Ma and Ma [31] used Faster R-CNN, while Sun and Chen [32] used YOLOv3 to train their respective neural network models for window feature extraction and object recognition. Although their models can detect window objects well, such region-based recognition approaches ignore the rich geometric ...