This repo is a collection of classifier, or model you would call, created by Tensorflow from different class of datasets. The purpose of this repo is enable quick inference for those who wanna get started in this field research but have less compute power to train the dataset from the scratch.
To improve the total accuracy of the models, the author reduce the classification class from 5 to 3. The author sees that there are 2 classes that looks similar to each other so it's better to skip it and focus on what differs the most.
Here you can find all the pre-trained models trained on difference platform that you can try for inference. The dataset used to made these pre-trained models are taken from Diabetic Retinopathy (resized) [https://www.kaggle.com/tanlikesmath/diabetic-retinopathy-resized] page on Kaggle.