Project: Create and Train a CNN for Classification

Image classification is a basic computer vision task. In this project, we’ll build a CNN using the layers we studied in the previous lessons. We’ll then train the CNN to classify 28x28 grayscale images of handwritten digits from the MNIST dataset of 6000 images.

Attempting to classify these images with traditional computer vision programming would be difficult. It is one of those cases where we can visually identify these images as 1 of the 10 digits without effort, but we would have a hard time explaining how we got to the correct answer. It is the perfect application for machine learning.

After training, our CNN will be put to the test: when fed with example images from the test dataset that were not used during training, the output should be the corresponding class, i.e., one of the 10 digits.