cannot find a way to encode image

[rotem154154]

stylegan2 has (18,512) latent vector and mobileStyleGAN map (1,512) vector to (1,512) style.
i dont get it how can i encode the image to the (1,512) style that the model takes as input.
thanks for help

[bes-dev]

@rotem154154 MobileSynthesisNetwork supports W+ space directly. Just throw Tensor with shape [batch_size, w_plus_size, latent_size] to the MobileSynthesisNetwork.

[rotem154154]

@rotem154154 MobileSynthesisNetwork supports W+ space directly. Just throw Tensor with shape [batch_size, w_plus_size, latent_size] to the MobileSynthesisNetwork.

but i want to edit an image so the tensor of that shape come from encoder that takes an image as input

[bes-dev]

@rotem154154 you can convert W to W+ like here