How nghiaho12/open3d_truedepth_registration works
This is an iPhone TrueDepth 3D-scanning pipeline. The user supplies calibration, RGB and depth captures plus registration and filtering options; the application builds per-frame point clouds, aligns them into a global reconstruction, then writes a point cloud or optional mesh and can visualize results. Image decoding and projection are separated into ImageDepth, while pose-graph optimization also has a C++/pybind11 implementation. Some process3d sections were omitted, so uncertain calls to the optimizer and visualization are not asserted as verified.
Main entry points: process3d.py Scan pipeline · run.py CLI options · image_depth.py Frame point clouds · pose_graph.cpp Pose optimization
Components 11
Application entry
- CLI options
run.py
Frame processing
- Frame point clouds
lib/image_depth.py
Registration
- Scan pipeline
lib/process3d.py - Frame alignment
lib/process3d.py - Pose optimization
cpp/pose_graph.cpp
Reconstruction output
- Global point cloud
lib/process3d.py - Camera visualization
lib/process3d.py - Poisson mesh
lib/process3d.py
Other components
- Scanner user
- Capture dataset
- PLY output
Connections 13
- Scanner user → CLI options: sets options
run.py - Capture dataset → CLI options: provides folder
run.py - CLI options → Scan pipeline: runs
run.py - Scan pipeline → Frame point clouds: loads frames
lib/process3d.py - Capture dataset → Frame point clouds: provides RGB depth
README.md - Scan pipeline → Frame alignment: registers frames
README.md - Scan pipeline → Pose optimization: may optimize poses
- Frame alignment → Global point cloud: merges frames
lib/process3d.py - Scan pipeline → Camera visualization: creates cameras
lib/process3d.py - Scan pipeline → Poisson mesh: optionally reconstructs
lib/process3d.py - Global point cloud → Poisson mesh: supplies points
lib/process3d.py - Scan pipeline → PLY output: writes result
README.md - Poisson mesh → PLY output: writes mesh
lib/process3d.py