search

ETH-PBL / TC-Driver

The Unlicense
17 stars 4 forks source link

how to record a video during training #1

Open BIT-KaiYu opened 3 months ago

BIT-KaiYu commented 3 months ago

how to record a video during training

niciBume commented 3 months ago

Can you be more specific?

BIT-KaiYu commented 3 months ago

Thank you. When I ran your code, I discovered the videos in the TC-Driver/TC_Driver/train/recording/ directory and was very interested. When I ran train.py, I set record_video to True and modified the relevant code. The final generated video has no pictures.

niciBume commented 3 months ago

This is usually a problem with mismatching gym versions. Are your dependencies matching these versions? https://github.com/ETH-PBL/TC-Driver/blob/main/Gym/requirements.txt

BIT-KaiYu commented 3 months ago

The environment is exactly the same. Could this be related to the fact that I commented out the code about self.visualize in the TC-Driver/Gym/gym/f110_gym/envs/f110_rl_env.py. This was the only change I made, as it resulted in the error: AttributeError: 'RLF110Env' object has no attribute 'create_obs_dict'

niciBume commented 3 months ago

It sounds relevant to visualization. We are working on a new version of the TC-Driver gym, which should be fixing this issue in the meanwhile.

BIT-KaiYu commented 3 months ago

Thank you, I will continue to pay attention.

eghignone commented 3 months ago

@BIT-KaiYu I have added a missing function and fixed an installation requirement that was worng. This should fix the video recording (I tested train.py and obtained such a recording).

While this is not the new version we are working on that @niciBume mentioned, this should fix at least the recording problem

BIT-KaiYu commented 2 months ago

I'm sorry to tell you that the problem still exists. Maybe there is a problem with my environment requirements.txt. Thank you anyway.

eghignone commented 2 months ago

Thank you for providing us with the requirements.txt file, I imagine that is the dump of your currently installed python packages. From it, it looks like most of the packages are the same version as those in the environment I tested the code, aside for the tf which is in my case tf==1.13.2, but I doubt this might be the source of the problem.

Could you provide us with more information? Specifically