This is a course project for the HTW Berlin and its finality is to analyze and classify with the help of a Residual Neuronal Network (50), a state-of-the-art architecture of Convolutional Neuronal Network, images of retinas which could be healthy or present some kind of sign of diabetic retinopathy.
This challenge was presented in Kaggle.
In this notebook you will find a custom implementation of ResNet-50, which was used, in help with the pretrained ImageNet weights, to classify the images.
All the image/label files can be found at the challenge in here. Download and extract the trainLabels.csv and save it as train_labels.csv . The test labels can be found here, download it and save it as test_labels.csv
To preprocess the images please refer to the implementation of my project partner in this repository. After you have followed his steps you should have a train_dst.zip and a test_dst.zip.
To be able to run this notebook you have to have a Google Drive account. The project structure in Google Drive should be as follow:
/retina_images
--/pretrained
--resnet50_weights_imagenet_no_dense.h5
--/history
--/checkpoints
--resnet50_dense_regression_last_384_final.h5
--train_labels.csv
--test_labels.csv
--train_dst.zip
--test_dst.zip
Open the notebook in Google Colab and start running it. At the 4th cell you will be prompted to enter a token for your Google Drive account, so the Notebook has access to the files you just uploaded.
Thanks to EyePACS for providing the retinal images for the original competition.
Cuadros, J., & Bresnick, G. (2009). EyePACS: an adaptable telemedicine system for diabetic retinopathy screening. Journal of diabetes science and technology, 3(3), 509–516.
Thanks to Tobias for providing the preprocessed images.