Logo Classification - COMP-546DL

Deep learning project for classifying brand logos using the Flickr Logos 27 dataset.

Files

• Logo_Classification.ipynb - Main notebook with all experiments and report
• final_logo_model.keras - Trained MobileNetV2 model
• class_names.json - List of 27 brand labels

How to run

1. Download the Flickr Logos 27 dataset from http://image.ntua.gr/iva/datasets/flickr_logos/
2. Extract it so you have a flickr_logos_27_dataset folder next to the notebook
3. Install dependencies: pip install tensorflow numpy pandas matplotlib pillow scikit-learn seaborn
4. Run the notebook cells in order

Models

• Basic CNN - 3 conv blocks, ~46% validation accuracy
• MobileNetV2 Transfer Learning - ~99.8% validation accuracy (selected as final model)