That’s fantastic! Please keep us in the loop. As you’ve probably seen in the technical report, we evaluated both models on Retina images, but there is so much room for further development. We’ve provided fine-tuning notebooks for both models. Please let me know if you need pointers.
I’m having the same idea and evaluate medgemma for zero shot classification on our own fundus dataset, though it is still limited. would love to know more about your progress