Question: Parameters for Full HD training

[buesma]

Dear authors,

Thank you for presenting this amazing method!
I want to evaluate this method (and also NSFF) on the full resolution images from the Nvidia dataset. As far as I understand it are the configuration files and pre-trained ckpts for the subsampled images with height=288.

I would be very grateful if you could let me know what parameters you used for training to get the maximum performance on the full resolution images
Optionally, do you have an idea how the method behaves if trained on the decreased resolution and then evaluated on the full resolution?

Best
Marcel