Skip to content
Search

Point cloud

lastmod 2026-09-14created 2020-05-2148 edits

A point cloud is a set of points in a coordinate system.

In a three-dimensional coordinate system, points are usually defined by X, Y, and Z coordinates and are often used to represent the surface of an object. Point clouds can be obtained with 3D scanners. A 3D scanner automatically measures a large number of points on the surface of an object and often outputs the resulting point cloud as a file. A set of points measured by such a machine may also be represented as a point cloud.

3D point clouds obtained this way are used in a variety of fields, such as CAD modeling for automation, meteorology, quality inspection, visualization, animation, rendering, and mass customization.

Point clouds can be rendered and viewed directly, but since it is hard to use point clouds directly in most 3D applications, they are converted into polygon meshes, triangle meshes, NURB models, or CAD models through a surface reconstruction process.

Categories

File formats

  • Point Cloud Data (PCD; .pcd)
  • PLY (file format) (Polygon File Format; PLY; .ply) - a polygon file format, developed at Stanford University by Turk et al
  • STL - a file format native to the stereolithography CAD software created by 3D Systems
  • OBJ - a geometry definition file format first developed by Wavefront Technologies
  • X3D - the ISO standard XML-based file format for representing 3D computer graphics data
  • npy, npz - Numpy file formats

Viewer

Deep learning based

Topics

Point Cloud Labeling Tools

List of Annotation Tools for Point Cloud:

  • Segments.ai
  • Deepen
  • Kognic
  • SageMaker Ground Truth
  • CVAT
  • Dataloop
  • Supervisely
  • Scale.ai
  • Understand.ai
  • Telus International
  • Sama
  • Appen
  • Scalabel

File Formats

PLY and PCD are both file formats for storing 3D point clouds, but they differ in their characteristics and intended use.

PLY (Polygon File Format)

A general-purpose 3D data format developed at Stanford University.

It can store not only point clouds but also mesh (mesh) data, so it has a wide range of uses. It supports two formats, ASCII and binary, and the header defines elements (element) such as vertex, face, and edge, along with the properties (property) of each element. Various properties such as color (RGB), normal vectors (normal), and texture coordinates can be added flexibly, and the file structure is relatively simple, making it easy to parse. It is widely used in many fields, including 3D scanning, computer graphics, and visualization.

PCD (Point Cloud Data)

A format defined by the PCL (Point Cloud Library) library, specialized for point cloud processing.

It supports three formats, ASCII, binary, and binary_compressed, and the header explicitly defines the data structure (VERSION, FIELDS, SIZE, TYPE, COUNT, WIDTH, HEIGHT, VIEWPOINT, POINTS, DATA). It can represent an organized point cloud (structured point cloud), so data can be stored in a WIDTH×HEIGHT structure like a 2D image. It is very well integrated with PCL and is mainly used for developing point cloud algorithms in areas such as robot vision, autonomous driving, and LIDAR data processing.

Comparison

In practice,

  • PCD is convenient for PCL-based projects,
  • while PLY offers better compatibility if the goal is general-purpose data exchange or visualization.

Converting between the two formats is also easy with libraries such as PCL or Open3D.

Models

  • PointNet
  • PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
  • KPConv: Flexible and Deformable Convolution for Point Clouds
  • Relation-Shape Convolutional Neural Network for Point Cloud Analysis
  • 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks
  • RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds

Datasets

  • ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes
  • Joint 2D-3D-Semantic Data for Indoor Scene Understanding
  • ShapeNet: An Information-Rich 3D Model Repository
  • 3D ShapeNets: A Deep Representation for Volumetric Shapes
  • 3DMatch: Learning the Matching of Local 3D Geometry in Range Scans

Industrial 3D Point Cloud Korean: 이상 탐지 (anomaly detection) datasets

MVTec 3D-AD - TIFF (organized point cloud), PNG (RGB)
A dataset for unsupervised 3D anomaly detection and localization, consisting of 4,147 high-resolution 3D point cloud scans across 10 real-object categories. Includes defects such as scratches, dents, holes, contamination, and deformation. Suitable for pre-training or transfer learning of weld defect detection models.
Class list: bagel, cable_gland, carrot, cookie, dowel, foam, peach, potato, rope, tire
(2112.09045) The MVTec 3D-AD Dataset for Unsupervised 3D Anomaly Detection and Localization 1
Download page - https://www.mvtec.com/research-teaching/datasets/mvtec-3d-ad/downloads
Google Drive Download Link - 14.2 GBytes compressed file
Evaluation Code - Mvtec-3d-ad-evaluation_code.tar.xz
Github - Real3D-AD (NeurIPS 2023)
A high-precision point cloud anomaly detection dataset with a total of 1,254 samples across 12 categories. Each item consists of tens of thousands to millions of points (resolution 0.0010mm~0.0015mm). Provides 360-degree coverage and complete prototypes. PCD and PLY files are provided.
Airplane, Car, Candybar, Chicken, Diamond, Duck, Fish, Gemstone, Seahorse, Shell, Starfish, Toffees
Liu et al., "Real3D-AD: A Dataset of Point Cloud Anomaly Detection", NeurIPS 2023
Eyecandies
A synthetic multimodal 3D anomaly detection dataset. Provides RGB + Depth + Normal Map. Often used as a benchmark together with MVTec 3D-AD.
Available for download from the Eyecandies Dataset | Download page.

Other 3D Point Cloud datasets

AAU Sewer Defect Point Cloud Dataset
This is the first public 3D dataset of AAU sewer pipes and defects. Most of the dataset consists of synthetic data, and the point clouds of real sewer pipes were recorded in a laboratory setting at Aalborg University in Denmark.
A set of constructed plastic pipes was used, and defects were created by displacing the pipes or placing rubber rings or bricks inside them. The data was collected using a Pico Flexx Time-of-Flight sensor.
OpenTrench3D - A Photogrammetric 3D Point Cloud Dataset for Semantic Segmentation of Underground Utilities
A photogrammetry-based 3D point cloud dataset for semantic segmentation of underground utilities
Github - https://github.com/SimonBuusJensen/OpenTrench3D
Google Drive Download Link - About 5.4 GBytes compressed file

Other sample data

  • For small-scale testing:
    • Open3D — Can be downloaded directly in code via the open3d.data module (bunny, eagle, etc.)
    • PCL (Point Cloud Library) — PCD files in GitHub PointCloudLibrary/data
  • Well-known 3D scans:
    • Stanford 3D Scanning Repository — graphics.stanford.edu/data/3Dscanrep/ (bunny, dragon, happy buddha, etc. in PLY format)
  • Large-scale / real-world
    • KITTI — cvlibs.net/datasets/kitti/ (autonomous driving LiDAR, velodyne point clouds)
    • ModelNet — modelnet.cs.princeton.edu (3D CAD models, 10/40 categories)
    • ShapeNet — shapenet.org (large-scale 3D shapes, registration required)
    • Semantic3D — semantic3d.net (large-scale outdoor colored point clouds, 4B+ points)
    • S3DIS — Stanford indoor 3D dataset

If you just want to grab one quickly, the Stanford bunny is the most classic.

See also

Favorite site

References

  1. 2112.09045v1.pdf ↩

author Zer0 sitemap