDICT_BOX_LISTS`. However, `ava_train_v2.2.csv` is already added into the `TRAIN_GT_BOX_LISTS`:
https://github.com/facebookresearch/SlowFast/blob/fd41618191d3c21c1ca21a61369ce9917646cf9c/slowfast/datasets/ava_helper.py#L82-L87
Though duplicate items can be processed in the following steps, I think it is unnecessary and unreasonable to treat `ava_train_v2.2.csv` as **predicted**, which is **gt** in fact.
Kindly ping @takatosp1
Contributor guide
Assessment
This issue has not been assessed yet.