Auto-Grade

Converting handwritten answers to text using Handwriting recognition technique and grading them accordingly.

Model

Dataset: MNIST dataset was used.

Architecture:

Training accuracy : 98%

Test accuracy : 99.01%

Answer Detection

First we convert the pdf to image
Then using opencv we detect the rectangles in which the answers should be given.
Cropping out the answers and processing it before passing it to the model for prediction.
Prediction

To run files on the website check out the format or use :

Data/answers-multipage.csv

Data/answers-multipage.pdf