Spaces:

iakarshu

/

docformer_for_document_classification

like 6

Runtime error

Hi there,

Actually, I trained this model on just 250 Document Images (since there were resource constraints). If you want that weight, definitely you can go to the app.py file, and see the line, which loads weights from checkpoints. But if you have resources (which I don't), you can train them yourself.

I am sharing the link for the tutorials here: https://github.com/uakarsh/docformer/tree/master/examples/docformer_pl
It contains the pre-training task code as well as RVL-CDIP Document Classification

The script for my training on RVL CDIP can be found here, https://www.kaggle.com/code/akarshu121/document-image-classification-with-docformer/notebook

Hope it helps,

Regards,
Akarsh

Thanks for the links! Are you pre-training your model on all three tasks?

Actually, that were was not any pre-training involved (for the model used in this space), but the script of pre-training is provided on github, so if you can do that, it would be great.

If you are talking about the tasks for pre-training, currently I have the script for MLM, and the function for TDR (Text Describe Image), so shortly I would be releasing the script for all the 3 pre-training tasks, and if I get a chance to use more resources to pre-train and share the checkpoints, I would definitely do that.