Joint Deep Learning and Information Propagation for Fast 3D City Modeling

In the field of geoinformation science, multiview, image-based 3D city modeling has developed rapidly, and image depth estimation is an important step in it. To address problems poor adaptability training models existing neural network methods long reconstruction time traditional geometric methods, we propose a general method for fast that combines prior knowledge information propagation. First, original downsampled input into to predict initial value. Then, plane fitting joint optimization are combined with superpixel optimized value upsampled resolution. Finally, propagation checked pixel-by-pixel obtain final estimate. Experiments were conducted using multiple datasets taken from actual indoor outdoor scenes. Our was compared analyzed variety widely used methods. The experimental results show our maintains high accuracy speed, it achieves better performance. This paper offers framework integrate networks which provide new approach obtaining geographic modeling.