collector-m / autoalignv2

[ECCV2022] AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection

AutoAlignV2

Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection

Zehui Chen, Zhenyu Li, Shiquan Zhang, Liangji Fang, Qinhong Jiang, Feng Zhao.

[Paper]

Performance

nuScenes Val set

Model	config	mAP	NDS
Baseline 1/8 subset		-	-
AutoAlignV2 1/8 subset		58.5	63.2

nuScenes Test Leaderboard

Model	mAP	NDS
AutoAlign	65.8	70.9
AutoAlignV2	68.4	72.4

Note

The code is released without rechecking. We will clean up and recheck the code recently.

Get Started

Prepare Dataset

python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes --version v1.0

Train Model

./tools/dist_train.sh aav2_cfg/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts.py 8

Test Model

./tools/dist_test.sh aav2_cfg/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts.py work_dirs/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts/epoch_20.pth 8 --eval bbox

Pretrained Model Weights

Model	Weights
Pretrained YOLOX-640	https://drive.google.com/file/d/1DuA16ocF8Dmfl0nC3NNSKG_Yqq-aXuLH/view?usp=sharing
Pretrained YOLOX-800	https://drive.google.com/file/d/1YPSjG38qXrV8gYlq5FuhQCjfRIUvhaju/view?usp=sharing
AutoAlignV2 1/8 subset	https://drive.google.com/file/d/1F_B4bq5Y6QdkUHQbqeLaEGl-IZHHe1WB/view?usp=sharing

Citation

If you find our work useful for your research, please consider citing the paper

@article{chen2022autoalignv2,
  title={AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection},
  author={Chen, Zehui and Li, Zhenyu and Zhang, Shiquan and Fang, Liangji and Jiang, Qinhong and Zhao, Feng},
  journal={ECCV},
  year={2022}
}