Real-Time-Semantic-Segmentation-Project

This project implements a real-time semantic segmentation system using a deep learning model. The system captures live video feed from a camera and performs pixel-wise classification to identify various objects and their boundaries in the scene.

Setup and Installation

1. Dependencies:

   • Python 3.x
   • OpenCV
   • PyTorch
   • PIL
   • torchvision
   • scipy
   • numpy

2. Installing Required Libraries:

pip install opencv-python-headless torch torchvision Pillow numpy scipy

3. Model Weights: Download the model weights (encoder_epoch.pth and decoder_epoch.pth) from the provided link and place them in the project directory.

4. Running the Application:

• Execute the main script to start the real-time semantic segmentation:

  python main_script.py
  

Usage

• The application opens a window displaying the live camera feed.
• Press Space to toggle real-time mode.
• Press TAB to toggle between different visualizations.
• Press 1-9 or a-f to choose specific classes for segmentation visualization.
• Press Esc to exit the application.
• Press s to save the current frame and segmentation result.

Project Purpose

The purpose of this project is to demonstrate the capabilities of semantic segmentation in real-time applications. It can be used for educational purposes or as a base for more complex computer vision projects.

Credits and Acknowledgments

• Model training and architecture are based on CSAILVision's Semantic Segmentation PyTorch.
• Data used: ADE20K MIT Scene Parsing Benchchmark.