A Deeper Look at Sheet Music Composer Classification Using Self-Supervised Pretraining

This article studies a composer style classification task based on raw sheet music images. While previous works recognition have relied exclusively supervised learning, we explore the use of self-supervised pretraining methods that been recently developed for natural language processing. We first convert images to sequences musical words, train model large set unlabeled “sentences”, initialize classifier with pretrained weights, and then finetune small labeled data. conduct extensive experiments International Music Score Library Project (IMSLP) piano data using range modern architectures. show substantially improves performance Transformer-based architectures perform best. also introduce two augmentation strategies present evidence learns generalizable semantically meaningful information.