autonomousvision / sledge

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

Map layers #7

Closed hamid54 closed 2 weeks ago

hamid54 commented 2 weeks ago

Hi,

As for input/output channels, Sledge has distinct channels for lanes (baseline, green, and red). However, there is no information available regarding the roads, which consist of a set of lanes and drivable areas. I am curious whether this aspect is specifically addressed or currently disregarded.

Thank you! Hamid

DanielDauner commented 2 weeks ago

Hi @hamid54

Thanks for this question!

We do not model the surface polygons and of the maps in SLEDGE. We assume a constant lane width of 3m in the Lane & Agent simulation (see here), which is similarly done in DriveSceneGen. In the Lane -> Agent mode, we just used the nuPlan map.

I hope that helps! Daniel