Robust Point Light Source Estimation Using Differentiable Rendering

Differentiable Rendering: A Survey

“…For the local ones, the purpose is to approximate useful gradients, therefore the optimization results should be used. However, in both cases, the lack of a Application Literature 3D object reconstruction [9], [20], [7], [8], [67], [22], [23], [68], [37], [40], [42], [61], [69], [70], [29], [71], [72], [73], [44] Body shape estimation [11], [10] Hand shape estimation [13], [74], [75] Face reconstruction [76] Object/camera pose estimation [77] Object part segmentation [78] Material estimation [17] Light/shading estimation [79], [80], [81], [82] Adversarial examples [83], [84], [85], [86], [87] Auto-labeling [45] Teeth modeling [88] common dataset prevents a fair comparison of the different algorithms. One possible solution is to create a set of toy problems which can be used to evaluate the derivative or optimization of geometry, material, light, and camera parameters.…”

“…Several works focus on estimating the light sources from images. Nieto et al [79] try to minimize the photometric error between the rendered image and the observed one, given the RGB and depth images of a scene. Unlike most previous works, which do not take cast shadows into account when estimating the illumination, their method is based on the Blinn-Phong reflection model [46].…”