How aevainc/doppler-icp works
Project type and purpose
• This is an open-source algorithmic tool (DICP: Doppler Iterative Closest Point) written in Python, built atop a custom fork of the Open3D library.
File structure
• Top-level:
• Open3D/ ← forked C++/Python library (submodule)
• dataset/ ← input point-cloud sequences + calibration + reference poses
• scripts/ ← Python entry points and modules
• images/ ← illustrations for README/poster
• requirements.txt, README.md, .gitmodules, etc.
• scripts/:
• run.py ← CLI for running ICP (Doppler or point-to-plane)
• registration.py ← core Doppler-ICP and geometry-ICP implementations
• utils.py ← I/O, file parsing, common routines
• Build/deploy:
• Open3D build (cmake → Python wheel)
• pip install -r requirements.txt
• git lfs pull for large dataset files
README for additional insights
• Describes algorithm purpose: joint optimization of Doppler and geometric objectives
Detail is key
• Build process of Open3D as a separate “build & install” component if desired
• File-format boundaries (.bin→PointCloud, JSON→calibration, TXT→poses)
• Versioning/branching only if relevant (e.g., git submodule pointer)