Fine-Grained Diabetic Retinopathy Classification: Challenges and Feature Learning

The classification of diabetic retinopathy involves small differences between lesion points, resulting in minute variations among different categories of retinas. Consequently, the model struggles to distinguish between similar lesion images accurately. To enhance the model's classification accuracy, it must possess a more refined classification ability. The success of fine-grained classification models hinges on the model's ability to learn the relevant target features effectively.