|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
base_model: mobilenetv2 |
|
|
datasets: |
|
|
- custom |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
pipeline_tag: image-classification |
|
|
library_name: tensorflow |
|
|
tags: |
|
|
- retinal-disease-detection |
|
|
- medical-imaging |
|
|
- fundus-images |
|
|
- mobilenetv2 |
|
|
- classification |
|
|
- grad-cam |
|
|
- retina |
|
|
model_name: RetinaVision-MNet |
|
|
--- |
|
|
# 🧠 RetinaVision-MNet |
|
|
|
|
|
**RetinaVision-MNet** is a custom-trained MobileNetV2-based deep learning model for multi-class retinal disease detection. |
|
|
It predicts **10 retinal conditions from fundus images** and includes **Grad-CAM heatmaps** to provide interpretable visual explanations for every prediction. |
|
|
|
|
|
The model is trained entirely from scratch and is hosted on Hugging Face due to GitHub’s file-size limitations. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔥 Key Features |
|
|
|
|
|
- **10-class retinal disease classification** |
|
|
- **MobileNetV2 backbone** — lightweight and efficient for medical imaging |
|
|
- **Grad-CAM interpretability** for understanding model decisions |
|
|
- **Custom-trained model (.h5)** using Keras / TensorFlow |
|
|
- **Optimized for FastAPI deployment** with async inference |
|
|
- Works seamlessly with secure JWT-protected backend |
|
|
|
|
|
--- |
|
|
|
|
|
## 📦 Usage |
|
|
|
|
|
Download the model file from the **Files and Versions** tab and place it in your project: |
|
|
|
|
|
```python |
|
|
from tensorflow.keras.models import load_model |
|
|
|
|
|
model = load_model("mobile_model.h5") |
|
|
|
