/Modified_MNIST_CNN

Given a dataset of 50,000 images containing MNIST digits (handwritten numerical digits from 0 - 9), where each image may contain multiple digits with different sizes and orientations, our goal was to design and validate a supervised classification model to find which number occupied the most space in each image. By utilizing widely used convolutional neural networks, we started from a basic network and slowly made it more and more complicated to get a final desirable score of 0.9310 on our validation set.

Primary LanguagePython

This repository is not active