Practical SVBRDF Acquisition of 3D Objects with Unstructured Flash Photography

摘要

Capturing spatially-varying bidirectional reflectance distribution functions(SVBRDFs) of 3D objects with just a single, hand-held camera (such as anoff-the-shelf smartphone or a DSLR camera) is a difficult, open problem.Previous works are either limited to planar geometry, or rely on previouslyscanned 3D geometry, thus limiting their practicality. There are severaltechnical challenges that need to be overcome: First, the built-in flash ofa camera is almost colocated with the lens, and at a fixed position; thisseverely hampers sampling procedures in the light-view space. Moreover,the near-field flash lights the object partially and unevenly. In terms ofgeometry, existing multiview stereo techniques assume diffuse reflectanceonly, which leads to overly smoothed 3D reconstructions, as we show inthis paper. We present a simple yet powerful framework that removes theneed for expensive, dedicated hardware, enabling practical acquisition ofSVBRDF information from real-world, 3D objects with a single, off-the-shelfcamera with a built-in flash. In addition, by removing the diffuse reflectionassumption and leveraging instead such SVBRDF information, our methodoutputs high-quality 3D geometry reconstructions, including more accuratehigh-frequency details than state-of-the-art multiview stereo techniques.We formulate the joint reconstruction of SVBRDFs, shading normals, and3D geometry as a multi-stage, iterative inverse-rendering reconstructionpipeline. Our method is also directly applicable to any existing multiview 3D reconstruction technique. We present results of captured objects withcomplex geometry and reflectance; we also validate our method numericallyagainst other existing approaches that rely on dedicated hardware, additionalsources of information, or both.