I'm looking into implementing this architecture. Did you try to interpolate through specific style inputs instead of the entire latent space? If so, did you observe style inputs at different layers affecting higher or lower level features of the generated image? If so, how strong was this observation compared to what the StyleGAN paper showed?
I'm looking into implementing this architecture. Did you try to interpolate through specific style inputs instead of the entire latent space? If so, did you observe style inputs at different layers affecting higher or lower level features of the generated image? If so, how strong was this observation compared to what the StyleGAN paper showed?