Model card for groupfree3d-l6-o256.sunrgbd.ze-liu

A Group-Free 3D 3D object detection model (transformer detector over sampled object candidates). Trained on SUN RGB-D.

Model Details

• Model Type: 3D object detection
• Model Stats:
  • Params (M): 14.4
  • Classes: 10
  • Features: 288
• Dataset: SUN RGB-D
• Metrics: mAP@0.25 64.1, mAP@0.5 45.65
• Paper: Group-Free 3D Object Detection via Transformers
• Converted from: zeliu98/Group-Free-3D (MIT)
• Library: torch-pointcloud

Trained and scored with the authors' SUN RGB-D protocol, which sizes each ground-truth box by the axis-aligned extent of the oriented box and keeps its heading. These scores do not compare with detectors scored against the oriented boxes themselves.

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(
    "groupfree3d-l6-o256.sunrgbd.ze-liu",
    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),
    "color": torch.rand(num_points, 3) * 255,
}
data = info["transform"](sample)
data = collate([data])

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

Feature extraction

with torch.no_grad():
    features = model.forward_features(data.get("x"), data["pos"], data["batch"])  # 288 channels

Citation

@inproceedings{liu2021groupfree,
  title   = {Group-Free 3D Object Detection via Transformers},
  author  = {Ze Liu and Zheng Zhang and Yue Cao and Han Hu and Xin Tong},
  booktitle = {ICCV},
  year    = {2021}
}

@inproceedings{song2015sunrgbd,
  title     = {{SUN RGB-D}: A {RGB-D} Scene Understanding Benchmark Suite},
  author    = {Song, Shuran and Lichtenberg, Samuel P. and Xiao, Jianxiong},
  booktitle = {CVPR},
  year      = {2015}
}

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