Deep Depth Completion of a Single RGB-D Image
This work addresses a domain-specific problem for users of commodity-depth cameras, such as in robotics or augmented reality, by providing an incremental improvement over existing depth completion techniques.
The paper tackles the problem of completing missing depth values in RGB-D images, which often occur for shiny, bright, transparent, and distant surfaces, by training a deep network to predict surface normals and occlusion boundaries from RGB input and combining them with raw depth data to fill holes, achieving better depth completions than alternative methods in experiments.
The goal of our work is to complete the depth channel of an RGB-D image. Commodity-grade depth cameras often fail to sense depth for shiny, bright, transparent, and distant surfaces. To address this problem, we train a deep network that takes an RGB image as input and predicts dense surface normals and occlusion boundaries. Those predictions are then combined with raw depth observations provided by the RGB-D camera to solve for depths for all pixels, including those missing in the original observation. This method was chosen over others (e.g., inpainting depths directly) as the result of extensive experiments with a new depth completion benchmark dataset, where holes are filled in training data through the rendering of surface reconstructions created from multiview RGB-D scans. Experiments with different network inputs, depth representations, loss functions, optimization methods, inpainting methods, and deep depth estimation networks show that our proposed approach provides better depth completions than these alternatives.