This repository hosts a state-of-the-art AI/ML-based desmoking and de-hazing algorithm designed to enhance the visibility of videos in real-time. This technology is particularly useful in scenarios involving fire and heavy smoke, providing clearer visuals that can be crucial for rescue operations and safety monitoring.
- Real-Time Video Processing: Delivers immediate enhancement of video frames.
- Advanced AI/ML Algorithms: Leverages cutting-edge machine learning techniques to effectively remove smoke and haze.
- High Performance: Optimized for efficient processing on standard hardware.
- Flexible Input Formats: Compatible with various video sources and formats.
- High-Quality Output: Maintains detail and clarity while improving visibility.
The desmoking/de-hazing algorithm is based on a convolutional neural network (CNN) trained on datasets containing smoky and clear images. The training process involves:
- Data Collection: Assembling a diverse dataset of images and videos with varying smoke and haze levels.
- Preprocessing: Normalizing and augmenting the dataset to enhance model robustness.
- Training: Using supervised learning to minimize the discrepancy between predicted clear images and ground truth.
- Evaluation: Testing the model on unseen data to ensure performance and generalization.