table structure recognition
table extraction

Adding config.json and preprocessor_config.json to make the model usable

#1
by alexneakameni - opened

When trying to load this model, I have this issue :
OSError: bsmock/TATR-v1.1-All does not appear to have a file named config.json. Checkout 'https://huggingface.co/bsmock/TATR-v1.1-Fin/main' for available files.

I think you have to define those files to make this repo usable. If it's the same architecture as https://huggingface.co/microsoft/table-transformer-detection/tree/main, could please add config.json and preprocessor_config.json as in that repo ?

We haven't converted these weights yet to be able to be loaded into the Hugging Face version of Table Transformer. These weights can be downloaded and used with the microsoft/table-transformer repository on GitHub. But making this model directly loadable in Hugging Face is on our todo list and we should get to it soon.

Best,
Brandon

Hey, great project! Any update on the config.json file yet? I'm able to download the weights, but without the model architecture used to generate the model, I'm left guessing how to build the model in PyTorch in order to apply the weights. Is there anywhere that has this information that I missed?

Hi @bsmock I was just wondering if there's any update on the timeline for when the weights for the Table Transformer(TATR-v1.1-All) will be ready for use in Hugging Face. Really excited about integrating this into my project as soon as it's available. Thanks for all the hard work you're putting into this!

@bsmock Any updates on this?

you can use this, microsoft have published the same; https://huggingface.co/microsoft/table-transformer-structure-recognition-v1.1-all

for preprocessor just use

from transformers import DetrImageProcessor

image_processor = DetrImageProcessor(
            do_resize=True, size=1000, longest_edge=1000
        )

Sign up or log in to comment