In the Limitation, ". Our initial experiments with a naive secondary reconstruction loss did not improve performance, but we believe a more sophisticated formulation to encourage the diffusion process to be aware of the surface being represented would improve its generative modeling capabilities". What is the meaning of the naive secondary reconstruction loss?
In the Limitation, ". Our initial experiments with a naive secondary reconstruction loss did not improve performance, but we believe a more sophisticated formulation to encourage the diffusion process to be aware of the surface being represented would improve its generative modeling capabilities". What is the meaning of the naive secondary reconstruction loss?