Computer Vision News - August 2019

Another nice comparison is the qualitative results. As you can see below, the depth maps generate with the proposed method (last column) looks much more appealing than the other methods. And here are some more results: The first row is the input image, the second is the MVS ground truth, the third is the method's depth from a single image and the fourth is the full model. To sum up, the paper shows a novel technique to generate a depth map with neural nets. It is important to mention that this paper by Google Research has won the Best Paper, Honorable Mention at the latest CVPR. We know some of the team members and admire their work very much! Learning the Depths 7