Hierarchical and Decoupled BEV Perception Learning Framework for Autonomous Driving

TL;DR

This paper introduces a hierarchical BEV perception framework with modular design and multi-module learning, significantly improving autonomous driving perception systems' development efficiency and performance.

Contribution

It proposes a novel hierarchical perception paradigm with a modular library, a user-friendly interface, and a multi-module training approach for better efficiency and accuracy.

Findings

Significant performance improvement on Nuscenes dataset

Enhanced development efficiency with modular design

Effective utilization of large-scale datasets through Pretrain-Finetune

Abstract

Perception is essential for autonomous driving system. Recent approaches based on Bird's-eye-view (BEV) and deep learning have made significant progress. However, there exists challenging issues including lengthy development cycles, poor reusability, and complex sensor setups in perception algorithm development process. To tackle the above challenges, this paper proposes a novel hierarchical BEV perception paradigm, aiming to provide a library of fundamental perception modules and user-friendly graphical interface, enabling swift construction of customized models. We conduct the Pretrain-Finetune strategy to effectively utilize large scale public datasets and streamline development processes. Moreover, we present a Multi-Module Learning (MML) approach, enhancing performance through synergistic and iterative training of multiple models. Extensive experimental results on the Nuscenes dataset demonstrate that our approach renders significant improvement over the traditional training scheme.