Model card for litept-small.scannet20.prs-eth

A LitePT point cloud semantic segmentation model (convolutions in the early stages, attention with 3D rotary embedding in the late ones). Trained on ScanNet (20 classes).

Model Details

• Model Type: Point cloud semantic segmentation
• Model Stats:
  • Params (M): 12.7
  • Input channels: 6
  • Classes: 20
  • Features: 72
• Dataset: ScanNet (20 classes)
• Metrics: mIoU 76.54, OA 91.42
• Paper: LitePT: Lighter Yet Stronger Point Transformer
• Converted from: prs-eth/LitePT (MIT)
• Library: torch-pointcloud

Install

pip install torch-pointcloud

This checkpoint also needs spconv and flash-attn, 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(
    "litept-small.scannet20.prs-eth",
    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),
    "instance": 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["pos_grid"], data["batch"])

Feature extraction

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

model.reset_classifier(num_classes=0)
with torch.no_grad():
    features = model(data.get("x"), data["pos_grid"], data["batch"])  # (N, 72)

Citation

@inproceedings{yue2026litept,
  title   = {LitePT: Lighter Yet Stronger Point Transformer},
  author  = {Yuanwen Yue and Damien Robert and Jianyuan Wang and Sunghwan Hong and Jan Dirk Wegner and Christian Rupprecht and Konrad Schindler},
  booktitle = {CVPR},
  year    = {2026}
}

@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},
}