This deep learning project focuses on building a highly accurate image classifier to distinguish between cats and dogs using transfer learning techniques. Our model achieves exceptional performance metrics:
- Accuracy > 90%
- Error Rate < 0.8
What makes our model particularly strong is its resilience - it maintains high accuracy even with unclear, blurry, or otherwise challenging images that would typically confuse less robust classifiers.
- Utilizes state-of-the-art transfer learning techniques
- Robust performance even with imperfect input images
- Comprehensive model evaluation and visualization
- Deployed as a user-friendly web application
We leveraged pre-trained models and fine-tuned them for our specific classification task. The project includes:
- Data preprocessing and augmentation
- Model architecture selection
- Transfer learning implementation
- Performance optimization
- Detailed evaluation metrics
- Ousmane KA
- Albert Sandokh BAKHOUM
Polytech Diamniadio / UAM - Master 2 en Ingénierie des Systèmes d'Informations et de données
- Jupyter Notebook: Complete implementation with detailed explanations (provided in HTML format)
- Web Application: Deployed model accessible online
🔗 https://ok-cat-dog-classifier.streamlit.app/
Try it with your own pet photos!
Our classifier demonstrates exceptional robustness with the following capabilities:
- Accurate classification of blurry or low-resolution images
- Consistent performance with partially obscured animals
- Reliable predictions for unusual angles or lighting conditions
- Handles various image artifacts and noise effectively
To explore the project implementation:
- Download the provided HTML notebook
- Open it in any modern web browser
- Follow along with the detailed comments and explanations
- Mobile application integration
- Real-time classification from video feeds
- Expanded animal classification capabilities
This project was developed as part of our Master's program in Information and Data Systems Engineering at Polytech Diamniadio / UAM.