Takes endoscope video and builds a simple 3D map. Parts the camera already passed are marked visited (green). Parts not reached yet are unvisited (red). It also prints a coverage percent.
No extra trackers on the scope. Only the video is used.
- Load frames from a dataset folder or a live camera.
- Mask bright glare (and tools if a YOLO model is present).
- Run SLAM to estimate camera path and 3D points.
- Put points in a voxel grid. Mark voxels near the camera path as visited.
- Save
coverage.ply,metrics.json, andminimap.png.
Batch entry point: python scripts/run_coverage.py --config configs/c3vd.yaml
| File | Topic |
|---|---|
| setup.md | Install and first run |
| datasets.md | C3VD, EndoSLAM, downloads |
| architecture.md | Code layout |
| evaluation.md | Metrics |
| configuration.md | YAML settings |
| live_capture.md | Webcam and live viewer |
| orb_slam3_wsl.md | Build SLAM on WSL2 |
| clinical-context.md | Medical background |
| faq.md | Common questions |
| roadmap.md | Planned work |
| references.md | Papers |
- Research code, not a certified medical device.
- Works best on short phantom clips with ground truth for testing.
- Default SLAM on Windows is a simple fallback. ORB-SLAM3 on WSL2 is better for longer clips.