facebookresearch / facebookresearch/sam2
Poor masking at edges of the frame
Open
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
When an object extends all the way to the edge of the frame, the mask returned by sam2 it often bumpy around the edge. Does anyone know if it would be possible to make a modification to extend the image edges by a few pixels immediately before it does the prediction, and then crop the extra pixels back off when it returns the result?
Contributor guide
Assessment
This issue has not been assessed yet.