How gargi-09/depthar works
DepthAR is a webcam AR application that estimates depth and person semantics, corrects foreground boundaries, stabilizes depth over time, and renders virtual overlays. The README documents a four-stage frame pipeline; the sampled sources implement its vision and rendering components, but the runtime orchestrator is not shown (api/main.py is empty in the excerpt). Accordingly, pipeline links rely on README evidence, and the separate face-reticle and exosuit features are shown without guessed wiring.
Main entry points: scripts Frame workflow · occlusion_reasoner.py Boundary correction · renderer.py Occlusion renderer · depth_estimator.py Depth estimation
Components 9
Vision inference
- Depth estimation
core/depth_estimator.py - Person segmentation
core/segmentor.py
Mask and stability
- Boundary correction
core/occlusion_reasoner.py - Temporal smoothing
core/temporal_smoother.py
AR rendering
- Occlusion renderer
ar/renderer.py - Face reticle
ar/face_anchor.py - Exosuit overlay
ar/exosuit.py
Runtime entry
- Frame workflow
scripts
Other components
- Webcam user
Connections 9
- Webcam user → Frame workflow: provides frames
README.md - Frame workflow → Depth estimation: estimates depth
README.md - Frame workflow → Person segmentation: segments person
README.md - Depth estimation → Boundary correction: provides depth
README.md - Person segmentation → Boundary correction: provides mask
README.md - Boundary correction → Temporal smoothing: corrects boundaries
README.md - Temporal smoothing → Occlusion renderer: stabilizes depth
- Boundary correction → Occlusion renderer: provides occlusion mask
ar/renderer.py - Frame workflow → Occlusion renderer: renders overlay