无人驾驶CVPR 20252025
BEVFormer: Bird's-Eye-View Perception with Transformers
Zhiqi Li, Wenhai Wang, Hongyang Li, Enze Xie et al.Shanghai AI Lab
摘要
BEVFormer learns bird's-eye-view representations from multi-camera images using spatial and temporal transformers for autonomous driving perception.
BEV perceptiontransformersmulti-cameraspatial-temporalautonomous driving
技术细节
模型骨架
BEV Transformer
编码器
Spatial Cross-Attention, Temporal Self-Attention
解码器
BEV Feature Decoder