search

Tencent / Real-SR

Real-World Super-Resolution via Kernel Estimation and Noise Injection
Apache License 2.0
788 stars 118 forks source link

question about clean-up #10

Open fedral opened 4 years ago

fedral commented 4 years ago

Hi, I have a question about clean-up section. realSR "adopt bicubic downsampling on the real image in the source domain to remove noise and make the image sharper".

But bicubic downsampling preserve as much information as possible, including noise and content, but rather than "remove noise". I tested on DIV2K train image 0048.png, which has severe noise. area / bicubic / bilinear downsampled 720*720 images are compared.

image

bicubic downsampled [bottom left] same visual quaility as HR; area downsampled [top right] eliminate most background noise; bilinear downsampled [bottom right] slightly denoise ;

or bicubic downsampling clean-up is just to fit with KernelGAN setting?

fedral commented 3 years ago

problem solved. opencv bicubic resize function using alpha = -0.75, whick tend to overshot, keep noise pattern from HR; while imresize bicubic in matlab using alpha = -0.5, generate denoised images as expected.