Open TZYSJTU opened 3 years ago
@csjcai @TZYSJTU Hi, I encounter the same problem. How to reproduce the Bicubic results reported in the paper?
I use the modified test code to evaluate the Version 1 dataset.
clear; clc; close all; warning off;
addpath(genpath('./.'));
count = 1;
folder2 = ['./Test/2'];
filepathsHR = [];
filepathsHR = cat(1,filepathsHR, dir(fullfile(folder2, 'HR', '*.png')));
filepathsLR = [];
filepathsLR = cat(1,filepathsLR, dir(fullfile(folder2, 'LR', '*.png')));
for i = 1 : 1: length(filepathsHR)
disp(filepathsHR(i).name)
disp(filepathsLR(i).name)
HR = imread(fullfile(folder2,'HR',filepathsHR(i).name));
HR_Ycbcr = im2double(rgb2ycbcr(HR));
HR_Y = im2single(HR_Ycbcr(:, :, 1));
LR = imread(fullfile(folder2,'LR',filepathsLR(i).name));
LR_Ycbcr = im2double(rgb2ycbcr(LR));
LR_Y = im2single(LR_Ycbcr(:, :, 1));
[PSNRCur, SSIMCur] = Cal_PSNRSSIM(im2uint8(HR_Y),im2uint8(LR_Y), 0, 0);
PSNRs(count) = PSNRCur;
SSIMs(count) = SSIMCur;
count = count + 1;
end
length(PSNRs)
mean(PSNRs)
mean(SSIMs)
I have reorganized Test
folder as follow:
./Test/2/HR/Canon_XXX_HR.png
./Test/2/LR/Canon_XXX_LR2.png
./Test/3/HR/Canon_XXX_HR.png
./Test/3/LR/Canon_XXX_LR3.png
./Test/4/HR/Canon_XXX_HR.png
./Test/4/LR/Canon_XXX_LR4.png
Scale | Reported | Reproduced | Diff |
---|---|---|---|
x2 | 32.61 | 32.12 | -0.49 |
x3 | 29.34 | 28.97 | -0.37 |
x4 | 27.99 | 27.59 | -0.40 |
Thanks in advance!
Why do HR and LR have the same resolution? How to calculate Bicubic baseline since HR and LR are the same size? Thanks!