facebookresearch / facebookresearch/sam2
Segmentation Tuning possibilities
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
Hello, thanks for this very impressive tool!
Trying the large model out of the box on fish cell grayscale images I see that sometimes it segment a bit more than what a human would do, for example:
and also there are disjoint regions for the same mask in some images for instance:
what are the tuning possibilities? I see that `SAM2ImagePredictor` only has a few arguments, playing with max_sprinkle_area and max_hole_area does not seems to have much impact, while mask_threshold does. Any suggestions?
Thanks
Contributor guide
Assessment
This issue has not been assessed yet.