ppaanngggg/layoutreader

Licensing and training a variant suitable for commercial use

shoshinL opened this issue · 3 comments

shoshinL commented

Hi Hantian Pang,

I love this project. I'm a Univeristy student from Germany and have the following questions:

  1. As this is a fine-tune of LayoutLMv3 I cannot use this in a commercial project, right?
  2. When trying to implement an approach like this, using only the bbox coordinates, but using a model with a more permissive license, like LiLT, do you think it could work? Could you disclose what you used to train, as I'd like to know if I need to have access to some strong GPU.

Thanks in advance for the answers!

ppaanngggg commented
  1. Thanks for your question, I found it use CC-BY-NC-SA, so it can't be used in commerical project. And I update LICENSE file now.
  2. Great advice, I think LiLT works, you can have a try. I train by 8*RTX3090, you can rent cheep GPUs from vast.ai
  • ❤️1
ppaanngggg commented

@shoshinL do you need any other help? If not, I will close this issue.

shoshinL commented

I don't have any further questions.
Thank you for your help and this great repo!

  • 😄1