search

fh2019ustc / DocTr-Plus

The official code for “Deep Unrestricted Document Image Rectification”, TMM, 2023.
https://project.doctrp.top/
MIT License
375 stars 40 forks source link

How can train with custom dataset? #4

Open WongVi opened 1 year ago

WongVi commented 1 year ago

Could you please let me know the training code so that I can train with custom dataset?

fh2019ustc commented 1 year ago

Due to commercial application reasons, we will not release the source training code for the time being.

drzraf commented 7 months ago

Quoting https://github.com/fh2019ustc/DocScanner:

Acknowledgement The codes are largely based on DocUNet and DewarpNet. Thanks for their wonderful works.

Indeed, and as you probably know, all your precursors have, for the last 2 decades, released their work in the open.

fh2019ustc commented 7 months ago

@drzraf DocScanner is a advanced research work. It was completed earlier than DocTr, but it has been maliciously rejected. I don't have to open source the code before it's accepted. In addition, I would like to ask what your contribution is to the field?

drzraf commented 7 months ago

I don't have to open source the code before it's accepted.

granted.

In addition, I would like to ask what your contribution is to the field?

I'm a just hobbyist... but I don't have to be an (involuntary) research's mere taxpayer to remind that 90% of western CV research is still done fully in the open (and had been so for the last 15+ years) while 90% of rising Chinese CV research popping up in the last 5 years ends up behind a curtain.

It's an unfortunate pattern. (But I do understand, now, that in this particular situation, you rightfully feel safer not releasing a PoC model to the crowd until the paper is actually published) I wish your work could find its way and hopefully allow you to do a regular public release at some point.

Best regards