Description / 描述

The objective of this competition is to create a model which can classifiy the severity level of Diabetic Retinopathy (DR) as accurately as possible from fundoscope images of individuals, and which can outperform the baseline and (hopefully) advanced baseline accuracy. The international clinical disease severity scale is used to define the different classes of DR severity, in accordance with the Early Treatment Diabetic Retinopathy Study (ETDRS). The different severity levels are defined as follows:

• No DR (0)
• Mild DR (1)
• Moderate DR (2)
• Severe DR (3)
• Proliferative DR (4)

A Convolutional Neural Network (CNN) which outperforms the advanced baseline accuracy has been created in this challenge, via the use of a DenseNet201 backbone, application of transfer learning and domain adaptation, together with model training hyper-parameter tuning after only a few iterations.

这个挑战的目标是创建一个能够尽可能准确地从个体的眼底相机图片中分类糖尿病视网膜病变（DR）的严重程度，并且能够超越基线和高级基线的准确性的模型。 DR 严重程度的定义是根据早期治疗糖尿病视网膜病变研究（ETDRS）的国际临床疾病严重程度刻度而定的。 不同的严重程度定义如下：

• 无 DR (0)
• 轻度 DR (1)
• 中度 DR (2)
• 重度 DR (3)
• 增殖性 DR (4)

在这次挑战中，通过使用 DenseNet201 骨架、应用迁移学习和领域适应性，以及仅经过几次迭代的模型训练超参数调优后，我们创建了一个超越了高级基线准确性的卷积神经网络（CNN）。