Direct content-based retrieval from music scores images
Researchers have developed new methods for content-based retrieval of music scores, moving beyond traditional metadata searches. The study explores characteristics relevant for search and proposes systematic ways to build query datasets. Experiments compare transcription-based Optical Music Recognition (OMR) with transcription-free Transformer and Large Language Models, finding OMR excels in-domain while transcription-free models handle variability better. AI
IMPACT Introduces novel approaches for searching visual music data, potentially improving accessibility for musicians and researchers.