README.md

🐶 Dog Vision

📘 Overview

A deep learning project that classifies dog breeds from images using TensorFlow and Keras.
The project follows a complete ML pipeline — from data preprocessing and augmentation to model training and evaluation.

🧠 Objectives

• Build a convolutional neural network (CNN) for dog breed classification.
• Apply transfer learning with pre-trained models.
• Optimize model performance using callbacks and early stopping.

⚙️ Tools & Libraries

• Python, TensorFlow, Keras
• NumPy, Matplotlib, Seaborn

🚀 Workflow

• Data Loading & Preprocessing
  • Loaded images from directories
  • Applied data augmentation
• Model Architecture
  • Built CNN from scratch
  • Integrated pre-trained MobileNetV2
• Model Training
  • Compiled with Adam optimizer
  • Trained with early stopping
• Model Evaluation
  • Plotted training/validation curves
  • Evaluated accuracy on test set

📊 Results

• Best Model: MobileNetV2 with fine-tuning
• Top Accuracy Achieved: ~90% (from notebook metrics)
• Key Influencing Features: Image resolution, augmentation techniques, transfer learning layers

📈 Visual Insights

• Transfer learning significantly improved accuracy.
• Data augmentation reduced overfitting.
• Fine-tuning pre-trained layers enhanced model generalization.

🧩 Key Takeaways

• Gained hands-on experience with deep learning and transfer learning.
• Improved skills in image data preprocessing and augmentation.
• Learned to build and fine-tune CNNs for image classification tasks.

📁 Notebook

• View Project Notebook

🛠️ Setup Instructions

• Clone the repository: git clone https://github.com/Aksachlisimo/dog_vision_project
• Install dependencies: pip install -r requirements.txt (create if not present: tensorflow, numpy, matplotlib, seaborn)
• Open the notebook in Jupyter and run all cells.

👨‍💻 Author

• Mohamed Mouhimine
• AI & Machine Learning Engineer