Awesome Point Cloud Learning
A curated list of papers & resources linked to deep learning on point sets.

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation[https://arxiv.org/abs/1612.00593] [CVPR2017]

[tensorflow][https://github.com/charlesq34/pointnet]

[pytorch][https://github.com/fxia22/pointnet.pytorch]

[pytorch][https://github.com/erikwijmans/Pointnet2_PyTorch]

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space[https://arxiv.org/abs/1706.02413] [NIPS2017]

[tensorflow][https://github.com/charlesq34/pointnet2]

[tensorflow][https://github.com/LinZhuoChen/pointnet2.multigpu]

[pytorch][https://github.com/erikwijmans/Pointnet2_PyTorch]

[pytorch][https://github.com/eriche2016/pointnet2.pytorch]

KCNet: Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling[https://arxiv.org/abs/1712.06760] [CVPR2018]

[caffe][http://www.merl.com/research/?research=license-request&sw=KCNet]

[pytorch][https://github.com/ftdlyc/KCNet_Pytorch]

Pointwise Convolutional Neural Networks[https://arxiv.org/abs/1712.05245] [CVPR2018]

[tensorflow][https://github.com/scenenn/pointwise]
SO-Net: Self-Organizing Network for Point Cloud Analysis[https://arxiv.org/abs/1803.04249] [CVPR2018]

[pytorch][https://github.com/lijx10/SO-Net]
Escape from Cells: Deep Kd-Networks for the Recognition of 3D Point Cloud Models[https://arxiv.org/abs/1704.01222] [ICCV2017]

[pytorch][https://github.com/fxia22/kdnet.pytorch]
Recurrent Slice Networks for 3D Segmentation of Point Clouds[https://arxiv.org/abs/1802.04402] [CVPR2018]

[pytorch][https://github.com/qianguih/RSNet]
SPLATNet: Sparse Lattice Networks for Point Cloud Processing[https://arxiv.org/abs/1802.08275] [CVPR2018]

[caffe][https://github.com/NVlabs/splatnet]

[pytorch][https://github.com/ftdlyc/splatnet_pytorch]

Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs[http://arxiv.org/abs/1711.09869] [CVPR2018]

[pytorch][https://github.com/loicland/superpoint_graph]
Dynamic Graph CNN for Learning on Point Clouds[https://github.com/WangYueFt/dgcnn] [arxiv2018]

[tensorflow][https://github.com/WangYueFt/dgcnn]
Local Spectral Graph Convolution for Point Set Feature Learning[https://arxiv.org/abs/1803.05827] [ECCV2018]

[tensorflow][https://github.com/fate3439/LocalSpecGCN]
SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters[https://arxiv.org/abs/1803.11527] [ECCV2018]

[tensorflow][https://github.com/xyf513/SpiderCNN]
Point Convolutional Neural Networks by Extension Operators[https://arxiv.org/abs/1803.10091] [SIGGRAPH 2018]

[tensorflow][https://github.com/matanatz/pcnn]
PointCNN: Convolution On X-Transformed Points[https://arxiv.org/abs/1801.07791] [NeurIPS 2018]

[tensorflow][https://github.com/yangyanli/PointCNN]

[mxnet][https://github.com/chinakook/PointCNN.MX]

[pytorch][https://github.com/hxdengBerkeley/PointCNN.Pytorch]

PointSIFT: A SIFT-like Network Module for 3D Point Cloud Semantic Segmentation[https://arxiv.org/abs/1807.00652] [arxiv2018]

[tensorflow][https://github.com/MVIG-SJTU/pointSIFT]

[pytorch][https://github.com/Tommymhz/PointNet.PointNet2.PointSIFT.Pytorch]

PointConv: Deep Convolutional Networks on 3D Point Clouds[https://arxiv.org/abs/1811.07246v1] [arxiv2018]

Software
PDAL - Point Data Abstraction Library is a C++ BSD library for translating and manipulating point cloud data.

Potree - Visualization in thge browser Popular tool for visualizing point cloud data.

MeshLab. Open source, portable, and extensible system for the processing and editing of unstructured 3D triangular meshes

CloudCompare. 3D point cloud and mesh processing software Open Source Project

PCL - Point Cloud Library is a standalone, large scale, open project for 2D/3D image and point cloud processing.

libLAS is a C/C++ library for reading and writing the very common LAS LiDAR format.

Algorithms
Pose Optimization Simultaneous Localization and Mapping (SLAM) problems can be posed as a pose graph optimization problem. We have developed a nonlinear optimization algorithm that solves this problem quickly, even when the initial estimate (e.g., robot odometry) is very poor. MIT’s Graph Based SLAM.

Point Set Registration The Orthogonal Procrustes Problem and Kabsch Algorithm requires correspondence between point sets as an input whereas Iterative Closest Point treats correspondence as a variable to be estimated.

Photogrammetry Well known method since the geographical survey days where photographs were used. Matches interest points and calculates the angles to approximate their locations using triangulation.

Pose Graph Optimization To optimize a set of poses, given relative pose between those poses. The difference is that point set registration problems only are moving "model" point set such that the difference between and the static "scene" set is minimized - without taking into consideration that the points in the model set may be captured while moving. During a single lidar scan rotation, the lidar may be moved (as mounted on a vehicle), so the points that are captured first and last differ with respect to their point of origin. In other words, using point cloud from a rotating laser scanner involves recovery of motion and correction of motion distortion in the lidar cloud.

Databases
CMU Databases group maintains a Database of Databases and it’s mostly well populated. It’s missing the geospatial databases like GeoMesa/GeoWave, but you can find other info.

A spatial data warehouse system is necessary for the effective handling of large-scale geospatial data.

Accumulo
GeoMesa Both are based on large key-value stores where large data can be partitioned automatically.
GeoWave developed in part by US National Geospatial-Intelligence Agency.
Formats
Point Cloud
I'll numerically order these in approximate ordering of what I commonly see, and leave bullets for those I've not personally used.

LAS The most common open file format for the interchange of 3-dimensional point cloud data from aerial imaging. Developed primarily for lidar point cloud data this format supports the exchange of any 3-dimensional x,y,z tuplet. The columns are quite fixed (well defined) so using it for terrestrial applications can be challenging.

EPT Entwine Point Tile (EPT) is a simple and flexible octree-based storage format for point cloud data, used in the Entwine and read/writable in PDAL.

PLY Uncompressed text based, and easily read, can export to text or binary file. Can also save the camera pose.

PCD The format used in Point Cloud Library (or PCL) is a large scale 2D/3D image and point cloud processing, free for commercial and research use. Supports viewpoint information.

ROS PointCloud2 In robotics, ROS format is common and point clouds messages are defined in sensor_msgs/PointCloud2. Most likely they will convert it to .pcd because PCL is the de facto library for working with point clouds in ROS.

XYZ This is a popular chemical format for describing atomic coordinates, used in computational chemistry, but it can be used for point cloud.

E57 A compact, vendor-neutral format for storing point clouds, images, and metadata produced by 3D imaging systems, such as laser scanners. The file format is specified by the ASTM, an international standards organization, and it is documented in the ASTM E2807 standard. E57 files do not have a projection. The E57 specification uses a subset of XML that has been extended to support efficient storage of large amounts of binary data. An E57 file is encoded as a hierarchical tree structure, some of which is encoded in XML, and some of which is encoded in a binary format that is not XML. The bulk of the data, including point data and images, is encoded in the binary sections for efficiency. Metadata, such as sensor pose information, is encoded in XML. The binary sections are not embedded in the XML section. Instead, they are located in separate sections of the file and referenced from the XML section.

Binary
TIFF is lossless and can store arbitrary data, commonly used in digital terrain models and ground labeling. See PDAL tutorial for point cloud ground classification
PNG is com- KML (Keyhole Markup Language) is an XML-based language schema for expressing geographic data in an Earth browser, such as Google Earth or Google Maps. KML uses a tag-based structure with nested elements and attributes. For GeoServer, KML files are distributed as a KMZ, which is a zipped KML file.
JPEG The JPEG is a compressed graphic file format, with some loss of quality due to compression. It is best used for photos and not recommended for exact reproduction of data.
GIF The GIF (Graphics Interchange Format) is a bitmap image format best suited for sharp-edged line art with a limited number of colors. This takes advantage of the format’s lossless compression, which favors flat areas of uniform color with well defined edges (in contrast to JPEG, which favors smooth gradients and softer images). GIF is limited to an 8-bit palette, or 256 colors.
SVG SVG (Scalable Vector Graphics) is a language for modeling two-dimensional graphics in XML. It differs from the GIF and JPEG in that it uses graphic objects rather than individual points.
TIFF TIFF (Tagged Image File Format) is a flexible, adaptable format for handling multiple data in a single file. GeoTIFF contains geographic data embedded as tags within the TIFF file.
PNG The PNG (Portable Network Graphics) file format was created as the free, open-source successor to the GIF. The PNG file format supports truecolor (16 million colors) while the GIF supports only 256 colors. The PNG file excels when the image has large, uniformly coloured areas.
OpenLayers is an open source JavaScript library for displaying map data in web browsers. The OpenLayers output has some advanced filters that are not available when using a standalone version of OpenLayers. Further, the generated preview contains a header with easy configuration options for display. Version 3 of the OpenLayers library is used by default. Version 3 can be disabled with the ENABLE_OL3 (true/false) format option or system property. For older browsers not supported by OpenLayers 3, version 2 is used regardless of the setting.
PDF (Portable Document Format) encapsulates a complete description of a fixed-layout 2D document,including any text, fonts, raster images, and 2D vector graphics.
GEOTIFF Georeferencing information to be embedded within a TIFF file. May include: coordinate transforms, map projection, coordinate systems, ellipsoids, datums, and anything else necessary to establish the exact spatial reference for the file. monly used for rastering large map tiles.
KML (Keyhole Markup Language) is an XML-based language schema for expressing geographic data in an Earth browser, such as Google Earth or Google Maps. KML uses a tag-based structure with nested elements and attributes. For GeoServer, KML files are distributed as a KMZ, which is a zipped KML file.
JPEG The JPEG is a compressed graphic file format, with some loss of quality due to compression. It is best used for photos and not recommended for exact reproduction of data.
GIF The GIF (Graphics Interchange Format) is a bitmap image format best suited for sharp-edged line art with a limited number of colors. This takes advantage of the format’s lossless compression, which favors flat areas of uniform color with well defined edges (in contrast to JPEG, which favors smooth gradients and softer images). GIF is limited to an 8-bit palette, or 256 colors.
SVG SVG (Scalable Vector Graphics) is a language for modeling two-dimensional graphics in XML. It differs from the GIF and JPEG in that it uses graphic objects rather than individual points.
TIFF TIFF (Tagged Image File Format) is a flexible, adaptable format for handling multiple data in a single file. GeoTIFF contains geographic data embedded as tags within the TIFF file.
PNG The PNG (Portable Network Graphics) file format was created as the free, open-source successor to the GIF. The PNG file format supports truecolor (16 million colors) while the GIF supports only 256 colors. The PNG file excels when the image has large, uniformly coloured areas.
OpenLayers is an open source JavaScript library for displaying map data in web browsers. The OpenLayers output has some advanced filters that are not available when using a standalone version of OpenLayers. Further, the generated preview contains a header with easy configuration options for display. Version 3 of the OpenLayers library is used by default. Version 3 can be disabled with the ENABLE_OL3 (true/false) format option or system property. For older browsers not supported by OpenLayers 3, version 2 is used regardless of the setting.
PDF (Portable Document Format) encapsulates a complete description of a fixed-layout 2D document,including any text, fonts, raster images, and 2D vector graphics.
GEOTIFF Georeferencing information to be embedded within a TIFF file. May include: coordinate transforms, map projection, coordinate systems, ellipsoids, datums, and anything else necessary to establish the exact spatial reference for the file.
Text
AtomPub (Atom Publishing Protocol) is an application-level protocol for publishing and editing Web Resources using HTTP and XML. Developed as a replacement for the RSS family of standards for content syndication, Atom allows subscription of geo data.
GeoRss RSS (Rich Site Summary) is an XML format for delivering regularly changing web content. GeoRss is a standard for encoding location as part of a RSS feed.supports Layers Preview produces a RSS 2.0 documents, with GeoRSS Simple geometries using Atom.
GeoJSON is a lightweight data-interchange format based on the JavaScript programming language. This makes it an ideal interchange format for browser based applications since it can be parsed directly and easily in to javascript. GeoJSON is a plain text output format that add geographic types to JSON.
CSV (Comma Separated Values) files are text files containing rows of data. Data values in each row are separated by commas. CSV files also contain a comma-separated header row explaining each row’s value ordering. GeoServer’s CSVs are fully streaming, with no limitation on the amount of data that can be outputted.

Tutorials
Data Structures for Large 3D Point Cloud Processing. Data Structures for Large 3D Point Cloud Processing Tutorial at the 13th International Conference on Intelligent Autonomous Systems

INF555 Geometric Modeling: Digital Representation and Analysis of Shapes: lecture 7.

3D Deep Learning on Point Cloud Data

Libraries
PCL - Point Cloud Library is a standalone, large scale, open project for 2D/3D image and point cloud processing.
3DTK - The 3D Toolkit provides algorithms and methods to process 3D point clouds.
PDAL - Point Data Abstraction Library is a C++ BSD library for translating and manipulating point cloud data.
libLAS is a C/C++ library for reading and writing the very common LAS LiDAR format.
entwine is a data organization library for massive point clouds, designed to conquer datasets of hundreds of billions of points as well as desktop-scale point clouds.
PotreeConverter is another data organisation library, generating data for use in the Potree web viewer.
lidR R package for Airborne LiDAR Data Manipulation and Visualization for Forestry Applications.
Software (Open Source)
Paraview. Open-source, multi-platform data analysis and visualization application.
MeshLab. Open source, portable, and extensible system for the processing and editing of unstructured 3D triangular meshes
CloudCompare. 3D point cloud and mesh processing software Open Source Project
OpenFlipper. An Open Source Geometry Processing and Rendering Framework
PotreeDesktop. A desktop/portable version of the web-based point cloud viewer Potree
Servers
LOPoCS is a point cloud server written in Python
Greyhound is a server designed to deliver points from Entwine octrees
Web-based point cloud viewers
Potree is a web-based octree viewer written in Javascript.
Conferences
International LiDAR Mapping Forum International LiDAR Mapping Forum (ILMF)
3D-ARCH is a series of international workshops to discuss steps and processes for smart 3D reconstruction, modelling, accessing and understanding of digital environments from multiple data sources.
Community
Laser Scanning Forum Laser Scanning Forum
Papers
Efficient Processing of Large 3D Point Clouds Jan Elseberg, Dorit Borrmann, Andreas N̈uchtre, Proceedings of the XXIII International Symposium on Information, Communication and Automation Technologies (ICAT '11), 2011

Data Structure for Efficient Processing in 3-D Jean-François Lalonde, Nicolas Vandapel and Martial Hebert, Robotics: Science and Systems I, 2005

An out-of-core octree for massive point cloud processing K. Wenzel, M. Rothermel, D. Fritsch, N. Haala, Workshop on Processing Large Geospatial Data 2014