Unifying single-image lighting estimation

We present a method for automatically estimating the lighting conditions from a single image. As opposed to most previous works which proposed methods that deal with individual aspects of the problem (e.g. indoors vs outdoors, parametric vs non-parametric), the proposed method unifies these ideas into a single, coherent framework. Our method will automatically estimate both parametric (individual light sources) and non-parametric (environment maps) lighting representations from both indoor and outdoor images. To this end, we will develop a novel deep learning architecture to automatically learn the mapping between an input image to its lighting conditions in an end-to-end fashion, by relying on a large dataset of indoor and outdoor high dynamic range images for training. The method will be quantitatively evaluated and compared to the most relevant approaches from the literature.

Henrique Weber
Superviseur universitaire: 
Jean-Francois Lalonde
Partner University: