Hi everyone, first time building a dataset. This is a v0.1, about 100 scans of book pages (both single and double-page per scan). The books are in the public domain. The intended use is for anyone looking to do image-to-text software work.<\/p>\n
The scans are in a .jpg format, with a PDF with the whole collection.<\/p>\n
I have also included 2 .txt files:<\/p>\n
1)”raw” (aka not corrected for halluciations, artifacts, etc.) .txt file for anyone looking to do a check. The file is in Markdown.<\/p>\n
2) A “corrected” .txt file, where the hallucinations, artifacts, errors, etc. were manually corrected. This file is in .txt, not Markdown.<\/p>\n
Looking for feedback if this is useful, how to make a dataset like this better, etc.<\/p>\n
Kaggle: https:\/\/www.kaggle.com\/datasets\/booksofjeremiah\/serbian-cyrillic-script-printed<\/a><\/p>\n Huggingface: https:\/\/huggingface.co\/datasets\/Books-of-Jeremiah\/raw-OCR-serbian-cyrillic<\/a><\/p>\n Any feedback on whether the set is useful for other use cases or how it can be made better is appreciated!<\/p>\n<\/div>\n submitted by \/u\/Books_Of_Jeremiah <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
