training-testing split for columbia dataset

[zechengtang]

Thanks for your awesome work! I am experiencing some issues regarding the training-testing splits on the four public datasets. Are you using the widely accepted protocol in RGB-N (Since ManTra-Net uses the entire dataset for testing, I presume you are referring to RGB-N as listed in ManTra-Net)? If so, what is the training-testing split for the Columbia dataset, considering that in RGB-N the model is trained on a synthetic dataset and tested on the entire dataset?