Perceptual deep depth super-resolution
This work addresses depth map upsampling for virtual reality and 3D reconstruction, offering a perceptual improvement over existing methods, though it is incremental in its approach.
The paper tackles the problem of depth map super-resolution by using color information to improve resolution, but this fusion can cause artifacts that affect 3D shape reconstruction for applications like virtual reality. They propose measuring quality through 3D surface renderings and show that a visual appearance-based loss with a CNN or deep prior yields significantly improved 3D shapes, as validated by perceptual metrics.
RGBD images, combining high-resolution color and lower-resolution depth from various types of depth sensors, are increasingly common. One can significantly improve the resolution of depth maps by taking advantage of color information; deep learning methods make combining color and depth information particularly easy. However, fusing these two sources of data may lead to a variety of artifacts. If depth maps are used to reconstruct 3D shapes, e.g., for virtual reality applications, the visual quality of upsampled images is particularly important. The main idea of our approach is to measure the quality of depth map upsampling using renderings of resulting 3D surfaces. We demonstrate that a simple visual appearance-based loss, when used with either a trained CNN or simply a deep prior, yields significantly improved 3D shapes, as measured by a number of existing perceptual metrics. We compare this approach with a number of existing optimization and learning-based techniques.