ALLINONE-Det is a general and strong 3D object detection codebase built on OpenPCDet, which supports more methods, datasets and tools (debugging, recording and analysis).

• Changelog
• Supported Features
• Introduction
• Model Zoo
• Installation
• Getting Started
• Acknowledgement

[2022-01-08] NEW: Update ALLINONE-Det to v0.1.0:

• Initial commit

[2022-01-09] NEW: Update ALLINONE-Det to v0.1.1:

• Support ONCE dataset and more supervised, semi-supervised, unsupervised domain adaptive learning models for it
• Support spconv 1.0~2.x for centerpoint_once detector
• Add PV-RCNN++, Voxel-RCNN related baseline configs on ONCE dataset
• Support CT3D, SA-Det3D, CIA-SSD (SSFA), centerpoint_kitti, PA_AUG, QualityFocalLoss, VariFocalLoss

• Support the latest version of OpenPCDet v0.5.2
• Support KITTI, ONCE, NuScnees, Lyft and Waymo datasets
• Support supervised, semi-supervised, unsupervised domain adaptive learning
• Support more models and modules than OpenPCDet
• Supports more Focal Loss variants, e.g., QualityFocalLoss and VeriFocalLoss
• Support plug-and-play remote visual debugging
• Support unified model configuration, training, recording and analysis
• Support Adaptive Object Augmentation Module
• Support Test Time Augmentation
• Support Balanced Sample Assignment and Objective Module

• Currently supported models:

  • supervised learning methods (total 15)

  Methods	Support datasets	Support features
  SECOND	kitti, once, waymo, nuscenes	center_head, resnet, IOU
  PintPillars	kitti, once, waymo, nuscenes	center_head, resnet
  CBGS	nuscenes, lyft
  PointRCNN	kitti, once	IOU
  Part A^2 Net	kitti, waymo	anchor-free
  PV-RCNN	kitti, once, waymo	center_head, resnet
  PV-RCNN++	once, waymo	center_head, resnet
  Voxel-RCNN	kitti, once, waymo	center_head, resnet, IOU
  CaDDN	kitti
  Centerpoint	kitti, once, waymo, nuscenes	center_head, resnet, RCNN
  PointPainting	once
  CT3D	kitti, once
  CIA-SSD	kitti
  SA-Det3D	kitti, waymo, nuscenes
  PA-AUG	kitti, once	class-separated augmentation

  • semi-supervised learning methods (total 6)

  Methods	Support datasets	Support features
  SemiSECOND	once	IOU
  Pseudo Label	once
  Noisy Student	once
  Mean Teacher	once
  SESS	once
  3DIoUMatch	once

  • unsupervised domain adaptation learning methods (total 3)

  Methods	Support datasets	Support features
  ST3D	once
  SN	once
  Oracle	once

Please refer to OpenPCDet for the installation of ALLINONE-Det.

Thanks to the strong and flexible OpenPCDet codebase maintained by Shaoshuai Shi (@sshaoshuai) and the reproduced benchmark (ONCE_Benchmark) on the ONCE (One Million Scenes) dataset by PointsCoder (@PointsCoder).