https://github.com/nevrohelios/deep-viz
🔬 Deep-Viz: Unveiling the Black Box of Deep Learning
https://github.com/nevrohelios/deep-viz
captum opencv pytorch streamlit torchvision
Last synced: 3 months ago
JSON representation
🔬 Deep-Viz: Unveiling the Black Box of Deep Learning
- Host: GitHub
- URL: https://github.com/nevrohelios/deep-viz
- Owner: NevroHelios
- Created: 2025-06-25T07:54:49.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2025-06-26T19:09:28.000Z (12 months ago)
- Last Synced: 2025-06-26T20:20:37.444Z (12 months ago)
- Topics: captum, opencv, pytorch, streamlit, torchvision
- Language: Python
- Homepage:
- Size: 8.79 KB
- Stars: 1
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🔬 Deep-Viz: Unveiling the Black Box of Deep Learning
*"Seeing is believing, but understanding is enlightening"* 🧠✨
## 🌟 What is Deep-Viz?
Deep-Viz is an interactive web application that makes AI interpretability accessible to everyone! Ever wondered what a neural network "sees" when it looks at an image? This tool pulls back the curtain on deep learning models, revealing the hidden patterns and decision-making processes that drive AI predictions.
## 🎯 Features
### 🔍 **Model Interpretability Made Simple**
- **Class Activation Maps (CAM)** using SmoothGradCAM++ 🎨
- **Integrated Gradients** for pixel-level feature attribution 📊
- **Real-time visualization** of model predictions 🚀
- **Multi-model support** (ResNet18 & ResNet50) ⚙️
### 🎨 **Interactive Experience**
- **Drag-and-drop** image upload 📤
- **Layer-wise exploration** of neural network activations 🔬
- **Top-5 predictions** with confidence scores 📈
- **Side-by-side comparison** of different visualization techniques 🔄
## 🧠 How It Works
```mermaid
graph TD
A[📷 Upload Image] --> B[🔄 Preprocessing]
B --> C[🤖 ResNet Model]
C --> D[📊 Predictions]
C --> E[🎨 SmoothGradCAM++]
C --> F[🔍 Integrated Gradients]
D --> G[📈 Top-5 Results]
E --> H[🗺️ Class Activation Map]
F --> I[⚡ Attribution Heatmap]
G --> J[📱 Interactive Dashboard]
H --> J
I --> J
style A fill:#ff6b6b
style C fill:#4ecdc4
style J fill:#45b7d1
```
## 🚀 Quick Start
### Prerequisites
Make sure you have Python 3.7+ installed on your system.
### Installation
1. **Clone the repository**
```bash
git clone https://github.com/nevrohelios/deep-viz.git
cd deep-viz
```
2. **Install dependencies**
```bash
pip install -r requirements.txt
```
3. **Launch the application**
```bash
streamlit run main.py
```
4. **Open your browser** and navigate to `http://localhost:8501` 🌐
## 📚 Understanding the Visualizations
### 🎨 Class Activation Maps (CAM)
CAMs highlight the regions in an image that are most important for the model's prediction. Think of it as the model's "attention map" - where is it looking when making decisions?
```
🖼️ Original Image → 🔥 Heat Map → 🎯 Important Regions
```
### ⚡ Integrated Gradients
This technique provides pixel-level attributions, showing how much each pixel contributes to the final prediction. It's like having X-ray vision into the model's decision process!
```
🔍 Pixel Analysis → 📊 Attribution Scores → 🎨 Visualization
```
## 🔧 Technical Architecture
### 📦 Core Components
| Component | Purpose | Technology |
|-----------|---------|------------|
| 🎨 **Frontend** | Interactive UI | Streamlit |
| 🤖 **Model Backend** | Deep Learning | PyTorch + TorchVision |
| 🔍 **Interpretability** | CAM Generation | TorchCAM |
| ⚡ **Attribution** | Gradient Analysis | Captum |
| 🖼️ **Image Processing** | Preprocessing | PIL + Transforms |
### 🏗️ Model Architecture
```
Input Image (224×224×3)
↓
ResNet18/50 Backbone
↓
Feature Extraction
↓
┌─────────────────┬─────────────────┐
│ CAM Extraction │ Gradient Flow │
│ (TorchCAM) │ (Captum) │
└─────────────────┴─────────────────┘
↓
Visualization Pipeline
↓
Interactive Dashboard
```
## 🎮 Usage Examples
### 🐕 Example 1: Animal Classification
Upload a photo of your pet and see which features the model uses to identify the animal!
### 🏞️ Example 2: Scene Understanding
Try landscape photos to understand how the model recognizes different environments.
### 🚗 Example 3: Object Detection
Upload images with vehicles to see how the model focuses on distinctive features.
## 🔬 Advanced Features
### 🎛️ **Layer Selection**
Explore different layers of the neural network to understand how features evolve from simple edges to complex patterns:
- **Early Layers**: Edge detection, basic shapes 📐
- **Middle Layers**: Textures, patterns 🌀
- **Deep Layers**: Complex objects, semantic features 🏗️
### 📊 **Model Comparison**
Switch between ResNet18 and ResNet50 to see how model depth affects interpretation:
| Model | Parameters | Speed | Accuracy |
|-------|------------|-------|----------|
| ResNet18 | 11.7M | ⚡ Fast | 🎯 Good |
| ResNet50 | 25.6M | 🐌 Slower | 🎯 Better |
## 🔮 What Makes This Special?
### 🎨 **Visual Excellence**
- Clean, intuitive interface designed for both beginners and experts
- Real-time processing with smooth animations
- High-quality visualizations that reveal model insights
### 🧪 **Educational Value**
- Perfect for students learning about AI interpretability
- Demonstrates cutting-edge explainable AI techniques
- Bridges the gap between theory and practice
### 🚀 **Production Ready**
- Robust error handling and user feedback
- Optimized for performance
- Extensible architecture for adding new models
## 🛠️ Dependencies
```python
# Core ML Framework
torch # PyTorch deep learning framework
torchvision # Computer vision utilities
torchcam # Class Activation Mapping
# Interpretability
captum # Model interpretability library
# Web Framework
streamlit # Interactive web applications
# Image Processing
pillow # Python Imaging Library
numpy # Numerical computing
matplotlib # Plotting library
```
## 🔄 Future Enhancements
- [ ] 🎯 **More Models**: Add support for Vision Transformers, EfficientNet
- [ ] 🎨 **Advanced Visualizations**: GradCAM, LIME, SHAP
- [ ] 📱 **Mobile Support**: Responsive design for mobile devices
- [ ] 🔗 **API Integration**: REST API for programmatic access
- [ ] 🎥 **Video Analysis**: Extend to video classification
- [ ] 🌐 **Multi-language**: Support for multiple languages
## 🤝 Contributing
We welcome contributions! Whether it's:
- 🐛 Bug fixes
- ✨ New features
- 📚 Documentation improvements
- 🎨 UI/UX enhancements
Please feel free to open issues and pull requests!
## 📄 License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
## 🙏 Acknowledgments
- **PyTorch Team** for the amazing deep learning framework 🔥
- **Streamlit** for making web apps incredibly simple 🚀
- **Captum** for state-of-the-art interpretability tools 🔍
- **TorchCAM** for excellent CAM implementations 🎨
---
**Made with ❤️ and lots of ☕**
*Star ⭐ this repo if you found it helpful!*