Skip to content

DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation (ICCV 2023)

Notifications You must be signed in to change notification settings

qcraftai/distill-bev

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation

Zeyu Wang, Dingwen Li, Chenxu Luo, Cihang Xie, Xiaodong Yang
DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation, ICCV 2023

Get Started

Installtion

Please refer to INSTALL to set up the environment and install the dependencies (see more details in Dockerfile).

Data Preparation

Please follow the instructions in DATA.

Training and Evaluation

Please follow the instructions in RUN.

Main Results

BEVDepth and BEVFormer are respectively used in this repo as CNNs and Transformers based students to exemplify the improvement by DistillBEV.

  • Backbone: ResNet-50 pre-trained on ImageNet-1K
Teacher Student mAP NDS Checkpoints
- BEVDepth 36.4 48.4 [Google Drive] [Baidu Cloud]
CenterPoint BEVDepth 39.0 50.6 [Google Drive] [Baidu Cloud]
MVP BEVDepth 40.3 51.0 [Google Drive] [Baidu Cloud]
Teacher Student mAP NDS Checkpoints
- BEVFormer 32.3 43.4 [Google Drive] [Baidu Cloud]
CenterPoint BEVFormer 35.9 46.8 [Google Drive] [Baidu Cloud]
MVP BEVFormer 36.7 47.6 [Google Drive] [Baidu Cloud]
  • Backbone: ResNet-101 pre-trained on ImageNet-1K
Teacher Student mAP NDS Checkpoints
- BEVDepth 40.7 52.2 [Google Drive] [Baidu Cloud]
CenterPoint BEVDepth 43.6 53.6 [Google Drive] [Baidu Cloud]
MVP BEVDepth 45.0 54.6 [Google Drive] [Baidu Cloud]
Teacher Student mAP NDS Checkpoints
- BEVFormer 34.9 46.0 [Google Drive] [Baidu Cloud]
CenterPoint BEVFormer 37.4 48.2 [Google Drive] [Baidu Cloud]
MVP BEVFormer 38.2 49.1 [Google Drive] [Baidu Cloud]

Citation

Please cite the following paper if this repo helps your research:

@InProceedings{Wang_2023_ICCV,
    author    = {Wang, Zeyu and Li, Dingwen and Luo, Chenxu and Xie, Cihang and Yang, Xiaodong},
    title     = {DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation},
    booktitle = {IEEE/CVF International Conference on Computer Vision (ICCV)},
    year      = {2023},
}

Acknowledgement

We thank the authors for the multiple great open-sourced repos, including MMDetection3D, CenterPoint, BEVDet, BEVDepth and BEVFormer.

License

Copyright (C) 2023 QCraft. All rights reserved. Licensed under the CC BY-NC-SA 4.0 (Attribution-NonCommercial-ShareAlike 4.0 International). The code is released for academic research use only. For commercial use, please contact [email protected].