3D-Point-Clouds

3D点云SOTA方法,代码,论文,数据集(点云目标检测&分割)

• 如有疑问，微信：shuangyu_ai
• 更多自动驾驶相关交流群，欢迎扫码加入：自动驾驶感知(PCL/ROS+DL)：技术交流群汇总(新版)

应同学建议，创建了星球 【自动驾驶感知(PCL/ROS+DL)】 专注于自动驾驶感知领域，包括传统方法(PCL点云库,ROS)和深度学习（目标检测+语义分割）方法。同时涉及Apollo，Autoware(基于ros2)，BEV感知，三维重建，SLAM(视觉+激光雷达) ，模型压缩（蒸馏+剪枝+量化等），自动驾驶模拟仿真，自动驾驶数据集标注&数据闭环等自动驾驶全栈技术，欢迎扫码二维码加入，一起登顶自动驾驶的高峰！

点云处理方法上主要包括两类方法

• 深度学习方法 [python]
  • 目标检测&语义分割&多目标跟踪（MOT）
  • 【202212done】目标检测最新论文实时更新
  • 【202304done】语义分割最新论文实时更新
• 传统上基于规则的方法 [c++]
  • PCL:https://github.com/HuangCongQing/pcl-learning
  • ROS: https://github.com/HuangCongQing/ROS
  • Apollo笔记：https://github.com/HuangCongQing/apollo_note

@双愚 , 若fork或star请注明来源

TODO

• 【202212done】目标检测最新论文实时更新
• 【202304done】语义分割最新论文实时更新
• 【202209done】目标检测框架(pcdet+mmdetection3d+det3d+paddle3d)文章撰写
• 【202208done】数据集调研总结
• 【202406done】3D语义分割框架综述(mmdetection3d|OpenPCSeg|Pointcept)
• 数据集详细剖析：kitti&waymo&nuScenes
• Apollo学习https://github.com/HuangCongQing/apollo_note

目录

0 目标检测框架(pcdet+mmdetection3d+det3d+paddle3d)

【202209done】目标检测框架(pcdet+mmdetection3d+det3d+paddle3d)文章撰写

代码注解笔记：

1. pcdet:https://github.com/HuangCongQing/pcdet-note
2. mmdetection3d:https://github.com/HuangCongQing/mmdetection3d-note
3. det3d: TODO
4. paddle3d: TODO

1 paper(code)

• paperswithcode: https://paperswithcode.com/

2 Datasets

自动驾驶相关数据集调研总结【附下载地址】(更新ing)

数据集基本处理: 数据集标注文件处理

部分数据下载脚本：https://github.com/HuangCongQing/download_3D_dataset

3 点云可视化

点云可视化笔记和代码：https://github.com/HuangCongQing/Point-Clouds-Visualization

3D点云可视化的库有很多，你的选择可能是：

• pcl 点云可视化 [c++]
• ROS topic可视化 [c++] [python]
• open3D [python]
• mayavi[python]
• matplolib [python]

4 点云数据标注

数据标注工具总结：https://github.com/HuangCongQing/data-labeling-tools

paper(code)

3D_Object_Detection

• >>>目标检测最新论文实时更新

• One-stage
• Two-stage

One-stage

Voxel-Net、SECOND、PointPillars、HVNet、DOPS、Point-GNN、SA-SSD、3D-VID、3DSSD

• Voxel-Net
• SECOND
• PointPillars
• HVNet
• DOPS
• Point-GNN
• SA-SSD
• 3D-VID
• 3DSSD

Two-stage

F-pointNet、F-ConvNet、Point-RCNN、Part-A^2、PV-RCNN、Fast Point RCNN、TANet

• F-pointNet
• F-ConvNet
• Point-RCNN
• Part-A^2
• PV-RCNN
• Fast Point RCNN
• TANet

3D_Semantic_Segmentation

• >>>语义分割最新论文实时更新

PointNet is proposed to learn per-point features using shared MLPs and global features using symmetrical pooling functions. Based on PointNet, a series of point-based networks have been proposed

Point-based Methods: these methods can be roughly divided into pointwise MLP methods, point convolution methods, RNN-based methods, and graph-based methods

1 pointwise MLP methods

PointNet++，PointSIFT，PointWeb，ShellNet，RandLA-Net

PointNet++ PointSIFT PointWeb ShellNet RandLA-Net

2 point convolution methods

PointCNN PCCN A-CNN ConvPoint pointconv KPConv DPC InterpCNN

• PointCNN
• PCCN
• A-CNN
• ConvPoint
• pointconv
• KPConv
• DPC
• InterpCNN

3 RNN-based methods

G+RCU RSNet 3P-RNN DAR-Net

• G+RCU
• RSNet
• 3P-RNN
• DAR-Net

4 graph-based methods

DGCNN SPG SSP+SPG PyramNet GACNet PAG HDGCN HPEIN SPH3D-GCN DPAM

• DGCNN
• SPG
• SSP+SPG
• PyramNet
• GACNet
• PAG
• HDGCN
• HPEIN
• SPH3D-GCN
• DPAM

3D_Instance Segmentation

Datasets

数据集下载

• shell脚本下载方式: https://github.com/HuangCongQing/download_3D_dataset

• https://hyper.ai/datasets
• https://www.graviti.cn/open-datasets

Graviti 收录了 400 多个高质量 CV 类数据集，覆盖无人驾驶、智慧零售、机器人等多种 AI 应用领域。举两个例子： 文章> https://bbs.cvmart.net/topics/3346

• Google数据集搜索：https://toolbox.google.com/datasetsearch
• Datahub，分享高质量数据集平台：https://datahub.io/
• 用于上传和查找数据集的机器学习数据集存储库：https://www.webdoctx.com/www.mldata.org
• datafountain收集数据集：https://www.datafountain.cn/dataSets
• tinymind收集数据集：https://www.tinymind.cn/sites#group_22 看到的一篇文章,里面有介绍很多数据集的：世界上最有价值的不是石油而是数据(附数据资源下载链接)
• https://www.graviti.cn/open-datasets

Datasets数据集汇总

https://github.com/Yochengliu/awesome-point-cloud-analysis#---datasets

• [KITTI] The KITTI Vision Benchmark Suite. [det.]**常用
• [ModelNet] The Princeton ModelNet . [cls.]
• [ShapeNet] A collaborative dataset between researchers at Princeton, Stanford and TTIC. [seg.]
• [PartNet] The PartNet dataset provides fine grained part annotation of objects in ShapeNetCore. [seg.]
• [PartNet] PartNet benchmark from Nanjing University and National University of Defense Technology. [seg.]
• [S3DIS**] The Stanford Large-Scale 3D Indoor Spaces Dataset. [seg.]**常用
• [ScanNet] Richly-annotated 3D Reconstructions of Indoor Scenes. [cls. seg.]
• [Stanford 3D] The Stanford 3D Scanning Repository. [reg.]
• [UWA Dataset] . [cls. seg. reg.]
• [Princeton Shape Benchmark] The Princeton Shape Benchmark.
• [SYDNEY URBAN OBJECTS DATASET] This dataset contains a variety of common urban road objects scanned with a Velodyne HDL-64E LIDAR, collected in the CBD of Sydney, Australia. There are 631 individual scans of objects across classes of vehicles, pedestrians, signs and trees. [cls. match.]
• [ASL Datasets Repository(ETH)] This site is dedicated to provide datasets for the Robotics community with the aim to facilitate result evaluations and comparisons. [cls. match. reg. det]
• [Large-Scale Point Cloud Classification Benchmark(ETH)] This benchmark closes the gap and provides a large labelled 3D point cloud data set of natural scenes with over 4 billion points in total. [cls.]
• [Robotic 3D Scan Repository] The Canadian Planetary Emulation Terrain 3D Mapping Dataset is a collection of three-dimensional laser scans gathered at two unique planetary analogue rover test facilities in Canada.
• [Radish] The Robotics Data Set Repository (Radish for short) provides a collection of standard robotics data sets.
• [IQmulus & TerraMobilita Contest] The database contains 3D MLS data from a dense urban environment in Paris (France), composed of 300 million points. The acquisition was made in January 2013. [cls. seg. det.]
• [Oakland 3-D Point Cloud Dataset] This repository contains labeled 3-D point cloud laser data collected from a moving platform in a urban environment.
• [Robotic 3D Scan Repository] This repository provides 3D point clouds from robotic experiments，log files of robot runs and standard 3D data sets for the robotics community.
• [Ford Campus Vision and Lidar Data Set] The dataset is collected by an autonomous ground vehicle testbed, based upon a modified Ford F-250 pickup truck.
• [The Stanford Track Collection] This dataset contains about 14,000 labeled tracks of objects as observed in natural street scenes by a Velodyne HDL-64E S2 LIDAR.
• [PASCAL3D+] Beyond PASCAL: A Benchmark for 3D Object Detection in the Wild. [pos. det.]
• [3D MNIST] The aim of this dataset is to provide a simple way to get started with 3D computer vision problems such as 3D shape recognition. [cls.]
• [WAD] [ApolloScape] The datasets are provided by Baidu Inc. [tra. seg. det.]
• [nuScenes] The nuScenes dataset is a large-scale autonomous driving dataset.用过
• [PreSIL] Depth information, semantic segmentation (images), point-wise segmentation (point clouds), ground point labels (point clouds), and detailed annotations for all vehicles and people. [paper] [det. aut.]
• [3D Match] Keypoint Matching Benchmark, Geometric Registration Benchmark, RGB-D Reconstruction Datasets. [reg. rec. oth.]
• [BLVD] (a) 3D detection, (b) 4D tracking, (c) 5D interactive event recognition and (d) 5D intention prediction. [ICRA 2019 paper] [det. tra. aut. oth.]
• [PedX] 3D Pose Estimation of Pedestrians, more than 5,000 pairs of high-resolution (12MP) stereo images and LiDAR data along with providing 2D and 3D labels of pedestrians. [ICRA 2019 paper] [pos. aut.]
• [H3D] Full-surround 3D multi-object detection and tracking dataset. [ICRA 2019 paper] [det. tra. aut.]
• [Argoverse BY ARGO AI] Two public datasets (3D Tracking and Motion Forecasting) supported by highly detailed maps to test, experiment, and teach self-driving vehicles how to understand the world around them.[CVPR 2019 paper][tra. aut.]
• [Matterport3D] RGB-D: 10,800 panoramic views from 194,400 RGB-D images. Annotations: surface reconstructions, camera poses, and 2D and 3D semantic segmentations. Keypoint matching, view overlap prediction, normal prediction from color, semantic segmentation, and scene classification. [3DV 2017 paper] [code] [blog]
• [SynthCity] SynthCity is a 367.9M point synthetic full colour Mobile Laser Scanning point cloud. Nine categories. [seg. aut.]
• [Lyft Level 5] Include high quality, human-labelled 3D bounding boxes of traffic agents, an underlying HD spatial semantic map. [det. seg. aut.]
• [SemanticKITTI] Sequential Semantic Segmentation, 28 classes, for autonomous driving. All sequences of KITTI odometry labeled. [ICCV 2019 paper**] [seg. oth. aut.]**常用
• [NPM3D] The Paris-Lille-3D has been produced by a Mobile Laser System (MLS) in two different cities in France (Paris and Lille). [seg.]
• [The Waymo Open Dataset] The Waymo Open Dataset is comprised of high resolution sensor data collected by Waymo self-driving cars in a wide variety of conditions. [det.]
• [A*3D: An Autonomous Driving Dataset in Challeging Environments] A*3D: An Autonomous Driving Dataset in Challeging Environments. [det.]
• [PointDA-10 Dataset] Domain Adaptation for point clouds.
• [Oxford Robotcar] The dataset captures many different combinations of weather, traffic and pedestrians. [cls. det. rec.]

常用分割数据集

• [S3DIS**] The Stanford Large-Scale 3D Indoor Spaces Dataset. [seg.] [常用]
• [SemanticKITTI] Sequential Semantic Segmentation, 28 classes, for autonomous driving. All sequences of KITTI odometry labeled. [ICCV 2019 paper**] [seg. oth. aut.] [常用]
• Semantic3d

常用分类数据集

todo

常用目标检测数据集

• [KITTI] The KITTI Vision Benchmark Suite. [det.]**常用
• [nuScenes] The nuScenes dataset is a large-scale autonomous driving dataset.用过
• [The Waymo Open Dataset] The Waymo Open Dataset is comprised of high resolution sensor data collected by Waymo self-driving cars in a wide variety of conditions. [det.]

References

• https://github.com/timzhang642/3D-Machine-Learning
• https://github.com/victorphd/autonomous-vahicles-learning-resource
• https://github.com/Yochengliu/awesome-point-cloud-analysis
• https://github.com/NUAAXQ/awesome-point-cloud-analysis-2021
• https://github.com/QingyongHu/SoTA-Point-Cloud
• https://arxiv.org/abs/1912.12033 : Deep Learning for 3D Point Clouds: A Survey
• https://github.com/zhulf0804/3D-PointCloud

License

Copyright (c) 双愚. All rights reserved.

Licensed under the MIT License.

---

微信公众号：【双愚】（huang_chongqing） 聊科研技术,谈人生思考,欢迎关注~

往期推荐：

1. 本文不提供职业建议，却能助你一生
2. 聊聊我们大学生面试
3. 清华大学刘知远：好的研究方法从哪来

本人创建星球 【自动驾驶感知(PCL/ROS+DL)】 专注于自动驾驶感知领域，包括传统方法(PCL点云库,ROS)和深度学习（目标检测+语义分割）方法。同时涉及Apollo，Autoware(基于ros2)，BEV感知，三维重建，SLAM(视觉+激光雷达) ，模型压缩（蒸馏+剪枝+量化等），自动驾驶模拟仿真，自动驾驶数据集标注&数据闭环等自动驾驶全栈技术，欢迎扫码二维码加入，一起登顶自动驾驶的高峰！