Bench2DriveZoo

Introduction

• This repo contains the training, open-loop evaluation, and closed-loop evaluation code for BEVFormer, UniAD , VAD in Bench2Drive.
• We merge multiple dependencies of UniAD and VAD including mmcv, mmseg, mmdet, and mmdet3d (v0.17.1) into a single library. As a result, it could support latest pytorch and advanced frameworks like deepspeed for acceleration.
• Use "git checkout tcp/admlp" to obtain their corresponding training and evaluation code.

Getting Started

• Installation
• Prepare Dataset
• Train and Open-Loop Eval
• Closed-Loop Eval in CARLA
• Convert Codes from Nuscenes to Bench2Drive

Results and Pre-trained Models

UniAD and VAD

BEVFormer

Related Resources

• Bench2Drive
• BEVFormer
• UniAD
• VAD