The MNIST is a well-known image recognition dataset, so much so that it is often used to benchmark new models. The dataset contains handwritten numbers from 0-9. This CNN uses a LeNet architecture, data augmentation, hyperparameter tuning and ensembling to achieve an accuracy of 99.567% - 94th on the Kaggle leaderboard of over 3300 submissions.
The training and test sets can be found on Kaggle: https://www.kaggle.com/competitions/digit-recognizer/data