Open3D-ML is an extension of the Open3D library that adds machine learning tools for 3D data. It provides datasets, pretrained models, and complete pipelines for tasks such as 3D object detection and semantic segmentation of point clouds, so researchers can train and run modern 3D perception models without assembling everything from scratch. The project has earned more than 2,300 stars on GitHub and was still actively updated in September 2026.
Three-dimensional perception โ understanding LiDAR scans, RGB-D captures, and point clouds โ underpins robotics, autonomous driving, and AR applications, but the tooling has historically been fragmented across papers and codebases. Open3D-ML unifies it inside the widely used Open3D ecosystem: it supports both PyTorch and TensorFlow backends, includes standard benchmark datasets with loaders and evaluation, and ships reference implementations of well-known 3D ML architectures. Visualization comes from Open3D itself, so inspecting predictions on point clouds is built in rather than bolted on. Models can be trained, evaluated, and run for inference through a consistent configuration-based pipeline system. The project also provides helper scripts for downloading datasets and converting common formats, which removes much of the setup work that usually precedes any experiment with a new 3D benchmark or a fresh sensor capture.
Open3D-ML is free and open source as part of the Open3D project. There are no licenses or fees; costs are limited to your own compute for training and inference.
Open3D-ML is for researchers and engineers working on 3D perception for robotics, autonomous vehicles, mapping, and AR. It is a strong starting point when you need reproducible baselines and ready pipelines instead of reimplementing papers one by one.