VAD End-to-End Inference

This repo is the implementation of VAD test client for HUGSIM benchmark

The implementation is based on:

VAD: Vectorized Scene Representation for Efficient Autonomous Driving

Bo Jiang¹*, Shaoyu Chen¹*, Qing Xu², Bencheng Liao¹, Jiajie Chen², Helong Zhou², Qian Zhang², Wenyu Liu¹, Chang Huang², Xinggang Wang^1,†

¹ Huazhong University of Science and Technology, ² Horizon Robotics

*: equal contribution, ^†: corresponding author.

arXiv Paper, ICCV 2023

Installation

Please refer to UniAD and VAD installation instructions. In practice, UniAD and VAD can share the same conda environment.

Please change ${VAD_PATH} in tools/e2e.sh as the path on your machine.

Launch Client

Manually Launch

zsh ./tools/e2e.sh ${CUDA_ID} ${output_dir}

Auto Lauch

Client can be auto lauched by the HUGSIM closed-loop script.