hustvl / VAD

[ICCV 2023] VAD: Vectorized Scene Representation for Efficient Autonomous Driving
https://arxiv.org/abs/2303.12077
Apache License 2.0
714 stars 78 forks source link

Questions on the scene encoder in VADv2 #48

Open zhanghui75 opened 8 months ago

zhanghui75 commented 8 months ago

I am wondering how to represent and get those tokens (map token, agent token, traffic element token) in VADv2.

  1. for map tokens and agent tokens, do you use the similar methods as VAD?
  2. for traffic elements, how you detect them? and do you extract and use the relation between traffic lights and lanes (e.g. straight-going light corresponds to the straight-going lane)?

Congratulations on your contineous exellent work, and hope to get your response.

joonjune88 commented 8 months ago

Hello. could you please answer the question above? I also really would like to know what parts of VAD version1 are re-used in VAD version2. Thank you in advance.