Transfer Learning

# Description

Complete transfer learning project for image classification with custom datasets. Uses ImageNet pre-trained models (VGG16, ResNet50, MobileNetV2, EfficientNetB0) and adapts them to a new classification problem via two approaches: (1) Feature Extraction: freeze the pre-trained backbone and train only new classifier heads, (2) Fine-Tuning: unfreeze part of the backbone and adjust the whole network with reduced learning rate. Compares accuracy, training speed and model size across architectures. Implements transfer learning-specific data augmentation and analyzes learning curves during fine-tuning.

$ ./deploy.sh $ git clone

# Key features

$ Pre-trained models: VGG16, ResNet50, MobileNetV2, EfficientNetB0

$ Feature extraction: frozen backbone + new dense heads

$ Fine-tuning: selective unfreezing with reduced learning rate

$ Accuracy, speed and size comparison across architectures

$ Transfer learning-specific data augmentation

$ Learning curve analysis and layer unfreezing strategy

# Gallery

Desktop view

Mobile view

# Technologies used

Python TensorFlow / Keras