KVALD: Kalman Vision-based Automated Local Dimming

KVALD is a real-time video processing application designed to reduce glare and stabilize brightness in video streams. It uses a combination of a deep learning-based mask generator and a differentiable Kalman filter to achieve this.

Table of Contents

• Architecture
• Running the Application

Architecture

The core of KVALD consists of the following components:

1. Mask Generation Network: A U-Net based model that generates a probability mask to identify glare hotspots.
2. Differentiable Kalman Filter: A Kalman filter that uses the generated mask to predict and update the state of the video stream, smoothing out brightness variations.
3. Spatial and Temporal Smoothing: Additional smoothing algorithms to further improve the quality of the output video.

Note on Architecture: This project has been refactored to be a pure Python implementation, removing the C++ components for simplicity and faster prototyping.

Running the Application

To run this project, you will need Python and the dependencies listed in requirements.txt. You can install them using pip:

pip install -r requirements.txt

You can then run the main application:

python proof_of_concept.py