Open jeonghomin opened 1 year ago
I wonder if i train the dataset that is not paired ( different pairs ) , matching does not work well right? and also, while training , does matching ( similarity scores) is learned also?
Hi, did you provide reference images who's content is similar with that of the input? If not, the reference image may not improve the final results.
I have a question for the testset.
When I train on the custom (paired dataset) and inference on the test set, it keeps making a strange issues.
Such as, when I test with the reference images of high resolution , the psnr and ssim shows same as giving reference image of low resolution. Even though when I give numpy zero array, the results are same.. can you answer why this occurs?