Training WDSR on custom input/output sizes

[michiyosony]

Thanks for this OS implementation! The results look pretty amazing. If I wanted to train a model for a custom super resolution scale, say 3.4x, would I be able to the following?

1. Create a dataset of downscaled images and 3.4x larger HR images structured as you describe here
2. Modify checks that scale be one of [2, 3, 4] in places like data.py and train.py
3. Run train.py with the argument --scale 3.4