nyud_layer: how to reduce data size

[gio]

Hi all, I've failed to train fcn32s on nyud dataset because I don't have enough memory on my gpu, and so the training end with "cuda out of memory error". I want to reduce the size of the image, depth, hha data. Does anyone know how to modify the nyud_layer for the loading of the dataset, in order to make this resize step?