I was wondering whether the authors of the paper have considered using spherical linear interpolation to combine the noise lens with the de-noises sequence. Theoretically, spherical linear interpretation might lead to a smoother blending of the latents as it respects the norm of the latent vectors.
I was wondering whether the authors of the paper have considered using spherical linear interpolation to combine the noise lens with the de-noises sequence. Theoretically, spherical linear interpretation might lead to a smoother blending of the latents as it respects the norm of the latent vectors.