https://github.com/assem-elqersh/alexnet-vgg-cifar10-comparison
Comparative analysis of AlexNet and VGG architectures trained on the CIFAR-10 dataset. The project explores differences in model design, feature extraction capabilities, and performance metrics across different neural network architectures.
https://github.com/assem-elqersh/alexnet-vgg-cifar10-comparison
alexnet batch-normalization cifar-10 cifar10 pytorch vgg16 vgg8
Last synced: 5 months ago
JSON representation
Comparative analysis of AlexNet and VGG architectures trained on the CIFAR-10 dataset. The project explores differences in model design, feature extraction capabilities, and performance metrics across different neural network architectures.
- Host: GitHub
- URL: https://github.com/assem-elqersh/alexnet-vgg-cifar10-comparison
- Owner: Assem-ElQersh
- License: mit
- Created: 2025-05-07T22:17:47.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2025-05-08T11:39:06.000Z (5 months ago)
- Last Synced: 2025-05-12T08:06:53.246Z (5 months ago)
- Topics: alexnet, batch-normalization, cifar-10, cifar10, pytorch, vgg16, vgg8
- Language: Jupyter Notebook
- Homepage:
- Size: 3.68 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# AlexNet and VGG Architectures Comparison on CIFAR-10
This repository contains implementations and comparative analysis of AlexNet and VGG architectures trained on the CIFAR-10 dataset. The project explores differences in model design, feature extraction capabilities, and performance metrics across different neural network architectures.
## Project Overview
The goal of this project is to compare and analyze the performance of several CNN architectures:
- AlexNet (adapted for CIFAR-10)
- VGG-16
- VGG-16 with Batch Normalization
- VGG-8 (reduced depth variant)
## Repository Structure
```
├── README.md
├── report/
│ └── alexnet_vgg_cifar10_report.pdf # Detailed analysis report
├── src/
│ └── implementations.ipynb # PyTorch implementation code
└── results/
├── 5 epochs results/ # Results from 5-epoch training
│ ├── results/
│ │ └── experiment_results.json
│ └── visualizations/ # Feature maps and training curves
└── 40 epochs results/ # Results from 40-epoch training
├── results/
│ └── experiment_results.json # Performance metrics
└── visualizations/ # Feature maps and training curves
```
## Key Findings
Based on our 40-epoch training results:
| Model | Parameters | Test Accuracy | Train Accuracy | Avg. Time per Epoch |
|-------|------------|---------------|----------------|---------------------|
| AlexNet | 23,238,474 | 87.81% | 94.14% | 36.09s |
| VGG-16 | 33,638,218 | 10.00% | 9.94% | 47.08s |
| VGG-16 BN | 33,646,666 | 91.94% | 99.43% | 50.20s |
| VGG-8 | 13,119,050 | 88.32% | 93.11% | 37.97s |
## Key Observations
1. **Batch Normalization Impact**: VGG-16 with Batch Normalization significantly outperforms standard VGG-16, achieving the highest test accuracy of 91.94%.
2. **Model Size vs Performance**: The smaller VGG-8 model (13M parameters) achieves comparable performance to AlexNet (23M parameters) while being more efficient.
3. **Training Difficulty**: Standard VGG-16 without batch normalization failed to learn effectively (10% accuracy), highlighting the importance of normalization in deep architectures.
4. **Speed vs Accuracy Tradeoff**: AlexNet has the fastest training time per epoch while VGG-16 BN delivers the highest accuracy but with the slowest training time.
## Dataset
This project uses the CIFAR-10 dataset, which consists of 60,000 32x32 color images across 10 classes:
- airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck
The dataset can be automatically downloaded using the PyTorch or TensorFlow API as shown in the implementation notebook.
## Usage
To run the code:
1. Clone this repository
2. Install dependencies:
```
pip install torch torchvision matplotlib numpy tqdm
```
3. Open and run the notebook in `src/implementations.ipynb`
## Future Work
- Implement other modern architectures (ResNet, EfficientNet)
- Apply advanced regularization techniques
- Explore performance on other datasets
- Implement attention mechanisms
## License
This project is licensed under the MIT License - see the [LICENSE](https://github.com/Assem-ElQersh/AlexNet-VGG-CIFAR10-Comparison/blob/main/LICENSE) file for details.