Can you provide some details about the training dataset?

[PeeBar]

Hi! Sorry to interrupt you.
I want to retrain your model, and I find that you crop 40960 image patches of size 256*256 from the PASCAL VOC dataset as your training set. But I can't get the similair number of image patches from the PASCAL VOC dataset, do you use both trainval and test folders of PASCAL VOC dataset? And can you provide some more details?
Thank you very much!