https://github.com/pierridotite/cifar10-compressai
Convolutional autoencoder for compressing and reconstructing CIFAR-10 images using TensorFlow.
https://github.com/pierridotite/cifar10-compressai
ai autoencoder cifar10 compression computer-vision deep-learning image-compression machine-learning neural-network opensource tensorflow
Last synced: about 2 months ago
JSON representation
Convolutional autoencoder for compressing and reconstructing CIFAR-10 images using TensorFlow.
- Host: GitHub
- URL: https://github.com/pierridotite/cifar10-compressai
- Owner: pierridotite
- License: mit
- Created: 2024-12-29T13:49:38.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2024-12-29T14:47:30.000Z (10 months ago)
- Last Synced: 2024-12-29T15:24:46.118Z (10 months ago)
- Topics: ai, autoencoder, cifar10, compression, computer-vision, deep-learning, image-compression, machine-learning, neural-network, opensource, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 176 KB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project
README
# CIFAR10-CompressAI
## Description
**CIFAR10-CompressAI** is a project that implements a convolutional autoencoder for compressing and reconstructing images from the CIFAR-10 dataset. The autoencoder is trained using a combination of perceptual loss and Mean Squared Error (MSE) loss, providing efficient compression while preserving the quality of reconstructed images.
## Features
- **Efficient Compression**: Utilizes a convolutional autoencoder to significantly reduce the size of CIFAR-10 images.
- **High-Quality Reconstructions**: Combines perceptual loss and MSE loss to maintain the visual quality of reconstructed images.
- **Data Augmentation**: Implements advanced techniques to enhance the model's robustness and performance.
- **GPU Support**: Optimized for training on GPUs using TensorFlow, accelerating the training process.
- **Modularity**: Organized codebase with modular components, facilitating easy contributions and extensions.
## Compression Metrics and Results
### Compression Performance
- **Original Size**: 61,440 bytes per image
- **Compressed Size**: 5,120 bytes per image
- **Compression Ratio**: 12.00
### Explanation of Results
The **compression ratio** of **12.00** indicates that each image is compressed to one-twelfth of its original size. This substantial reduction in size demonstrates the effectiveness of the convolutional autoencoder in minimizing storage requirements without compromising the quality of the images.
**Why This Model Excels:**
- **Balanced Loss Functions**: By leveraging both perceptual loss and MSE loss, the model ensures that reconstructed images retain essential visual features and textures, providing a balance between compression efficiency and image fidelity.
- **Advanced Architecture**: The convolutional layers in the autoencoder are adept at capturing spatial hierarchies and patterns in images, enabling effective compression.
- **Data Augmentation**: Enhancing the training data with augmentation techniques makes the model more robust and improves its generalization capabilities.
- **Optimized Training**: Utilizing GPU acceleration with TensorFlow significantly speeds up the training process, allowing for faster iterations and model improvements.
Overall, **CIFAR10-CompressAI** offers a powerful solution for image compression tasks, achieving high compression ratios while maintaining the quality of the original images.
## Project Structure
```
CIFAR10-CompressAI/
├── data/ # Folder for data
├── models/ # Storage for trained models and images
├── notebooks/ # Jupyter notebooks for exploration
├── src/ # Source code
│ ├── data_preprocessing.py
│ ├── models.py
│ ├── train.py
│ ├── evaluate.py
├── .gitignore # Files and folders to ignore by Git
├── README.md # Project documentation
├── requirements.txt # Project dependencies
├── LICENSE # Project license
├── CONTRIBUTING.md # Contribution guide
```
## Installation
1. **Clone the Repository**
```bash
git clone https://github.com/pierridotite/CIFAR10-CompressAI.git
cd CIFAR10-CompressAI
```
2. **Create a Virtual Environment (Optional but Recommended)**
```bash
python -m venv venv
venv\Scripts\activate # On Windows
source venv/bin/activate # On macOS/Linux
```
3. **Install Dependencies**
```bash
pip install -r requirements.txt
```
## Usage
### Training the Model
To train the autoencoder, run:
```bash
python src/train.py
```
### Evaluating the Model
To evaluate and compare the original and reconstructed images, run:
```bash
python src/evaluate.py
```
## Contribution
Contributions are welcome! Please refer to the [CONTRIBUTING.md](CONTRIBUTING.md) file for more details.
## License
This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for more details.
## Examples
### Training
### Compression Comparison
## Advanced Usage
You can explore the notebooks in the `notebooks/` folder for additional analyses and visualizations.
---
Thank you for using **CIFAR10-CompressAI**! Feel free to contribute and share this project with the community.