snskrt/Shiv_puran_OCR

Dataset Card for Shlok vs Non Shlok in Shiv puran

This dataset will enable a object detection model to differentiate between shlok vs Non-Shlok in this types of books, after that shloks can be cropped and a parallel corpus of image-text can be created.

Dataset Details

Dataset Sources

PDF from internet archive was used, will link it here later

Uses

To further develop Sanskrit OCR

Direct Use

Write a python script to crop the shlok boxes and then manually link them to the text, either by annotating(too lengthy) or make a json key-value pair.

Dataset Structure

the annotation files have all the coordinates and that stuff.

• from page_0 to page_102 ---> Vidhyeshwar Samhita(manually annotated, and then trained a model on it, will link the model later)

• from page_103 to page_463 ---> Rudra Samhita(Used the model to do infernece, and then some outliers were manually annotated)

• from page_464 to page_508 ---> Shat Rudra Samhita(Used the model to do infernece, and then some outliers were manually annotated)

For any queries contact me on X

Aluminium