octformer-base.scannet20.octree-nn

Model card for octformer-base.scannet20.octree-nn

An OctFormer point cloud semantic segmentation model (octree-based windowed transformer). Trained on ScanNet (20 classes).

Model Details

• Model Type: Point cloud semantic segmentation
• Model Stats:
  • Params (M): 44.0
  • Input channels: 10
  • Classes: 21
  • Features: 168
• Dataset: ScanNet (20 classes)
• Metrics: mIoU 74.78, OA 90.9
• Paper: OctFormer: Octree-based Transformers for 3D Point Clouds
• Converted from: octree-nn/octformer (MIT)
• Library: torch-pointcloud

Install

pip install torch-pointcloud

This checkpoint also needs ocnn and dwconv, which need a build matching your torch and CUDA: see the installation guide.

Usage

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

model, info = tp.create_model(
    "octformer-base.scannet20.octree-nn",
    pretrained=True,
    return_info=True,
)
model = model.cuda().eval()  # GPU-only kernels

# synthetic sample with the keys a dataset provides
num_points = 8192
sample = {
    "pos": torch.randn(num_points, 3),
    "color": torch.rand(num_points, 3) * 255,
    "normal": torch.randn(num_points, 3),
    "segment": torch.zeros(num_points, dtype=torch.long),
}
data = info["transform"](sample)
data = collate([data])
data = {key: value.cuda() for key, value in data.items()}

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

Feature extraction

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

model.reset_classifier(num_classes=0)
with torch.no_grad():
    features = model(
        data.get("x"),
        data["octree"],
        data["octree"].depth,
        data["pos"],
        data["batch"],
    )  # (N, 168)

Citation

@article{wang2023octformer,
  title   = {OctFormer: Octree-based Transformers for 3D Point Clouds},
  author  = {Peng-Shuai Wang},
  journal = {ACM Transactions on Graphics},
  volume  = {42},
  number  = {4},
  year    = {2023}
}

@inproceedings{dai2017scannet,
  title   = {ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes},
  author  = {Angela Dai and Angel X. Chang and Manolis Savva and Maciej Halber and Thomas Funkhouser and Matthias Nießner},
  booktitle = {CVPR},
  year    = {2017}
}

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