Three years of lessons compressed into a toolkit you can actually run on your machine today.
🐍
12 Weeks of Runnable Python Code
Each week builds on the last — from NumPy vectorization to a complete end-to-end 3D detector. Every file runs out of the box. No tutorial-ware.
Python · NumPy · PyTorch
🐳
Zero-Setup Docker Environment
A single docker-compose up command gives you a GPU-ready container with CUDA, PyTorch, Open3D, and all sensor fusion libraries pre-installed.
Docker · CUDA · Conda
🗂
Real Dataset Parsing Scripts
TFRecord readers for Waymo Open Dataset and NuScenes loaders — including the calibration math most tutorials skip. Parse your first real LiDAR frame in under 10 minutes.
Waymo · NuScenes · Protobuf
🚧
The Pitfall Guide
A brutally honest doc of every mistake I made in my first year — wrong coordinate frames, miscalibrated extrinsics, leaky validation sets. Avoid 3 months of pain in 20 minutes.
Field Notes · Anti-patterns
📊
Evaluation & mAP Framework
Production-style evaluation scripts with IoU-based matching, precision-recall curves, and the KITTI mAP metrics — the same metrics used in industry benchmarks.
mAP · IoU · NMS
🔗
Private GitHub Repo Access
Full commit history so you can see how the code evolved week by week. Not a final polished snapshot — the messy, real process of building something that works.
GitHub · Version Control
