Super resolution - Direct comparison of the selected region, magnified image 1:1 output

[Faberman]

Hello.
Thank you for this useful tool, but I can't figure out if it is possible to compare a part of multiple very similar images on the original scale (100%).

For example, I have 8 high resolution output images from 8 super-resolution GANs.
click for original image

They are very similar, yet I'd like to spot the differences. So, I would like to select a region and view that region from all the images at 100% scale.

Just like this - only the selected region, without the original images (because they are huge).

[nachifur]

Thank you for your interest in MulimgViewer. I understand what you want to do. I am considering solving your problem. Can you wait for me for a few minutes? If the picture does not involve privacy, you can send a few test images to my email. liujiawei@sia.cn

[Faberman]

I can provide the real images if you need them, but I think they are a bit heavy for debugging purposes. Just use those lightweight test images I used for my examples: test_images.zip

[nachifur]

Now, I can give you an imperfect solution.

• Select all pictures for comparison, show scale can be set to 0.2, 0.2, output scale can be set to 1,1.
• Then save image, Output mode can be set to 7.
• Find the magnifier_images folder in the save directory, then compare the pictures in the magnifier_images folder and set an appropriate show scale to avoid distortion.

Unfortunately, after these operations, there is no guarantee that magnifier_images is on the original scale (100%).
Your question is very good, I will try to solve it in the next version!

[nachifur]

At present, for the output of the magnifier image, I forcibly resize it, which is the reason why magnifier images on the original scale (100%) cannot be obtained. This is unreasonable and I will modify it. Thank you for pointing this out!

MulimgViewer/utils.py

Line 823 in 808f043

def magnifier_preprocessing(self, img, img_mode=0):

[Faberman]

Thank you! Yes, unfortunately, at the moment the magnifier_images are getting resized. Upscaling with a subsequent downscaling is not that good for a pixel-perfect comparison.

By the way, here is an updated pack of test images - easier to test multi-region options.
test_images_multi.zip

[nachifur]

Yes, this is very challenging, I will think of a perfect way to solve it. Thank you!

[nachifur]

If there are any improvements or bugs, please call me, and I will try my best to improve MulimgViewer.
You can recommend MulimgViewer to those around you who need it. My personal propaganda power is seriously insufficient. Thanks you!

[Faberman]

Your tool is really unique and I will definitely recommend it to anyone. Now there is only one option for an image comparison - FastStone Image Viewer. But it's able to compare only 4 images. Probably sufficient for an average user, but it's not enough for those who deal with GANs.

[nachifur]

I am not a native speaker, but I hope MulimgViewer has a fluent and readable English version of the readme. I don't know if you are willing to maintain the English version of the readme. If you want to do this, I sincerely appreciate you.

[Faberman]

I can't edit wiki directly and GitHub doesn't support pull requests for the wiki, but I've edited some confusing parts, so here is an updated Readme: https://gist.github.com/Faberman/e7533e7bab499ebddaca7f27f86609a9

[nachifur]

Thank you for your polish of the English readme. In order to have greater rights to contribute to MulimgViewer, I have invited you to be a collaborator.

[nachifur]

Sorry, due to my busy schedule recently, I may not be able to update the functions in time in the recent period. You now have the right to commit directly, if you want, you can try to create a new branch to complete this function. Thank you for any contributions.
Besides, I also hope that we will become in-depth collaborators, both in this project or an academic article.

[nachifur]

I have implemented this feature, you need to use python script to run.

Process Flow:

1. Show original=True, input mode=Sequential, Num per img=-1, 🔍️Scale=1,1. Use multi-box zoom mode. Then refresh GUI by ctrl+r.
2. Show original=False, Then refresh GUI by ctrl+r.
3. If you only have one box, by setting Show box in 🔍️=False, you can eliminate the box of the enlarged image.

In addition, you can set 🔍️Scale=1,1 to achieve saving on the original scale (100%) for the output of the magnifier image.

test file:
https://github.com/nachifur/MulimgViewer/files/6742637/test_images_multi.zip

pip installation:

pip install wxpython pillow pytest-shutil

run:

python3 MulimgViewer.py

Has your problem been solved?
Finally, sincerely thank you for your contribution to MulimgViewer.