search

opendilab / InterFuser

[CoRL 2022] InterFuser: Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer
Apache License 2.0
514 stars 42 forks source link

Lightweight docker #50

Open Mariusmarten opened 1 year ago

Mariusmarten commented 1 year ago

Hi, thank you for making the code available!

I would like to run the trained InterFuser model on a small set of pre-collected CARLA data. Can I download such data from somewhere without collecting them myself? How would you go about setting up such a lightweight setting?

deepcs233 commented 1 year ago

Hi! We have a plan to release a small set (about 10-20G) of pre-collected CARLA data, but it may take some time. I think it's hard to get a acceptable performance with such small data.

For a lightweight setting, i recommend you use a smaller network and more training epochs.

Mariusmarten commented 1 year ago

Thank you for the information. Do I understand correctly that with those weights: http://43.159.60.142/s/p2CN I will not need to train the model at all? I am mainly interested in the driving capabilities of the model (to run it on my own data).

deepcs233 commented 1 year ago

Hi! If you just want to visualize/test the driving capabilities on your data for convenience, it's ok to use this pretrain weights. But it you want to eval the performance in some benchmarks, you need to re-collect the correspondind dataset and re-train it.

Mariusmarten commented 11 months ago

Hi! Thank you for the response. How good are the pre-trained weights? Are they close to the actual model performance?