CNN Vision Explorer 🧠

An interactive Streamlit application to visualize and understand the core components of Convolutional Neural Networks (CNNs).

---

🌐 Live Demo

Access the deployed application here: https://cnn-vision-explorer.streamlit.app

---

Creator Details

• Created by: Vishav Gupta
• Roll No.: 102497018
• SubGroup: 3Q2F
• Branch: Computer Science & Engineering
• Course: UCS668 - EDGE AI: Data Center Vision

---

✨ Features

This application is divided into four distinct modules:

1. 🖼️ Augmentation Sandbox:

   • Interactively apply various image augmentation techniques (rotation, brightness, blur, crop, etc.) to an uploaded image.
   • See a real-time side-by-side comparison of the original and augmented images.
   • Generates the corresponding PyTorch code snippet for the selected transformations.

2. 📈 Activation Function Lab:

   • Visualize common activation functions (ReLU, Sigmoid, Tanh, Leaky ReLU) and their derivatives.
   • Use the interactive "Single Neuron Simulator" to see how input, weight, and bias affect a neuron's output and gradient, demonstrating concepts like vanishing gradients and the dying ReLU problem.

3. 👁️ CNN Architecture Inspector:

   • Upload an image and select a pre-trained CNN model (VGG16, ResNet50, MobileNetV2).
   • Explore the model's architecture and choose specific convolutional layers.
   • Visualize the top activated feature maps from the selected layer to understand hierarchical feature extraction (edges -> textures -> parts -> objects).
   • Includes model-specific explanations of what features are typically learned at different depths.

4. 🧠 Classifier's Decision (XAI):

   • See the model's top 5 predictions for your uploaded image, displayed with confidence scores in a bar chart.
   • Implements Grad-CAM (Gradient-weighted Class Activation Mapping) to generate a heatmap overlay.
   • This heatmap visually explains which parts of the image the model focused on to make its top prediction, providing Explainable AI (XAI).

---

🛠️ Tech Stack

• Framework: Streamlit
• Deep Learning: PyTorch, Torchvision
• Plotting: Plotly, Matplotlib
• Image Processing: OpenCV, Pillow
• Numerical: NumPy
• UI/Navigation: streamlit-option-menu

---

⚙️ Setup

1. Clone the repository:

   git clone https://github.com/VishavGupta01/cnn-vision-explorer
   cd cnn-vision-explorer

2. Create a virtual environment (Recommended):

   python -m venv venv
   # On Windows
   venv\Scripts\activate
   # On macOS/Linux
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

---

▶️ Running the App

1. Navigate to the project directory in your terminal.
2. Run the Streamlit application:

   streamlit run app.py

3. The application should automatically open in your web browser.

---

Project Structure

CNN_Vision_Explorer/
│
├── .streamlit/
│   └── config.toml         # Streamlit theme configuration
│
├── assets/
│   └── style.css           # Custom CSS
│
├── modules/
│   ├── module_1.py         # Code for Augmentation Sandbox
│   ├── module_2.py         # Code for Activation Function Lab
│   ├── module_3.py         # Code for CNN Architecture Inspector
│   └── module_4.py         # Code for Classifier's Decision (XAI)
│
├── app.py                  # Main application script with navbar & routing
├── requirements.txt        # Python dependencies
└── README.md               # This file

---