Few-Shot Music Symbol Classification via Self-Supervised Learning and Nearest Neighbor

Abstract

The recognition of music symbols within score images represents one of the main stages in Optical Music Recognition systems. While current state-of-the-art methods based on Deep Learning are capable of adequately performing this task, they generally require a vast amount of data that has to be manually labeled. Such a particularity generally limits their applicability when addressing historical manuscripts with early music notation, for which annotated data is considerably scarce. In this paper, we propose a self-supervised learning-based method that addresses this task by training a neural-based feature extractor with a set of unlabeled documents and performs the recognition task considering just a few reference samples. Experiments on a reference early music corpus report that the proposal outperforms the contemplated baseline strategies even with a remarkably reduced number of labeled examples for the classification task.