This project involves implementing semantic segmentation using the U-Net model, specifically designed for person segmentation tasks.
The U-Net architecture is trained on a dataset of person images using TensorFlow, enabling precise identification and segmentation of people in images.
In addition to training the model, the project integrates OpenCV to create a live segmentation system.
This system processes live video streams, applying the segmentation model to detect persons in real-time, allowing for background modification or blurring effects.
The project demonstrates practical applications such as background replacement in video calls or privacy-enhancing video processing.
1. Train a U-Net model using TensorFlow on a person dataset to achieve high-precision person segmentation.
2. Apply semantic segmentation techniques to accurately differentiate persons from the background in images.
3. Utilize OpenCV to build a real-time person segmentation system capable of processing live video feeds.
4. Implement live background modification or blurring of segmented persons in real-time video streams.
5. Test the accuracy and efficiency of the live segmentation system in various scenarios.
6. Develop practical applications for person segmentation, such as privacy-focused video processing and dynamic background replacement.
