Overview

This project implements a convolutional neural network (CNN) model to classify retinal images into four categories. The model utilizes transfer learning with the pre-trained MobileNetV3Large architecture, fine-tuned on the retinal image dataset. The objective is to achieve high accuracy and F1 score on both training and validation datasets.

Data Preparation

Dataset Loading

• Training and validation datasets are loaded using TensorFlow's image_dataset_from_directory method.
• Both datasets contain images classified into four categories.
• Dataset details:
  • Training set: 76,515 images
  • Validation set: 21,861 images
• Images are resized to 224x224 pixels with three color channels (RGB).
• Batch size is set to 32.
• Shuffling is enabled during data loading for randomness.

Code Sample

python
CopyEdit
train_data = image_dataset_from_directory(
    r"..\\data\\retina\\train",
    labels="inferred",
    label_mode="categorical",
    color_mode="rgb",
    batch_size=32,
    image_size=(224, 224),
    shuffle=True
)

valid_data = image_dataset_from_directory(
    r"..\\data\\retina\\val",
    labels="inferred",
    label_mode="categorical",
    color_mode="rgb",
    batch_size=32,
    image_size=(224, 224),
    shuffle=True
)

Model Architecture

Base Model: MobileNetV3Large

• MobileNetV3Large serves as the backbone architecture.
• It is initialized with ImageNet pre-trained weights.