facebookresearch / facebookresearch/sam2

Community tool for long-video batch processing with SAM2 (simple-sam2)

Open
#757 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Jupyter Notebook
Stars
19.9k
Forks
2.5k
PR merge metrics
No merged PRs in 30d

Description

Hi SAM2 team,

I built a lightweight wrapper around SAM2 video predictor that solves
long-video segmentation - SAM2 currently loads the entire video into
memory which makes it unusable for longer clips.

simple-sam2 adds:
- Batch processing :- only `batch_size` frames in memory at once, keeping
GPU/CPU memory usage constant regardless of video length

- Unified prompt API :- mask, points, and box in a single call (SAM2 natively
requires mask OR points+box separately)

- Direct video file input:- pass any .mp4/.avi/.mov directly, frames are
extracted automatically via OpenCV, no manual preprocessing needed

- Selective frame range processing :- specify `start_frame_idx` and `end_frame_idx`
to segment only a portion of the video instead of processing the entire clip,
useful for long videos where only a specific segment is of interest

- Persistent storage layout :- extracted frames and output masks are organized
under a canonical directory structure, so re-runs skip re-extraction and
masks are always easy to locate

- Carry-over mask mechanism :- object identity is maintained across batch
boundaries by propagating the last frame's mask as the seed for the next batch

PyPI: https://pypi.org/project/simple-sam2/
GitHub: https://github.com/varun-kolluru/simple_sam2

Sharing here in case it's useful to others in the community. Happy to
answer questions or take feedback.

Contributor guide

Open the contributing guide

Research direction

Review the linked simple_sam2 GitHub project and PyPI package first; the issue does not identify a SAM2 repository file, test, or entry point to change. No repository change or completion condition is specified, so clarify whether this is intended as an integration, documentation link, or community-tool announcement.

Written by the indexing model from the issue text.

Assessment

Tech stack
opencv
Domain
computer-vision
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.