set its intensity with a brightness adjustment. This new process is fundamentally different from the traditional one because the capture session is not concerned with directly producing a visually pleasing image, it only seeks to record useful data for the later editing step. From this perspective, it is related to recent work in computational photography such as coded apertures, e.g., [Levin et al. 2007; Bishop and Favaro 2011], and lightfield cameras, e.g., [Ng et al. 2005; Adelson and Bergen 1991], in which computation is an integral part of the image formation process. This motivates us to name
this modern approach computational lighting design. See [Kelley 2011; Guanzon and Blake 2011] for examples of this workflow from professional photographers that inspired us (unaffiliated with the project). These examples demonstrate the use of this workflow in architectural photography, and product photography. However, one major disadvantage of this workflow is that it is quite cumbersome even for experienced photographers. When the number of images grows to several tens, or even above a hundred, navigating the corresponding layer stack becomes impractical. Further, with
large scenes, most images show the main subject mostly in the dark with only a small part lit (see Figure 1a). Finding the useful images in the stack, setting their relative intensities, and blending them together to get visually pleasing results are highly challenging tasks that require advanced photography and image editing skills.