BEVFormer: Learning Bird’s-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers

Zhiqi Li^1,2∗ , Wenhai Wang^2∗ , Hongyang Li^2∗ , Enze Xie³ , Chonghao Sima² , 

Tong Lu¹ , Yu Qiao² , Jifeng Dai^2✉

 ¹Nanjing University  ²Shanghai AI Laboratory  ³The University of Hong Kong

Figure 1: We propose BEVFormer, a paradigm for autonomous driving that applies both Transformer and Temporal structure to generate bird’s-eye-view (BEV) features from multi-camera inputs. BEVFormer leverages queries to lookup spatial/temporal space and aggregate spatiotemporal information correspondingly, hence benefiting stronger representations for perception tasks.

Abstract

3D visual perception tasks, including 3D detection and map segmentation based on multi-camera images, are essential for autonomous driving systems. In this work, we present a new framework termed BEVFormer, which learns unified BEV representations with spatiotemporal transformers to support multiple autonomous driving perception tasks. In a nutshell, BEVFormer exploits both spatial and temporal information by interacting with spatial and temporal space through predefined grid-shaped BEV queries. To aggregate spatial information, we design a spatial cross-attention that each BEV query extracts the spatial features from the regions of interest across camera views. For temporal information, we propose a temporal self-attention to recurrently fuse the history BEV information. Our approach achieves the new state-of-the-art 56.9% in terms of NDS metric on the nuScenes test set, which is 9.0 points higher than previous best arts and on par with the performance of LiDAR-based baselines. We further show that BEVFormer remarkably improves the accuracy of velocity estimation and recall of objects under low visibility conditions. The code will be released at https://github.com/zhiqi-li/BEVFormer