Each dataset can be created by using the line below where source
is the directory containing images
which we want to train and validate. The optional argument --recursive
will recursively search for
all images in source
. scale
denotes the the values of factors used to downsample LR images.
The interpolation method used to downsample images is bicubic, by default.
$ python make_dataset.py --src source --dst dest --scale 2 3 4 --val_r 0.1 --recursive