MuSViT — foundation vision model for sheet music (MAESTRO mirror)
Mirror of PRAIG's MuSViT encoders (ECCV 2026), laid out one subdirectory per size for MAESTRO's download-on-demand loader:
| subdir | upstream | params | hidden |
|---|---|---|---|
musvit/ |
PRAIG/musvit | ~0.1 B | 768 |
musvit-light/ |
PRAIG/musvit-light | small | 384 |
Both are standard vit_mae checkpoints (config.json + model.safetensors),
byte-identical to upstream (sha256-verified at mirror time). Pre-trained with
Masked Autoencoders on 9.7 million sheet-music page images from IMSLP; load
through transformers.ViTModel for deterministic page embeddings
(1024×1024 input → CLS token).
License — CC BY-NC-SA 4.0 (non-commercial)
These weights are released by PRAIG under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 license and are redistributed here unchanged under the same terms. Non-commercial use only. Attribution: PRAIG (Pattern Recognition and Artificial Intelligence Group, University of Alicante).
Citation
@inproceedings{penarrubia2026musvit,
title = {MuSViT: A Foundation Vision Model for Sheet Music Representation},
author = {Penarrubia, Carlos and Rios-Vila, Antonio and Fuentes-Martinez, Eliseo
and Martinez-Sevilla, Juan C. and Castellanos, Francisco J. and
Alfaro-Contreras, Maria and Calvo-Zaragoza, Jorge},
booktitle = {European Conference on Computer Vision (ECCV)},
year = {2026}
}
The authors acknowledge Edward Guo / IMSLP (Petrucci Music Library) for the training data. Mirrored for MAESTRO by AEmotionStudio.