This repo contains the model weights for TATR (Table Transformer) v1.1, trained on the PubTables-1M and FinTabNet.c datasets, using the training details in the paper: "Aligning benchmark datasets for table structure recognition".

These model weights are intended to be used with the Microsoft implementation of Table Transformer (TATR).

This model was trained to work best on tightly cropped table images (5 pixels or less). Images with significant padding included around the table were not included in the training, and so the model may not perform well on these at inference. Use a table detection model to detect and tightly crop the table prior to passing to this model.

Model weights that can be loaded into the Hugging Face implementation of TATR are coming soon.

FinTabNet.c will be officially released soon. Please see our GitHub repo for a script to create FinTabNet.c from the original FinTabNet dataset.

Please see the following for more details: