Image Classification with EfficientNet and ResNet

Overview

This project is an implementation of an image classification model using EfficientNetB0 and ResNet50 architectures. It employs TensorFlow and Keras for model training and evaluation. The models are trained on an augmented dataset with binary classification (Fake vs Real).

Features

• Uses EfficientNetB0 and ResNet50 pretrained models
• Implements image augmentation for better generalization
• Utilizes Early Stopping, Learning Rate Reduction, and Model Checkpointing
• Applies fine-tuning after initial training to improve model accuracy
• Saves trained models and history for further analysis
• Includes visualization functions to plot accuracy trends
• Provides a function for real-time image prediction

Dataset Preparation

Place your dataset in a folder named dataset, structured as follows:

dataset/
    ├── class_1/  # Images belonging to class 1
    ├── class_2/  # Images belonging to class 2

Update dataset_path in the script if your dataset location differs.

Dependencies

Ensure you have the following libraries installed:

pip install tensorflow numpy matplotlib seaborn

Code Structure

• Data Augmentation: Extensive transformations applied to training images.
• Model Building: Two models (EfficientNet & ResNet) built dynamically with custom dense layers.
• Training & Fine-Tuning: Models are first trained, then fine-tuned for better performance.
• Visualization: Accuracy trends plotted for both training and validation.
• Prediction Function: Allows real-time image classification.

Training Process

1. Load the dataset using ImageDataGenerator
2. Train EfficientNet and ResNet models
3. Fine-tune both models
4. Plot accuracy graphs
5. Save the trained models

Model Training & Fine-Tuning

# Train EfficientNet model
efficientnet_model = build_model(EfficientNetB0, "efficientnet")
history_efficientnet = train_model(efficientnet_model, train_data, val_data, "efficientnet")

# Train ResNet model
resnet_model = build_model(ResNet50, "resnet")
history_resnet = train_model(resnet_model, train_data, val_data, "resnet")

# Fine-tune models
history_efficientnet_fine = fine_tune_model(efficientnet_model, train_data, val_data, "efficientnet")
history_resnet_fine = fine_tune_model(resnet_model, train_data, val_data, "resnet")

Model Evaluation

The training history is stored and can be plotted using:

plot_history(
    [history_efficientnet, history_resnet],
    ["EfficientNet Training", "ResNet Training"]
)

Predicting a New Image

To classify a new image, use:

result = predict_image(model, "path_to_image.jpg")
print(result)  # Output: "Fake" or "Real"

Checkpoints & Model Saving

• The best model is saved automatically as {model_name}_best.h5
• The trained model object is pickled for later use (.pkl file)

Contributors

Feel free to contribute to this project by submitting pull requests!

License

This project is licensed under the MIT License.