Image encoder clarification

[gbenms]

Image encoder clarification

[wl-zhao]

Hi,

Thanks for your careful reading. We use the kl-autoencoder for the first-stage model, the same as the text-to-image diffusion model in stable-diffusion. Perhaps because many other models in stable-diffusion (such as ffhq256, cin256, etc) adopt the VQGAN as the first-stage model, VQGAN is often used to indicate the first stage of a stable-diffusion model, despite the slight difference in actual implementation. We will modify it in the updated version of our paper to avoid misunderstanding.

Thanks,
Wenliang