search

wayveai / mile

PyTorch code for the paper "Model-Based Imitation Learning for Urban Driving".
MIT License
330 stars 31 forks source link

How to increase speed when performing closed-loop validation at Carla #26

Closed KevinDu1 closed 10 months ago

KevinDu1 commented 10 months ago

Seeing that the code sets the Carla's FPS to 25Hz causes it to take a day to run one round of validation when running evaluate.py, how can I set it up so that the model can be validated faster?

anthonyhu commented 10 months ago

According to the CARLA developers, the minimum FPS you can set is 10Hz. For faster evaluation, I set the constant CARLA_FPS=10 here https://github.com/wayveai/mile/blob/main/mile/constants.py#L4.

If you are evaluating single-frame models, it should work with a pretrained model. But if you're using temporal models, you have to make sure that the model was trained on the correct temporal stride.

KevinDu1 commented 10 months ago

Thanks！