Question about Runtime

[ssuralcmu]

Hi,

Thanks for your great work in planning for autonomous driving. The real-time capability is very impressive. The paper reports a inference speed of 8.2 FPS on NuScenes, and 45 FPS on NAVSIM. It also mentions that both of these speeds are on a single 4090 GPU. Could you confirm if this is the expected inference speed? Also, could you explain why there is such a large gap in FPS depending on the dataset? I might be misunderstanding some aspects of the paper.