🔗 GitHub
Description
This project demonstrates how to train an image classifier from scratch using TensorFlow and Keras on the Kaggle Cats vs Dogs dataset. The goal is to classify images into two categories: cats and dogs. The project covers the following key steps:
- Setting up the data: Downloaded the Kaggle Cats vs Dogs dataset and prepared it for training.
- Building the model: Developed a Convolutional Neural Network (CNN) from scratch.
- Training and evaluation: Trained the model and evaluated its performance on a test set.
- Visualization: Visualized the performance through accuracy and loss graphs, as well as kernel feature maps.
Dataset
The model is trained on the Kaggle Cats vs Dogs dataset. You can access it here.
Results
- Achieved 92% accuracy on the test set.
- Visualized kernel feature maps to understand how the model makes decisions.
Next Steps
- Data Augmentation: Experiment with data augmentation techniques to improve model generalization.
- Model Architecture: Refine the architecture to boost performance.
- Hyperparameter Tuning: Explore tuning for optimal results.
- Transfer Learning: Implement transfer learning to reduce training time and improve results.
- Deployment: Save the model and deploy it for real-world applications.
Technologies Used:
- Deep Learning: TensorFlow, Keras
- Data Preprocessing: NumPy, Matplotlib
- Dataset: Kaggle Cats vs Dogs