About Encoding Annotations

[jetd1]

Thanks for releasing this great work! Could you provide more details on how your encode the annotations (such as depth) with the pretrained VAE? Does the model require any fine-tuning / additional training in order to map the annotation into the latent space.

[jingheya]

Sorry for my late reply. We do not fine-tune the pre-trained VAE and only adapt the annotations into 3-channel input, similar to Marigold and GeoWizard. Depth maps are repeated to 3 channels, surface normal maps (which originally are 3-channels) remain unchanged.