A general 3D Object Detection codebase in PyTorch.

Det3D is the first 3D Object Detection toolbox which provides off the box implementations of many 3D object detection algorithms such as PointPillars, SECOND, PIXOR, etc, as well as state-of-the-art methods on major benchmarks like KITTI(ViP) and nuScenes(CBGS). Key features of Det3D include the following aspects:

• Multi Datasets Support: KITTI, nuScenes, Lyft
• Point-based and Voxel-based model zoo
• State-of-the-art performance
• DDP & SyncBN

Please refer to INSTALATION.md.

Please refer to GETTING_STARTED.md.

The original model and prediction files are available in the CBGS README.

car  AP @0.70, 0.70,  0.70:
bbox AP:90.54, 89.35, 88.43
bev  AP:89.89, 87.75, 86.81
3d   AP:87.96, 78.28, 76.99
aos  AP:90.34, 88.81, 87.66

car  [email protected],  0.70,  0.70:
bbox AP:90.63, 88.86, 87.35
bev  AP:89.75, 86.15, 83.00
3d   AP:85.75, 75.68, 68.93
aos  AP:90.48, 88.36, 86.58

• Lyft Config

• Models
  • VoxelNet
  • SECOND
  • PointPillars
• Features
  • Multi task learning & Multi-task Learning
  • Distributed Training and Validation
  • SyncBN
  • Flexible anchor dimensions
  • TensorboardX
  • Checkpointer & Breakpoint continue
  • Self-contained visualization
  • Finetune
  • Multiscale Training & Validation
  • Rotated RoI Align

• To Be Released

  • CGBS on Lyft(val) Dataset

• Models

  • PointRCNN
  • PIXOR

• Support Waymo Dataset.
• Add other 3D detection / segmentation models, such as VoteNet, STD, etc.

Benjin Zhu , Bingqi Ma

Det3D is released under the Apache licenes.

Det3D is a derivative codebase of CBGS, if you find this work useful in your research, please consider cite:

@article{zhu2019class,
  title={Class-balanced Grouping and Sampling for Point Cloud 3D Object Detection},
  author={Zhu, Benjin and Jiang, Zhengkai and Zhou, Xiangxin and Li, Zeming and Yu, Gang},
  journal={arXiv preprint arXiv:1908.09492},
  year={2019}
}

• mmdetection
• mmcv
• second.pytorch
• maskrcnn_benchmark