EFIQA: Explainable Fundus Image Quality Assessment via Anatomical Priors

We ground fundus quality assessment in anatomical integrity rather than subjective labels, enabling an unsupervised approach with spatial explainability and robust cross-dataset generalization.

Currently under review in MIDL2026: https://openreview.net/forum?id=b9TBF3O88T

Code

Currently you can access on GitHub https://github.com/penway/EFIQA.

Inference

Test our model directly online at: https://huggingface.co/spaces/penway47/EFIQA. You can also download script and weight there.