A PyTorch implementation for extracting deep features from pre-trained ResNet50 and visualizing them using t-SNE dimensionality reduction technique. This project is particularly useful for understanding learned representations and analyzing feature distributions across different classes.
resnet50-tsne-visualization/
│
├── README.md # Project documentation (this file)
├── requirements.txt # Python dependencies
├── LICENSE # MIT License
├── .gitignore # Git ignore rules
├── setup.py # Package setup file
│
├── src/ # Source code directory
│ ├── __init__.py # Package initializer
│ ├── feature_extractor.py # Main feature extraction class
│ ├── model.py # ResNet50 model definition
│ └── utils.py # Utility functions
│
├── configs/ # Configuration files
│ └── default_config.yaml # Default configuration
│
├── models/ # Pre-trained model weights
│ ├── .gitkeep # Placeholder for directory
│ └── README.md # Instructions for model files
│
├── data/ # Dataset directory
│ ├── .gitkeep # Placeholder for directory
│ └── README.md # Data preparation instructions
│
├── results/ # Output visualizations
│ ├── .gitkeep # Placeholder for directory
│ └── sample_results/ # Sample visualization results
│ └── tsne_cifar10.jpg # Example t-SNE visualization
│
└── scripts/ # Utility scripts
├── download_pretrained.py # Download pre-trained weights
└── prepare_data.py # Data preparation script
- 🚀 Feature Extraction: Extract high-dimensional features from any layer of ResNet50
- 📊 t-SNE Visualization: Reduce dimensions and visualize feature distributions
- 🎨 Customizable Plots: Beautiful, publication-ready visualizations with color-coded classes
- 📦 Modular Design: Clean, reusable code structure with OOP design
- 🔧 Configurable: Easy to adjust parameters via configuration files
- 📈 Batch Processing: Efficient processing of large datasets
- 💾 Checkpoint Support: Save and load extracted features for faster experimentation
- Clone the repository
git clone https://github.com/dcx1110/resnet50-tsne-visualization.git
cd resnet50-tsne-visualization- Install dependencies
pip install -r requirements.txt- Download pre-trained weights (optional)
python scripts/download_pretrained.pyfrom src.feature_extractor import FeatureExtractor
# Initialize the feature extractor
extractor = FeatureExtractor(
model_path='./models/resNet50.pth', # Path to pre-trained weights
data_path='./data/cifar10', # Path to your dataset
batch_size=128, # Batch size for processing
num_classes=10, # Number of classes
device='cuda' # Use 'cpu' if GPU not available
)
# Run the complete pipeline
tsne_features, labels = extractor.run(
save_path='./results/my_visualization.jpg'
)
# The extracted features and labels are also returned for further analysis
print(f"t-SNE features shape: {tsne_features.shape}")
print(f"Labels shape: {labels.shape}")from src.feature_extractor import FeatureExtractor
# Create extractor with custom configuration
extractor = FeatureExtractor(
model_path='./models/resNet50_custom.pth',
batch_size=256,
num_classes=100 # For CIFAR-100 or custom dataset
)
# Step-by-step execution for more control
dataloader = extractor.load_data(split='val')
model = extractor.load_model()
# Extract features from a specific layer (optional)
features, labels = extractor.extract_features(model, dataloader)
# Apply t-SNE with custom parameters
tsne_features = extractor.apply_tsne(
features,
n_components=2,
perplexity=30,
random_state=42
)
# Create custom visualization
extractor.visualize_tsne(
tsne_features,
labels,
save_path='./results/custom_tsne.jpg',
figsize=(15, 15),
point_size=10
)The following image shows a t-SNE visualization of ResNet50 features extracted from CIFAR-10 dataset:
Each color represents a different class, and the spatial proximity indicates feature similarity learned by the model.
You can modify the default configuration in configs/default_config.yaml:
model:
architecture: resnet50
num_classes: 10
pretrained_path: ./models/resNet50.pth
data:
dataset: cifar10
data_root: ./data
batch_size: 128
num_workers: 4
tsne:
n_components: 2
perplexity: 30
random_state: 42
n_iter: 1000
visualization:
figsize: [20, 20]
dpi: 300
colormap: jet-
FeatureExtractorClass: The core class that handles the entire pipelineload_data(): Loads and preprocesses the datasetload_model(): Initializes and loads the pre-trained ResNet50extract_features(): Extracts features from the modelapply_tsne(): Applies t-SNE dimensionality reductionvisualize_tsne(): Creates and saves the visualization
-
Data Processing:
- Automatic data augmentation for training
- Normalization using ImageNet statistics
- Efficient batch processing with DataLoader
-
Feature Extraction:
- Extracts features from the final layer before classification
- Supports extraction from intermediate layers
- Handles GPU/CPU computation automatically
-
Visualization:
- Color-coded points for different classes
- Customizable plot parameters
- High-resolution output for publications
- Python 3.7+
- PyTorch 1.9.0+
- torchvision 0.10.0+
- NumPy 1.19.0+
- Matplotlib 3.3.0+
- scikit-learn 0.24.0+
- tqdm 4.62.0+
- Pillow 8.0.0+
- PyYAML 5.4.0+ (for configuration files)
Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.
- Fork the repository
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -m 'Add some AmazingFeature') - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
This project is licensed under the MIT License - see the LICENSE file for details.
- ResNet50 architecture from Deep Residual Learning for Image Recognition
- t-SNE algorithm from Visualizing Data using t-SNE
- CIFAR-10 dataset from Learning Multiple Layers of Features from Tiny Images
If you find this project useful, please consider giving it a star ⭐️
Made with ❤️ by [Your Name]