Gaussian blur not applied to input images

Edwardlzy / SRCNN

Tensorflow implementation of single image super-resolution using a Convolutional Neural Network

MIT License

47 stars 17 forks source link

Open sdv4 opened 4 years ago

sdv4 commented 4 years ago

I noticed that you missed the step of bluring the images between down-sampling and up-sampling. From prepare_train.py:

im_temp = scipy.misc.imresize(im_label, 1/scale, interp='bicubic')
im_input = scipy.misc.imresize(im_temp, size, interp='bicubic')

This still created distorted images, but I think it makes the task easier when the blur isn't applied.