All Tools
P
OtherFreeOpen Source
PADDLE3D
3D detection and BEV perception toolkit on PaddlePaddle
Apache-2.0
ABOUT
Lidar, camera, and BEV 3D detection usually live in separate research repos. Paddle3D puts PointPillars, CenterPoint, BEVFormer, PETR, and related models on PaddlePaddle with KITTI, nuScenes, and Waymo pipelines so driving teams train and deploy one 3D perception stack.
INSTALL
pip install paddle3dINTEGRATION GUIDE
1. Train PointPillars or CenterPoint on lidar point clouds
2. Run monocular 3D detection with SMOKE or CaDDN on camera images
3. Train BEVFormer or PETR multi-camera 3D detection for driving
4. Segment point clouds and export models toward Apollo deployment
TAGS
pythoncomputer-vision3d-detectionautonomous-drivingpaddlepaddlebevlidar