AI-Mobility-Research-Lab/centerpoint-nuscenes-roadside-8k

CenterPoint — NuScenes Custom Roadside (8k, rebuilt consecutive val 6401–7200)

This checkpoint was trained with CenterPoint (pillar-based) implemented in OpenPCDet on a custom roadside NuScenes-format dataset.

Model Details

Property	Value
Architecture	CenterPoint (PillarVFE + PointPillarScatter + BaseBEVBackbone + CenterHead)
Classes	car, truck, pedestrian, bicycle
Point cloud range	[−51.2, −51.2, −5.0] → [51.2, 51.2, 3.0] m
Voxel size	0.2 × 0.2 × 8.0 m (pillar)
Training epochs	80
Point features	x, y, z, intensity, timestamp (5-dim)

Files

File	Description
checkpoint_epoch_80.pth	Full model checkpoint (weights + optimizer state)
centerpoint_8k_rebuilt_consecutive_val_6401_7200.yaml	Model + training config
nuscenes_custom_8k_rebuilt_consecutive_val_6401_7200.yaml	Dataset config

Inference Demo

An interactive Gradio demo is available in the companion demo/ folder of the OpenPCDet repository.
Run it with:

conda activate openpcdet_torch25
python demo/gradio_demo.py

Citation

If you use this model, please cite OpenPCDet:

@misc{openpcdet2020,
    title={OpenPCDet: An Open-source Toolbox for 3D Object Detection from Point Clouds},
    author={OpenPCDet Development Team},
    howpublished={\url{https://github.com/open-mmlab/OpenPCDet}},
    year={2020}
}