Motive and Goal is developing an AI based eye diagnostic machine that can detect eye problems of a person.
Data Collection
Data collected from: https://www.kaggle.com/datasets/gunavenkatdoddi/eye-diseases-classification
This dataset contains huge data, almost 4500+ images found.
Data Modification
This dataset contains quality images. Since all the images are the same color but different in shade, I convert the images to GrayScale. This process reduces training time, facilitates internal computation, increases Accuracy.
Using RGB images I've got 81% accuracy.
But for GrayScale images Accuracy increased to 86%.
FrameWork
For this project I use Google Colab, FastAi, Pytorch libraries. These are really helpful to implement code.
Model selection
In this project I've experiment with different models. Finally, FastAi provides xresnet34deeper model, that gave me better result than another model.
I've tried Resnet34, Resnet50, vgg16 and lastly xresnet34_deeper.
Train and validation
I tried 4 times to train models and I've got some findings.
| Model | Accuracy |
|---|---|
| Resnet34 | 80% |
| Resnet50 | 81% |
| vgg16 | 79% |
| xresnet34_deeper with rgb image | 84% |
| xresnet34_deeper with GrayScale image | 86% |
Confusion Matrix
Prediction
I deployed my project on HuggingFace website. Link is : https://huggingface.co/spaces/Rimi98/EyeDiseaseClassifier
You can use my project from this website: https://aklimarimi.github.io/website/
This project is completed, deployed, integrated.