opendilab/dos

[CVPR 2023] ReasonNet: End-to-End Driving with Temporal and Global Reasoning

dos's Issues

数据收集

作者您好，我有个问题想请问一下，像DOS_0X_town05.json 这种json文件和对应的xml文件是运行哪个脚本生成的？

occlusion scenes

Hello, I used DOS_ 01_ Town05.xml to run some routes experiment and I couldn't see any occlusion scenes in the window. The agent toke auto_pilot. py. May I ask why?

record the inputs during the evaluation

Hi, I wondered how I could record sensor inputs in the evaluation in case I want to investigate the failure cases.

DOS_X_town05.json

Hello, may I ask what content is saved in this file? For example, what does “available_event_configurations” mean, and what does each “transform” mean？Thank you！

内存缓存机制

Hi，我对这段描述不是很理解，主要有3点：

1. the last memory frame is full，这句话指的什么意思？是代表内存满了吗？可是下文也有提到帧数满废弃旧内存的情况。
2. initialize one new frame with the zero vector and set it as the last frame to store new features. 这句话的工程性逻辑我没太理解，请问可以再解释一下吗，麻烦了
3. When the number of frames reachesthe limit Tl, the obsolete memory will be removed. 请问这里的obsolete memory是怎么定义的？就是直接把long-term 内存的原先的都清理掉吗？

模型代码公开

请问后续会公开模型训练和验证的代码吗？我对您的工作很感兴趣，向您学习，感谢

code about ReasonNet

作者您好，请问ReasonNet的代码以及环境配置可以开源分享吗，殷切盼复呢，我的邮箱[email protected]

训练过程中的缓存机制

非常感谢博主的论文，深受启发，有一些问题想咨询一下博主
1）论文中的 Interaction modeling 是图神经网络吗？能否进一步详细介绍
2）训练过程中的时间步长是多少呢？假设步长为4，是每一帧都有一个输出和loss作梯度下降，还是只采用最后一帧的输出作梯度下降？
感谢博主

carla版本

请问使用的carla是哪个版本？

DOS_ 02_ Town05.xml

Hello, I am using DOS_ 02_ Town05.xml ran a route experiment and I couldn't see any occlusion scenes in the window. My agent called auto_ Pilot. py. May I ask why? In addition, the car has been parked on this interface for a long time, and Carla has not been stuck. I don't know why？

Question about an error: CHALLENGE_TRACK_CODENAME=SENSORS

Hi, thanks for your great work!

I have a question while trying to run dos benchmark, I always have

The sensor's configuration used is invalid:
> Illegal sensor extrinsics used for Track [Track.SENSORS]!

error.

I don't know what is wrong with my setting. Do you have any idea, why this is happening and what is the solution?
Thanks a lot in advance!!

The error:

My setting

训练数据

你好，我对训练时的数据处理有疑惑，可以帮忙解答一下吗？
按照论文的思路，是要输入连续帧的图像进行训练吗？因为同一个场景连续帧数据的存储记忆训练才有意义？如果是的话，要怎么样输入训练数据呢？因为数据包括了多个城镇和多条路线，存储的连续帧信息可能不是同一场景