search

Boyan-Lenin-Xu / GCResNet

PyTorch implementation of Graph Convolutional Networks in Feature Space for Image Deblurring and Super-resolution, IJCNN 2021.
12 stars 0 forks source link

Little question #2

Open w-z-hub opened 2 years ago

w-z-hub commented 2 years ago

Hello, this is an excellent job. When I was reading your paper, I didn’t quite understand the setting of degree in the article. Did you change the setting in WattsStrogatz.m and adjacency.m to make the graph you need. Another question is whether the three data of full.mat, fullSR.mat, and sr64.mat correspond to the cases where the number of channels is 96, 128, and 64. I'm looking forward to your answer.Thank you.

Boyan-Lenin-Xu commented 2 years ago

Thanks for your attention. Yes, the structure of graph can be changed by changing WattsStrogatz.m and adjacency.m. deblur_2_full.mat is for 128 channels. fullSR and sr64 is for super-resolution. For other settings in deblurring, you may need to change the second line in adjacency.m and save FULL (which is calculated in the 7th line in adjacency.m) as a matrix.

w-z-hub commented 2 years ago

Hello, there is a small question about the sr64.mat data set. Have you made 800 different graphs with matlab to form the sr64.mat file?

Boyan-Lenin-Xu commented 2 years ago

Hi. 800 images are the training dataset of super-resolution. The graph networks (which are saved in .mat files) are produced by adjacency.m.