Equivariant Neural Rendering

learned representation transforms like a real 3D scene

• require relative transformations
• train in a style transfer way