I would like to ask regarding the initialization of the network in training. In the paper, you mentioned both base-0123 and refine-0123 are initialized from the Zero123-XL weights. However, in the README file, you mention the pretrained_model_name_or_path is lambdalabs/sd-image-variations-diffusers, which means you basically train the both base-0123 and refine-0123 from scratch.
Could you please clarify this? If you initialize from "kxic/zero123-xl" in diffusers, could you please endorse it and update in the README?
Hi,
thanks for open sourcing the great work!
I would like to ask regarding the initialization of the network in training. In the paper, you mentioned both
base-0123andrefine-0123are initialized from the Zero123-XL weights. However, in the README file, you mention thepretrained_model_name_or_pathislambdalabs/sd-image-variations-diffusers, which means you basically train the bothbase-0123andrefine-0123from scratch.Could you please clarify this? If you initialize from
"kxic/zero123-xl"indiffusers, could you please endorse it and update in the README?Best, Yuxuan