result looks strange on text area about real-esrgan

[xinntao]

C2-Matching is a reference-based SR framework.
You can perform C2-matching (with super-resolution) and then down-sample it to the desired size. Or you can retrain your own model without SR>

[tachikoma777]

the reference image is not the gt, but another image with sightly parallax

[xinntao]

@tachikoma777

I will improve the Real-ESRGAN by fine-tuning on images with text.
The preliminary results (on English text) are promising, and I will extent it.

For reference-based SR, you may check: https://github.com/yumingj/C2-Matching

[tachikoma777]

@xinntao
thanks! I have been testing on C2-Matching, but I am new to this, do you know how to do only restoration not super resolution on C2-Matching? I ask this since basicSR is also your great work, thanks!

[pyl62112991]

@tachikoma777

I will improve the Real-ESRGAN by fine-tuning on images with text. The preliminary results (on English text) are promising, and I will extent it.

For reference-based SR, you may check: https://github.com/yumingj/C2-Matching

Hello, I would like to ask, this EDRGAN and C2-Matching , two papers are respectively designed for what kind of super-resolution task, whether the image with text more fit for C2-Matching?