about UMD dataset

Hello, author. I’d like to ask: in the evaluation on the UMD dataset, is the training strategy consistent with the one used for the other dataset? Also, when selecting images for training, are they randomly chosen as well?
