Exploring AlexNet for ImageNet Classification

Unleashing Vision, One Image at a Time

This github repository demonstrates the implementation of AlexNet for image classification on the ImageNet dataset. By leveraging transfer learning and fine-tuning techniques, it achieves high accuracy and robust classification performance. The project consolidates key results, metrics, and visualizations into a single Jupyter/Colab notebook notebook, making it accessible and easy to use.

---

Features

• Model: AlexNet architecture adapted for 10 ImageNet classes.
• Performance: Achieved 93.25% accuracy with strong precision, recall, and F1-scores.
• Visualization: Includes Confusion matrix, Learning curves, Metric breakdowns.
• Optimization: Leveraged transfer learning, fine-tuning, and a one-cycle learning policy for efficient training.

---

Usage

To explore and run this project, follow these steps:

1. Open the Colab notebook using the link below:

2. Execute the cells sequentially to:

   • Load and preprocess the ImageNet dataset.
   • Train the AlexNet model using transfer learning and fine-tuning.
   • Visualize the results, including performance metrics and training progress.

3. Download the project documentation for in-depth insights and methodology: Project Documentation

Results Summary

• Accuracy: Achieved 93.25% across 10 ImageNet classes.
• Metrics: High precision, recall, and F1-scores across all classes, indicating robust classification performance.
• Learning Curve: Consistent training loss reduction and early convergence of validation accuracy.

Visualizations

Below are key visualizations from the project:

1. ImageNet Dataset Sample

2. Training Loss Over Epochs

3. Validation Accuracy Over Epochs

4. Learning Curve

5. Overall Metrics

6. Confusion Matrix

7. Model Performance Visualization

8. Model Architecture

---

Insights

• Transfer Learning: Utilized pre-trained AlexNet weights, reducing training time and improving accuracy.
• Fine-Tuning: Adaptive learning rates enhanced performance on the target dataset.
• Efficiency: Combined transfer learning with the one-cycle learning policy for faster convergence and minimized overfitting.

---

Requirements

To run the notebook, ensure you have the following installed in your environment (if running locally):

• Python: Version 3.8 or later
• Libraries:
  • PyTorch
  • torchvision
  • NumPy
  • Matplotlib
  • pandas
  • scikit-learn

To install the required libraries, execute:

pip install torch torchvision numpy matplotlib pandas scikit-learn