Multi-Camera Unified Pre-training via 3D Scene Reconstruction

(for DETR3D, BEVFormer, BEVDet, BEVDepth and Semantic Occupancy Prediction)

Repository for our RAL 2024 paper Paper arXiv

Abstract

Multi-camera 3D perception has emerged as a prominent research field in autonomous driving, offering a viable and cost-effective alternative to LiDAR-based solutions. The existing multi-camera algorithms primarily rely on monocular 2D pre-training. However, the monocular 2D pre-training overlooks the spatial and temporal correlations among the multi-camera system. To address this limitation, we propose the first multi-camera unified pre-training framework, called UniScene, which involves initially reconstructing the 3D scene as the foundational stage and subsequently fine-tuning the model on downstream tasks. Specifically, we employ Occupancy as the general representation for the 3D scene, enabling the model to grasp geometric priors of the surrounding world through pre-training.

Methods

Getting Started

Model Zoo

Backbone	Method	Pre-training	Lr Schd	NDS	mAP	Config
R101-DCN	BEVFormer	ImageNet	24ep	47.7	37.7	config/[model]
R101-DCN	BEVFormer	ImageNet + UniScene	24ep	50.0	39.7	config/model
R101-DCN	BEVFormer	FCOS3D	24ep	51.7	41.6	config/model
R101-DCN	BEVFormer	FCOS3D + UniScene	24ep	53.4	43.8	config/pre-trained model/fine-tune-model

Bibtex

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{UniScene,
  title={Multi-Camera Unified Pre-Training Via 3D Scene Reconstruction},
  author={Min, Chen and Xiao, Liang and Zhao, Dawei and Nie, Yiming and Dai, Bin},
  journal={IEEE Robotics and Automation Letters},
  year={2024},
  publisher={IEEE}
}

Acknowledgement

Many thanks to these excellent open source projects: