MNIST: Hello World of Computer Vision
MINST is a defacto standard benchmark for classification in Computer Vision. It is now dated, because current models and hardware easily achieve high accuracies (nearly 100%), which does not give any meaningful information while comparing them to each other. However, it's simplicity is still a great advantage for Hello World type of programs while familiarising with a new framework.
I did exactly that, and I achieved over 99% accuracy with a relatively simple model (63,330 trainable parameters).
