Zanina

I m working on similar project and I find your work interesting.. can you share it with me ? this is my email : [email protected] thank you in advance

Hello, I couldn't find the exact input data to train the model on the icdar dataset , can you provide explanations for it ? thanks.

the model accepts both image and bounding boxes and corresponding transcripts as input, you can't only rely on image itself.

it would be better to check a model for detecting table shape, and then you can parse the content and arrange it, I think the PICK model would be more...

I think you ended up with an overfitting problem, how many images did you use for train/test data ?

you need to process your image using an ocr and then extract each field bounding boxes

I can't understand what you mean

I am using tesseract as text detection and text recognition tool , the author said that it wasn't good for result accuracy , I still don't know why

@AtulKumar4 thanks for the answer , can you please provide more informations about the format of IOB labels if the model perfroms sequence tagging in character level ?

yes thank you so much, I still have confusion .. in the code they used both word and character embedding , so from what I understood the BiLSTM layer accecpts...