Supplementary Material for “Dense Non-Rigid Structure-from-Motion and Shading with Unknown Albedos”

In this document, we provide supplementary information about the ‘boundariness map’ Et computation for the boundary contour term. This term encourages the boundary of the surface to project to strong intensity edges in the image and it is a very useful constraint for poorly-textured surfaces. However, boundary may be attracted by false surface edges, which corresponds to background clutter or surface texture. The segmentation cue to remove such false positives from the ‘boundariness map’ depends on the particular dataset. According to the dataset, we use two segmentation cues: the image projection of the estimated surface and the color information of the image. We define an intermediate ‘boundariness map’ et where strong image gradients are represented by a low value. This map acts such as a potential well. We compute it in two steps. First, we compute a blurred grayscale version of the image It using a Gaussian filter (h, σ). We denote it I (h,σ) t . Then, we use the following formula: