Training with your own lr data

[pihchikk]

Hello, thanks a lot for your work!

I currently work on grayscale CT-images super-resolution. As I went through both steps of the training process using pre-trained stable diffusion the results turned out to be far from desired. I have a dataset of paired lq and hq images that I want to use to train stage1 and stage2 models. Is it possible to do this in the current design with minor additions to the code? Will the use of a SwinIR model that was pretrained with my dataset, improve the quality of generation?