i-Code
i-Code copied to clipboard
Fintuning code on FUNSD and CORD
Thanks for your contribution for Document AI. There are scripts for fine-tuning RVL-CDIP and DUE-benchmark,but the one for FUNSD and CORD seems missing. Could you also share this script?
Besides, the CORD F1 score of DONUT reported in your paper is much higher than the one in original paper (91.6 vs 84.1). Could you given some explanations?
I would like to echo this question as well.
I've had trouble replicating the results from the Information Extraction task on FUNSD and CORD. I suspect there is something subtly different about the prompts we are using or about how you tie the generation back to the original data.
Hi, I am trying to replicate the results from the Information Extraction task on FUNSD. Any updates on the script that can be used?
Any updates?