This is pulling data from an image of a table into a browsable table. It definitely seems promising. Far from perfect, but this is a proof of concept after all. Huge thanks to @jakobrosin for the idea of using a table.
Based on my previous tweet, this could very well be novel. If anyone who relies on a screen reader could try it out and give me any feedback, it'd be so appreciated.
@hannah@jakobrosin I don’t have a suitable image to test this RN, but have you thought of exporting the table to markdown and then putting it through an LLM to fix up the results? This is a simple enough task that ChatGPT should have no problem whatsoever, and it probably wouldn’t be easy to do any other way.
@miki I think there's potential there, but there's some definite lower-hanging fruit. It has coordinate data for every character, but the API is grouping them across lines. If someone more familiar with image processing wanted to step up and write a quick "find all the dividing lines in an image", that would do wonders.
Add comment