Shallow depth of field (sdof) rendering

Abstract

This disclosure relates to the lightweight and efficient synthesis of shallow depth of field (SDOF) renderings. According to some embodiments, coarse focus information and semantic segmentation information may be leveraged to generate SDOF renderings in a 'live preview' or 'streaming' mode. Semantic segmentation may be defined as a process of creating a mask over an image, wherein pixels are segmented into a predefined set of semantic classes. Segmentations may include as many classes as are desired by a given implementation (e.g., a 'foreground' class and a 'background' class). In some embodiments, the rendering of synthetic SDOF images according to the techniques described herein may be completed using only a single camera of a device and without the use of dedicated depth-sensing technology (e.g., structured light cameras, stereo cameras, time-of-flight cameras, etc.), thereby allowing the SDOF rendering process to operate in a fashion that is not unduly time-, processing-, and/or power-intensive.