HDETR / H-PETR-3D

[CVPR2023] This is an official implementation of paper "DETRs with Hybrid Matching".

H-PETR-3D

Main Results

Model	Epoch	mAP	NDS	config	download
PETRv2	24epoch	41.00%	50.30%	config	model
H-PETRv2	24epoch	41.93%	51.23%	config	model
PETRv2	36epoch	41.07%	50.68%	config	model
H-PETRv2	36epoch	42.59%	52.38%	config	model

Preparation

Please refer to PETR for environment and dataset preparation.

Train

tools/dist_train.sh projects/configs/petrv2/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800.py 8 --work-dir work_dirs/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800/

Evaluation

tools/dist_test.sh projects/configs/petrv2/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800.py work_dirs/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800/latest.pth 8 --eval bbox

Modified files compared to vanilla PETRv2

To support Hybrid-branch

• projects/mmdet3d_plugin/models/dense_heads/hybrid_petrv2_head.py
• projects/mmdet3d_plugin/models/dense_heads/__init__.py
• projects/mmdet3d_plugin/models/utils/petr_transformer.py
• projects/configs/petrv2/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800.py
• projects/configs/petrv2/hybrid_petrv2_vovnet_gridmask_p4_800x320_lambda1_group4_t1800_ep36.py

Citation

@article{jia2022detrs,
  title={DETRs with Hybrid Matching},
  author={Jia, Ding and Yuan, Yuhui and He, Haodi and Wu, Xiaopei and Yu, Haojun and Lin, Weihong and Sun, Lei and Zhang, Chao and Hu, Han},
  journal={arXiv preprint arXiv:2207.13080},
  year={2022}
}

@article{liu2022petrv2,
  title={PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images},
  author={Liu, Yingfei and Yan, Junjie and Jia, Fan and Li, Shuailin and Gao, Qi and Wang, Tiancai and Zhang, Xiangyu and Sun, Jian},
  journal={arXiv preprint arXiv:2206.01256},
  year={2022}
}