We implement Dynamic Voxelization proposed in and provide its results and models on KITTI dataset.
@article{zhou2019endtoend,
title={End-to-End Multi-View Fusion for 3D Object Detection in LiDAR Point Clouds},
author={Yin Zhou and Pei Sun and Yu Zhang and Dragomir Anguelov and Jiyang Gao and Tom Ouyang and James Guo and Jiquan Ngiam and Vijay Vasudevan},
year={2019},
eprint={1910.06528},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
| Model | Class | Lr schd | Mem (GB) | Inf time (fps) | mAP | Download |
|---|---|---|---|---|---|---|
| SECOND | Car | cyclic 80e | 5.5 | 78.83 | model | log | |
| SECOND | 3 Class | cosine 80e | 5.5 | 65.10 | model | log | |
| PointPillars | Car | cyclic 80e | 4.7 | 77.76 | model | log |