Scene Representation Networks: Continuous 3D-Structure-Aware Neural\n Scene Representations

Unsupervised learning with generative models has the potential of discovering\nrich representations of 3D scenes. While geometric deep learning has explored\n3D-structure-aware representations of scene geometry, these models typically\nrequire explicit 3D supervision. Emerging neural scene representations can be\ntrained only with posed 2D images, but existing methods ignore the\nthree-dimensional structure of scenes. We propose Scene Representation Networks\n(SRNs), a continuous, 3D-structure-aware scene representation that encodes both\ngeometry and appearance. SRNs represent scenes as continuous functions that map\nworld coordinates to a feature representation of local scene properties. By\nformulating the…