SillyBoba
SillyBoba

OCR + transformers

hi everyone, I have a freelance data science project where I need to read 200k medical records PDFs , and extract the text data out of it. The current accuracy of the project is around 60% The client is hell bent to use some sort of transformers to extract the data.

If someone has knowledge of OCRs and Transformers and interested in collaborating on this project with me then please email me your resume and contact number to remotelute36@gmail.com

17mo ago
Find out if you are being paid fairly.Download Grapevine
FluffyTaco
FluffyTaco

TrOCR

PrancingPancake
PrancingPancake

This actually does not give good results as you expect when in say low lighting conditions

GroovyQuokka
GroovyQuokka

Hi, can you please share what you did in this project?

ZippyNarwhal
ZippyNarwhal

my team has worked on a similar project, we had the same problem statement(extracting name,age, doses, ICD codes(and also normalising this) from given hospital records.)

PrancingCoconut
PrancingCoconut

Transformers are for text-to-text. Try ViT. It's supposed to work well on image classification tasks so maybe you'd have to draw bounding boxes on each character and then classify it and append to the output text

Discover more
Curated from across