Acknowledge training datasets (incl. Leuven-Haifa) in README and clarify data-use terms for model weights
Hi VascX team,
Thanks for releasing VascX. It's great to see the models and weights openly available.
I'm one of the authors of the Leuven-Haifa High-Resolution Fundus Image Dataset. Your paper (arXiv 2409.16016, Table 1) says it was used to train the vessel and artery/vein segmentation models. Thank you for citing it in the paper. I have two small requests:
1. Dataset acknowledgement in the README
The README doesn't currently say which datasets the released weights were trained on. Could you add a short "Training data" / "Acknowledgements" section listing them, with their citations? Most VascX users will only see the repo and the Hugging Face card, not the paper. For Leuven-Haifa, the dataset terms ask that the following be cited:
bibtex
@article {vaneijgen2024leuvenhaifa,
title = {Leuven-Haifa High-Resolution Fundus Image Dataset for Retinal Blood Vessel Segmentation and Glaucoma Diagnosis},
author = {Van Eijgen, Jan and Fhima, Jonathan and Billen Moulin-Roms{'e}e, Marie-Isaline and Behar, Joachim A. and Christinaki, Eirini and Stalmans, Ingeborg},
journal = {Scientific Data},
volume = {11},
number = {1},
pages = {257},
year = {2024},
doi = {10.1038/s41597-024-03086-6}
}
@article {fhima2024lunet,
title = {LUNet: Deep Learning for the Segmentation of Arteriesolution Fundus Images}, author = {Fhima, Jonathan and Van Eijgen, Jan and Billen Moulne and Brackenier, Helo{"i}se and Kulenovic, Hana and Debeuf, author = {Fhima, Jonathan and Van Eijgen, Jan and Billen Moulne and Brackenier, Helo{"i}se and Kulenovic, Hana and Debeuf,Val{'e}rie and Vangilbergen, Marie and Freiman, Moti and Stalmaachim A.}, journal = {Physiological Measurement}, year = {2024},
doi = {10.1088/1361-6579/ad3d28} }Dataset record: https://doi.org/10.48804/Z7SHGO (KU Leuven RDR)It would also help if the README asked users of the vessel and Aning datasets alongside VascX.
2. Data-use terms for the released weights
Leuven-Haifa is distributed under custom terms through a Data Trts use to non-commercial purposes. The weights on Hugging Faceare currently released under AGPL-3.0, which allows commercial use. Could you add a note to the README and model card saying that the weights were trainepartly on datasets with non-commercial restrictions, and that usms before any commercial use? Several other public datasets inyour training mix carry similar restrictions.
Happy to help with wording or a PR. Thanks again!