Fast retinal disease diagnosis through fundus camera image classification, trying three different approaches.
GitHub: https://github.com/PhysGuy01/DeepLearning2026

Simple custom CNN with residual blocks without any pretraining. (11M params)

Same encoder as Acmonides' but pretrained on vessel segmentation (brontes_segmentation). (11M params)

ResNet50's backbone finetuned on this task. (24M params)

Training dataset from the article: https://www.nature.com/articles/s41597-022-01564-3

Francesco Specchia