Semantic segmentation with OpenCV and deep learning