What to build
Tracking issue for video object tracking using SAM 2 / SAM 3 memory across frames, so labelme's video/multi-frame annotation can propagate an object across frames instead of re-annotating each frame. SAM 3 (released 2025-11) adds promptable concept segmentation and tracking across video; the underlying models are already reachable via osam.
This is a large bet and a HITL tracking issue: it needs design (frame model, propagation UX, storage) before implementation slices are cut.
Acceptance criteria
Blocked by
None - design discussion can start immediately
What to build
Tracking issue for video object tracking using SAM 2 / SAM 3 memory across frames, so labelme's video/multi-frame annotation can propagate an object across frames instead of re-annotating each frame. SAM 3 (released 2025-11) adds promptable concept segmentation and tracking across video; the underlying models are already reachable via osam.
This is a large bet and a HITL tracking issue: it needs design (frame model, propagation UX, storage) before implementation slices are cut.
Acceptance criteria
Blocked by
None - design discussion can start immediately