Free-form scanning of non-planar appearance with neural trace photography

We propose neural trace photography, a novel framework to automatically learn high-quality scanning of non-planar, complex anisotropic appearance. Our key insight is that free-form appearance can be cast as geometry learning problem on unstructured point clouds, each which represents an image measurement and the corresponding acquisition condition. Based this connection, we carefully design network, jointly optimize lighting conditions used in acquisition, well spatially independent reconstruction reflectance from measurements. not tied specific setup, adapt various factors data-driven manner. demonstrate effectiveness our number physical objects with wide variation The are captured light-weight mobile device, consisting single camera RGB LED array. also generalize other common types light sources, including point, linear area light.