Regarding the usage of SimScale training data

[wangxiyang2022]

Hi,

I have one more question regarding the usage of SimScale simulation data.

The paper only uses 134K simulation samples, while the full SimScale dataset contains 236K samples in total. Excluding the 33K samples without camera images filtered in the repo, there should still be around 200K valid simulation data in theory. Could you please explain why only 134K of them were adopted in the paper?

Best regards,

[yihongXU]

Hi,

Indeed, we did not use the full simscale data due to the compute resource limit. We're preparing the ckpt with full simscale as trainset.