Find Cat Breed is a computer vision application that uses on-device machine learning to identify the breed of a cat on a photo. A ResNet classifier bundled in a CoreML is capable of differentiating 44 different cat breeds with a 51% accuracy.
To get best results, several tips have been used such as training with data augmentation. Also for better stability, the app takes several photos a in short time frame and automatically aggregates related classfier predictions.