Model card for pv-rcnn.kitti.mmdet3d

A PV-RCNN 3D object detection model (two-stage detector over voxel features and point keypoints). Trained on KITTI.

Model Details

• Model Type: 3D object detection
• Model Stats:
  • Params (M): 13.1
  • Input channels: 4
  • Classes: 3
  • Features: 512
• Dataset: KITTI
• Metrics: mAP 72.16, AP/Car 83.54, AP/Pedestrian 59.73, AP/Cyclist 73.2
• Paper: PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection
• Converted from: open-mmlab/mmdetection3d (Apache-2.0)
• Library: torch-pointcloud

Install

pip install torch-pointcloud

This checkpoint also needs spconv and pyg-lib, 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(
    "pv-rcnn.kitti.mmdet3d",
    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),
    "intensity": torch.rand(num_points, 1),
}
data = info["transform"](sample)
data = collate([data], cat_keys=["pos_voxel"])
data = {key: value.cuda() for key, value in data.items()}

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

Feature extraction

with torch.no_grad():
    features = model.forward_features(
        data["voxel"],
        data["pos_voxel"],
        data["voxel_num_points"],
        data["batch_pos_voxel"],
    )  # 512 channels

Citation

@inproceedings{shi2020pvrcnn,
  title   = {PV-RCNN: Point-Voxel Feature Set Abstraction for {3D} Object Detection},
  author  = {Shaoshuai Shi and Chaoxu Guo and Li Jiang and Zhe Wang and Jianping Shi and Xiaogang Wang and Hongsheng Li},
  booktitle = {CVPR},
  year    = {2020}
}

@inproceedings{geiger2012kitti,
  title     = {Are we ready for Autonomous Driving? The {KITTI} Vision Benchmark Suite},
  author    = {Geiger, Andreas and Lenz, Philip and Urtasun, Raquel},
  booktitle = {CVPR},
  year      = {2012}
}

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