Det3D

A general 3D Object Detection codebase in PyTorch

Call for contribution.

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

Introduction

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

Installation

Please refer to INSTALATION.md.

Quick Start

Please refer to GETTING_STARTED.md.

Model Zoo and Baselines

3DBN on KITTI(val) Dataset 3:1

bbox AP:90.55, 89.42, 88.24
bev  AP:90.20, 88.30, 79.59
3d   AP:89.43, 85.48, 77.36
aos  AP:89.85, 88.14, 86.94

To Be Released

1. PointPillars on NuScenes(val) Dataset
2. CGBS on NuScenes(val) Dataset
3. CGBS on Lyft(val) Dataset

Currently Support

• 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

TODO List

• Models
  • PointRCNN
  • PIXOR

Det3D is released under the Apache licenes.

Acknowledgement

• mmdetection
• mmcv
• second.pytorch
• maskrcnn_benchmark