facebookresearch / facebookresearch/sam2
image_size vs window_spec
Open
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
The default image_size appears to be 1024 for all models, but the window spec for T, S & B+ appear to be optimimal for image sizes that are multiples of 224 (ie 896 or 1120 would be better), only the L will work with multiples of 256 w/ extra padding overhead for the windowing...
EDIT: for L, that should have read 'w/o' 'without extra padding overhead'
Curious if there was a reason for this?
Contributor guide
Assessment
This issue has not been assessed yet.