Sun-sky model estimation from outdoor images

When a virtual object is inserted into an outdoor image, the recovery of scene illumination has a critical effect on the mix of virtual objects and actual reality. There are two main parts of the object in the outdoor scene: the sun and the sky. In order to represent the illumination conditions of these two natural illumination, this paper uses the Lalonde-Matthew outdoor illumination model to perform the sky and sun in the image. Model use seven parameters represent the illumination of the scene. So the original illumination estimation problem is transformed into a prediction problem of seven illumination parameters. For this problem, this paper proposes a new two-branch network structure, one branch is used to estimate the sun orientation, and the other branch is used to estimate the remaining six parameters. This paper also introduces convolution block attention module (CBAM) based on this structure. The introduction of this module enables the network to select the most important information for the current task target from a large number of information when extracting image features, while suppressing other useless information.