Models¶

torch-pointcloud ships architectures for point cloud classification, semantic and part segmentation, object detection and self-supervised pretraining, behind one timm-style factory:

import torch_pointcloud as tp

model = tp.create_model("pointnext-sm.scanobjectnn-hardest.openpoints", pretrained=True)
tp.list_models(task="classification", pretrained=True)  # or "semantic-segmentation", "detection", ...

Weights are downloaded from the Hugging Face Hub on first use and cached under ~/.cache/torch-pointcloud/models (TORCH_POINTCLOUD_CACHE_DIR moves the cache). Each checkpoint keeps the license of its source, listed on the task pages, and a few are restricted to non-commercial use: see THIRD_PARTY_NOTICES.md.

Five pretrained checkpoints on two committed sample clouds: object classification, part segmentation, scene segmentation, 3D detection, and LiDAR segmentation

Tasks¶

Classification

One label per cloud: run, evaluate and fine-tune a classifier.
Semantic segmentation

One label per point: voxelization, full-resolution predictions, mIoU.
Part segmentation

Category-conditioned part labels and the ShapeNetPart protocol.
Object detection

Oriented boxes: decode, filter with NMS, score with mAP.
Feature maps

The representation under the head: embeddings, retrieval, PCA.

Scores¶

Every pretrained checkpoint is evaluated with the protocol of its reference implementation. The plot compares the measured score with the published one, and each task page lists the numbers.

Architectures¶

Classification¶

Best for shape classification (ModelNet40, ScanObjectNN, ShapeNet objects). Inputs are single object scans, outputs are scene-level class predictions.

Model	Year	Checkpoints
PointNet PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation	2016	–
PointNet++ PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space	2017	4
DGCNN Dynamic Graph CNN for Learning on Point Clouds	2018	2
PointCNN PointCNN: Convolution On \(\mathcal{X}\)-Transformed Points	2018	–
PointConv PointConv: Deep Convolutional Networks on 3D Point Clouds	2018	1
KPConv KPConv: Flexible and Deformable Convolution for Point Clouds	2019	–
Point Transformer V1 Point Transformer	2020	–
Point-BERT Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling	2021	6
Point-M2AE Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training	2022	3
Point-MAE Masked Autoencoders for Point Cloud Self-supervised Learning	2022	5
PointMLP Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework	2022	4
PointNeXt PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies	2022	2
OctFormer OctFormer: Octree-based Transformers for 3D Point Clouds	2023	1
PointGPT PointGPT: Auto-regressively Generative Pre-training from Point Clouds	2023	15
PointMamba PointMamba: A Simple State Space Model for Point Cloud Analysis	2024	4

Segmentation¶

Best for dense per-point labeling (S3DIS, ScanNet, SemanticKITTI). Inputs are large scenes, outputs are per-point class predictions.

Model	Year	Checkpoints
PointNet PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation	2016	–
PointNet++ PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space	2017	7
DGCNN Dynamic Graph CNN for Learning on Point Clouds	2018	7
PointCNN PointCNN: Convolution On \(\mathcal{X}\)-Transformed Points	2018	–
KPConv KPConv: Flexible and Deformable Convolution for Point Clouds	2019	8
PVCNN Point-Voxel CNN for Efficient 3D Deep Learning	2019	3
PVCNN++ Point-Voxel CNN for Efficient 3D Deep Learning	2019	2
RandLA-Net RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds	2019	6
SpUNet 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks	2019	1
Point Transformer V1 Point Transformer	2020	1
SPVCNN Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution	2020	3
Point Transformer V2 Point Transformer V2: Grouped Vector Attention and Partition-based Pooling	2022	–
PointMLP Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework	2022	–
PointNeXt PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies	2022	23
SPFormer-UNet Superpoint Transformer for 3D Scene Instance Segmentation	2022	1
OctFormer OctFormer: Octree-based Transformers for 3D Point Clouds	2023	2
Point Transformer V3 Point Transformer V3: Simpler, Faster, Stronger	2023	4
SphereFormer Spherical Transformer for LiDAR-based 3D Recognition	2023	–
Concerto Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations	2025	1
Sonata Sonata: Self-Supervised Learning of Reliable Point Representations	2025	1
Utonia Utonia: Toward One Encoder for All Point Clouds	2026	1

Part segmentation¶

Best for object parts (ShapeNetPart). Inputs are single objects and their category, outputs are per-point part labels.

Model	Year	Checkpoints
DGCNN Dynamic Graph CNN for Learning on Point Clouds	2018	1
Point-M2AE Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training	2022	1
Point-MAE Masked Autoencoders for Point Cloud Self-supervised Learning	2022	1
PointNeXt PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies	2022	3

Detection¶

Predict 3D bounding boxes for indoor scenes (ScanNet, SUN RGB-D) or driving scenes (KITTI, nuScenes, Waymo).

Model	Year	Checkpoints
PointPillars PointPillars: Fast Encoders for Object Detection from Point Clouds	2018	2
PointRCNN PointRCNN: 3D Object Proposal Generation and Detection from Point Cloud	2018	1
SECOND SECOND: Sparsely Embedded Convolutional Detection	2018	2
VoteNet Deep Hough Voting for 3D Object Detection in Point Clouds	2019	2
3DETR An End-to-End Transformer Model for 3D Object Detection	2021	3
VoxelNeXt VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking	2023	1
LION LION: Linear Group RNN for 3D Object Detection in Point Clouds	2024	1
Voxel-Mamba Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection	2024	–

Self-supervised pretraining¶

Backbones pretrained without labels, registered with task="pretraining"; the fine-tuned classification / segmentation heads are registered under their downstream task.

Model	Year	Checkpoints
Point-BERT Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling	2021	2
Point-M2AE Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training	2022	1
Point-MAE Masked Autoencoders for Point Cloud Self-supervised Learning	2022	1
PointGPT PointGPT: Auto-regressively Generative Pre-training from Point Clouds	2023	3
PointMamba PointMamba: A Simple State Space Model for Point Cloud Analysis	2024	1
Concerto Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations	2025	7
Sonata Sonata: Self-Supervised Learning of Reliable Point Representations	2025	2
Utonia Utonia: Toward One Encoder for All Point Clouds	2026	3