fh2019ustc / doctr-plus Goto Github PK

hello, i had realized myself code base on your open code and get good results, but i cant understand the mathematical meaning of convex upsample? could you tell me or some knowledge points？ @fh2019ustc

2019

您好，您分享的代码不完全，inference.py是空的

Thank you for open source your research code but didn't find a paper download link. Where can I find the paper? Thanks

作者您好，我阅读了GeoTr结构框架，发现每一层encoder和decoder都用了两层的MultiheadAttention，论文中说encoder是标准的transformer框架中的encoder，但是transformer中的encoder只用了一层的MultiheadAttention，decoder用了两层。因为我最近在做论文复现的工作，想咨询一下，期待回复！

Hello, I have filled in the form to obtain the doc3d data set. It has been more than a week since I filled it out, but I have not received the voucher yet. Can you make a doc3d dataset for me? Thank you very much for your contribution. My email is: [email protected]

hi， I want to know where the pre-trained weights is, and in addition, why is the inference code empty?

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

The loss function is defined as the L1 distance between the predicted warping flow f_b and its given ground truth f_{gt}. However, how is the ground truth obtained? In the link UDIR test set, there are only original and scanned images, but there is no ground truth matrix of pixel-wise corresponding between the two images.

inference.py代码是空的，还不恢复吗？

hello,can you reveal the pre-training model

hello, did you generate incomplete document data yourself or based on an existing data set? Does this model have a negative effect on document rectification with boundaries?

I git clone the repository and just read the code, but I found that the content of inference.py has been deleted. Is it because of the publication of the paper? Thank you!

How to crop Doc3D with corresponding backward flow？