Prepare a non-standard text page for machine learning analysis - Deep Learning, Vision
- Status: Closed
- Præmier: $250
- Modtagne indlæg: 3
- Vinder: garybutler
We are a startup that specializes in AI solutions for businesses. Our fast growing business works in the legal-, real estate- and services industry. For the area ML Vision we are searching for talents who can help us solve challenges like the following:
The overall task is to apply deep learning to pages of text that are formatted in all kinds of strange ways - there are thousands of them and each one is different. You have blocks of text with addresses, invoice numbers, reference numbers etc. in various places. You have tables, different font sizes, headers, footers etc. Basically everything that you can think of in an invoice.
You can assume that they are all clean and perfect OCR.
In order to prepare the page for deep learning, the elements on the page have to be separated, identified, their position on the page has to be identified, their surroundings have to be identified, etc.
What would be your logical approach and what would be the ML Vision technologies you would use to automatically prepare each page for the deep learning, so you can train addresses, account numbers, invoicing amounts, VAT percentages etc. as entities from thousands of differently formatted invoices?
Subsequent and frequent hiring of talents with the best submissions for the contest is very likely.
“Gary posted the only high quality submission to our contest and was available for a conf call to elaborate on his suggested solution. ”
Bedste indlæg fra denne konkurrence