Recovering Real-World Reflectance Properties and Shading From HDR Imagery

Deep scene-scale material estimation from multi-view indoor captures

“…The only other methods designed to handle our input at scene-scale are differentiable rendering approaches [8,14]…”

“…The recent development of differentiable rendering algorithms has reinvigorated research on inverse rendering, as pioneered by Yu et al [7] for scene-scale material recovery. Given a 3D model of the scene, modern approaches estimate material and lighting by simulating complex global illumination effects using differentiable path-tracing [13,8,14]. We see our approach as complementary to such optimization-based algorithms that work on scene-scale.…”

“…We observe that it is possible to create renderable assets from a multi-view dataset with approximately 30 minutes to an hour of computation depending on number of images and their resolution. Previous works dealing with scene-scale material estimation [8,14] report around 10 − 12 hours for a complete scene optimization. Thus, our method is able to produce renderable material maps in a fraction of time as compared to previous works.…”