A training objective that measures how well a reconstructed 3D scene matches the appearance of the original images when rendered.