tianweiy/CenterPoint

Health v2 · maintenance only

Adoption not part of the score

2182 stars · 476 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official PyTorch implementation of CenterPoint, a CVPR 2021 method that performs 3D object detection and tracking from LiDAR point clouds by representing objects as center keypoints in bird's-eye view. It includes training and evaluation pipelines plus pretrained model zoos for the Waymo Open Dataset and nuScenes, where tracking reduces to greedy closest-point matching.

Use cases

• detect 3d objects in lidar point clouds
• run multi-object tracking on autonomous driving data
• train and evaluate models on nuScenes or Waymo Open Dataset
• reproduce a strong lidar-only 3d detection baseline for research
• benchmark real-time 3d detection performance
• explore center-based keypoint detection for point clouds

When to choose

• you need a proven, high-accuracy LiDAR 3D detection and tracking baseline for Waymo or nuScenes
• you are doing research on 3D object detection and want reproducible CVPR 2021 results and pretrained checkpoints
• you want tracking as simple greedy point matching without separate tracking models

When to avoid

• your data is 2D images rather than 3D point clouds
• you need a turnkey production system rather than a research training/evaluation codebase
• you cannot work with PyTorch and GPU/CUDA training setups

Facets

library · maturity maintenance

deep-learning machine-learning computer-vision deep-learning computer-vision autonomous-vehicles machine-learning robotics python lidar point-cloud 3d-object-detection object-tracking nuscenes waymo-open-dataset pytorch model-zoo research-code bird-eye-view keypoint-detection cvpr-2021 gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="tianweiy/CenterPoint")