facebookresearch / facebookresearch/sam2
Are there any method for reducing gpu memory overhead?
Open
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
As I execute the SAM2 Model for [2:30 second Video with 24fps]
It takes GPU Memory almost 60GB.
I want to reduce gpu memory usage overhead.
For reducing GPU Memory Overhead
I used **offload_video_to_cpu=True**
But it's still overhead [GPU memory 21GB usage..]
Are there any method to reduce gpu memory's overhead??
1. Like only take few frame's state for reducing gpu overhead
2. Lazy Image Storing to GPU for reducing gpu memory overhead
Contributor guide
Assessment
This issue has not been assessed yet.