Demonstration of training and developing CNNs to classify Fashion-MNIST dataset using Pytorch.
Demonstration and code provided as a single notebook.
Fashion-mnist is a small dataset for fashion product classification consisting 60,000 training images and 10,000 test images. Here I demonstrate how you can train your own CNN network(s) which reach(es) similar accuracy (~95) as the benchmarks and yet still “efficiently” (e.g. without millions of parameters). I also show the process of developing a Deep learning architecture and tuning its hyperparameters by testing at least 3 model variations using different strategies, e.g. network structures, data augmentations, regularizations, etc.