Equivariant Neural Rendering

Emilien Dupont, Miguel Angel Bautista, Alex Colburn +4

We propose a framework for learning neural scene representations directly from images, without 3D supervision. Our key insight is that 3D structure can be imposed by ensuring that…