Could I train realesrgan as a document image denoising model?

[Yikai-Liao]

Recently, I noticed that there is a great document denoising dataset (actually a augmentation pipline) https://github.com/sparkfish/shabby-pages

I want to train a document denoising model, and I wonder if realesrgan would be a good choice for backbone architecture(also the loss function).

[bjxpen]

I'm interested in this idea too as I was reading scanned PDFs, and they are a pain to stare at.

[Yikai-Liao]

I'm interested in this idea too as I was reading scanned PDFs, and they are a pain to stare at.

Currently, there are some traditional methods to deal with this problem. But you will need to adjust the parameters manually (That's why I want to train an ai model for pdf denoising).

I have recorded a vedio about this, and uploaded it bilibili.

https://www.bilibili.com/video/BV1Yp4y1e7jb/?spm_id_from=..search-card.all.click&vd_source=9630992a410cea5e816b2568126c0371

[meihao5631]

你好，请问兄弟你实验了没，效果咋样啊

[Yikai-Liao]

你好，请问兄弟你实验了没，效果咋样啊

没空整啊