README.md

metadata

license: mit
library_name: torch-pointcloud
tags:
  - point-cloud
  - 3d
  - pytorch
  - torch-pointcloud
  - pointnext
  - segmentation
datasets:
  - shapenetpart
model-index:
  - name: pointnext-sm.shapenetpart.openpoints
    results:
      - task:
          type: point-cloud-segmentation
        dataset:
          name: ShapeNetPart
          type: shapenetpart
        metrics:
          - name: ins_mIoU
            type: mean_iou
            value: 86.88
          - name: cls_mIoU
            type: mean_iou
            value: 84.48

Model card for pointnext-sm.shapenetpart.openpoints

A PointNeXt point cloud segmentation model (scaled PointNet++ with inverted residual blocks). Trained on ShapeNetPart.

Model Details

• Model Type: Point cloud semantic segmentation
• Model Stats:
  • Params (M): 1.0
  • Input channels: 7
  • Classes: 50
  • Features: 96
• Dataset: ShapeNetPart
• Metrics: ins_mIoU 86.88, cls_mIoU 84.48 (reference 86.7)
• Paper: PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies
• Converted from: guochengqian/PointNeXt (MIT)
• Library: torch-pointcloud

Install

pip install torch-pointcloud

Usage

import torch
import torch_pointcloud as tp
from torch_pointcloud.utils.data import collate

model, info = tp.create_model(
    "pointnext-sm.shapenetpart.openpoints",
    task="segmentation",
    pretrained=True,
    return_info=True,
)
model = model.eval()

# synthetic sample with the keys a dataset provides
num_points = 8192
sample = {
    "pos": torch.randn(num_points, 3),
    "normal": torch.randn(num_points, 3),
    "category": torch.tensor(0),
    "segment": torch.zeros(num_points, dtype=torch.long),
}
data = info["transform"](sample)
data = collate([data])

with torch.no_grad():
    logits = model(data.get("x"), data["pos"], data["batch"], data["category"])

Feature extraction

with torch.no_grad():
    features = model.forward_features(data.get("x"), data["pos"], data["batch"])

model.reset_classifier(num_classes=0)
with torch.no_grad():
    features = model(data.get("x"), data["pos"], data["batch"], data["category"])  # (N, 96)

Citation

@inproceedings{qian2022pointnext,
  title   = {PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies},
  author  = {Guocheng Qian and Yuchen Li and Houwen Peng and Jinjie Mai and Hasan Abed Al Kader Hammoud and Mohamed Elhoseiny and Bernard Ghanem},
  booktitle = {NeurIPS},
  year    = {2022}
}

@article{yi2016shapenetpart,
  title   = {A Scalable Active Framework for Region Annotation in {3D} Shape Collections},
  author  = {Yi, Li and Kim, Vladimir G. and Ceylan, Duygu and Shen, I-Chao and Yan, Mengyan and Su, Hao and Lu, Cewu and Huang, Qixing and Sheffer, Alla and Guibas, Leonidas},
  journal = {ACM Transactions on Graphics (TOG)},
  volume  = {35},
  number  = {6},
  year    = {2016}
}

@software{dujardin2026pytorchpointcloud,
  author  = {Arthur Dujardin},
  title   = {PyTorch PointCloud},
  year    = {2026},
  doi     = {10.5281/zenodo.22159632},
  url     = {https://github.com/arthurdjn/pytorch-pointcloud},
}