Training a high-performance retinal foundation model with half-the-data and 400 times less compute

Open in new window