facebookresearch / facebookresearch/SlowFast
Effect of data augumentation REVERSE_INPUT_CHANNEL and INV_UNIFORM_SAMPLE
Open
question
- Dominant language
- Python
- Stars
- 7.4k
- Forks
- 1.3k
- PR merge metrics
- No merged PRs in 30d
Description
I notice the config in Charades set DATA.REVERSE_INPUT_CHANNEL and DATA.INV_UNIFORM_SAMPLE as True. Is it to match the data pre-processing in Caffe2 pre-training? So if I want to reproduce the results on Kinetics, I also need to set them as True? (It seems the config in Kinetics/c2 does not do so).
Contributor guide
Assessment
This issue has not been assessed yet.