FLSS Dataset

# FLSS Dataset: Full-Page Grand Staff Optical Music Recognition via YOLO11m-Based Segmentation and Sheet Music Transformer [](https://doi.org/10.5281/zenodo.19068882) ## 1. Dataset Overview The **FLSS Dataset** is a high-fidelity, large-scale dataset specifically constructed to address the persistent challenges of spatial–temporal alignment in full-page grand staff Optical Music Recognition (OMR). It contains **45,000 densely annotated full-page piano score images**, spanning diverse musical periods from Baroque to contemporary repertoires. Unlike existing OMR datasets that primarily focus on isolated monophonic segments or cropped staves, FLSS explicitly provides comprehensive hierarchical annotations. It serves as a rigorously labeled benchmark for layout analysis (e.g., staff-group detection, brace recognition), region-level symbol transcription, and end-to-end global score reconstruction. This dataset is the official accompanying data repository for the paper:> **"Full-Page Grand Staff Optical Music Recognition via YOLO11m-Based Segmentation and Sheet Music Transformer"** (Under Review at IJDAR). ## 2. Data Source and Copyright Compliance To strictly ensure structural reproducibility and facilitate open research, all raw sheet music images within this dataset are sourced exclusively from the **International Music Score Library Project (IMSLP)**. **Copyright Statement:** All previously included proprietary or copyrighted materials have been completely removed. The curated images are in the **public domain**, and the corresponding structural annotations created by our research team are released under the open-access license specified below. There are no third-party licensing restrictions for non-commercial academic use. ## 3. Dataset Structure The dataset is organized into paired folders containing the original scanned images and their corresponding machine-readable annotations. ```textFLSS_Dataset/│├── images/ # High-resolution full-page score images│ ├── train│ ├── val #(up to 5000 images)│ │└── labels/ # Hierarchical structural and sequential labels ├── train # (e.g., bounding boxes ) ├── val