search

autonomousvision / sledge

[ECCV'24] SLEDGE: Synthesizing Driving Environments with Generative Models and Rule-Based Traffic
Apache License 2.0
146 stars 6 forks source link

Why is nuplan data required. #3

Closed deo-abhijit closed 3 weeks ago

deo-abhijit commented 3 weeks ago

Hi! Thanks for sharing the code!

I am trying to understand the paper. so in the abstract of the paper, it says, 

Compared to nuPlan, SLEDGE requires 500× less storage to set up (<4 GB), making it a more accessible
option and helping with democratizing future research in this field.

But in the installation, there is a step to download the nuplan data. Can i use sledge without this data?

Thanks.

deo-abhijit commented 3 weeks ago

is it for training the autoencoder?

DanielDauner commented 3 weeks ago

Hi @deo-abhijit

Yes, the complete nuPlan dataset is only required if you want to train the autoencoder from scratch and with input features different from the default. We provide a download of the default autoencoder features with the ~500k samples in download_cache.sh.

The <4 GB refers to the SLEDGE simulation with a trained diffusion model and the nuPlan maps. We will make this more clear in the updated docs and are working on the checkpoint release.

Best, Daniel

deo-abhijit commented 3 weeks ago

Understood! Thanks