self.latent_scaling = self.latent_scaling / (self.latent_scaling - 1) * 2.0

[MichaleGo]

In encoder.py, line 163, self.latent_scaling = self.latent_scaling / (self.latent_scaling - 1) * 2.0

self.latent_scaling should bet a scale factor around 2, this makes no sense, because in line 98 scale = self.latent_scaling / image_size,

self.latent_scaling should be a absolute resolution number, I think they are contraditory.

[MichaleGo]

Another question, Why is there an additional minus one here? uv = uv * scale - 1.0

[primecai]

Hi,

I'm wondering if you figured it out? I do not quite understand what this step is doing at all. If the input uv is already in [-1, 1] as mentioned here, why do we even need these latent scaling things?

Thanks in advance.