This talk discusses methodologies for reconstructing illumination information of real scenes from images. Three techniques are presented that enable illumination capturing and radiometric reconstruction. The first approach utilizes specialized hardware to capture wide-angle images of the real scene and subsequent image processing to identify light sources. The second method enables mobile capturing of omnidirectional scene illumination in high dynamic range using standard mobile phone camera and accumulation of angular light information. Finally, the third approach uses deep learning to train a network for illumination estimation from single image of a real scene. The presented methods for illumination estimation can be used in various application areas including augmented reality, scene understanding, and 3D scene relighting.