facebookresearch / facebookresearch/sam2
Multiple Occlusions , Segmentation of sports video.
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
I'm using SAM2 for tracking in fast-paced hockey videos and encountering persistent occlusion issues affecting accuracy. I'm interested in practical "how-to" guidance on effectively handling occlusions with fine-tuning the model or possibly with SAM2Long or Samurai?
Could someone please share:
Key parameters or configurations recommended for fast-moving sports scenarios.
Any practical tips or experiences dealing with similar challenges.
Clear steps for patching SAM2 to run SAM2Long or Samurai.
Here's a short video illustrating my issue:
https://youtu.be/MHLFzKLSEec
Contributor guide
Assessment
This issue has not been assessed yet.