search

microsoft / table-transformer

Table Transformer (TATR) is a deep learning model for extracting tables from unstructured documents (PDFs and images). This is also the official repository for the PubTables-1M dataset and GriTS evaluation metric.
MIT License
2.31k stars 256 forks source link

Issues with row and cell detection (inconsistent behaviour) #124

Closed mspondon closed 1 year ago

mspondon commented 1 year ago

I have attached the sample image from where I am interested in table , row and finally cell detection. I am using inference.py script to do the same. I am attaching the final visualize output from the script inference.py sample_page-0001 sample_page-0001_0_fig_tables sample_page-0001_0_fig_cells

My question is without finetuning our model , is their any way to improve that , btw thanks a lot to the community for such a wonderful model.

bsmock commented 1 year ago

Hi,

Sorry for the delayed response. We would expect some amount of fine-tuning of the PubTables-1M model to be needed for tables that differ from those in the PubTables-1M dataset.

However, we did just release a multi-domain pre-trained model, trained on both PubTables-1M and FinTabNet.c. You can try that model and see if your results improve on this case without doing your own fine-tuning.

Best, Brandon

mspondon commented 1 year ago

@bsmock multi-domain pre-trained model helps. Closing this.