Skip to content

PyTorch PointCloud¶

PyTorch PointCloud

Deep learning on point clouds, with PyTorch. Models, pretrained weights, datasets, transforms and inferers, in the style of timm and PyG.

Get started Browse the models pip install torch-pointcloud

Classification Classification
Classificationpointnet2-ssg.modelnet40.xu-yan
Part segmentation Part segmentation
Part segmentationpointnext-sm.shapenetpart.openpoints
Indoor segmentation + detection Indoor segmentation + detection
Indoor segmentation + detectionptv3-base.scannet20.pointcept3detr-m.scannet.fair
Feature similarity Feature similarity
Feature similaritysonata-base.pretrain.fair
LiDAR segmentation + detection LiDAR segmentation + detection
LiDAR segmentation + detectionspvcnn-119gmacs.semantickitti.mit-han-labvoxelnext.nuscenes.openpcdet
Large-scale features, 30M points Large-scale features, 30M points
Large-scale features, 30M pointssonata-base.pretrain.fair

In a few lines¶

Every checkpoint is one create_model call away, and ships the transform that turns a raw point cloud into what the network expects:

import torch_pointcloud as tp

model, info = tp.create_model(
    "ptv3-base.scannet20.pointcept",
    task="semantic-segmentation",
    pretrained=True,
    return_info=True,
)
info["transform"]  # the preprocessing pipeline of that checkpoint
info["weights"]["metrics"]  # {"mIoU": 77.42, "OA": 92.00}

tp.list_models(task="detection", pretrained=True)  # all detection checkpoints

Why torch-pointcloud?¶

torch-pointcloud is a library of pretrained models that makes common layers and backbones easy to reuse. It does not replace research-first codebases such as Pointcept, OpenPCDet and MMDetection3D; it complements them with a common interface that makes benchmarking and interoperability across architectures easier.

What's inside¶

  • Get Started

    Install, run a first model, and learn the data format.

  • Models

    PointNet, PointNet++, RandLA-Net, KPConv, PointNeXt, OctFormer, Point Transformer, SPVCNN, and more.

  • Datasets

    ModelNet, ScanNet, S3DIS, ShapeNetPart, ScanObjectNN, SemanticKITTI, Semantic3D, and more.

  • Transforms

    Composable, non-mutating dict transforms inspired by MONAI.

  • Inferers

    Sliding windows, voxel partitions, potential spheres and test-time augmentation for scenes of any size.

  • Tutorials

    Ready-to-use notebooks, from a first classification to survey-scale inference.

  • API Reference

    Auto-generated reference for every public class and function.

  • About

    Why the library exists, how to cite it, and the changelog.

License¶

Apache 2.0, see LICENSE. Pretrained weights keep the license of their source, see THIRD_PARTY_NOTICES.md.