Weird result when using this checkpoint with Zero123's live demo code
Hi, thank you for releasing this checkpoint!
Instead of directly using the model with threestudio, I wanted to test the novel view synthesis performance of this checkpoint.
I thought that the camera model remained the same as the original zero123 for this checkpoint.
However, when I tested on the dinosaur image, I got a weird result when I only changed the azimuth angle:
When using the zero123-xl checkpoint, things work as expected:
Is this an expected behavior? If so, how should I change my camera condition?
Best,
Ryu
To my knowledge, they changed the camera parameters little bit, its best to check threestudio provided inference code for running the stable-zero123
I checked out the inference code for stable-zero123 in the threestudio repo.
But, as mentioned in issue https://github.com/threestudio-project/threestudio/issues/360, the only difference in the camera conditioning between the original zero123 and stable-zero123 seems to be the camera distance. In stable-zero123, they just abnegated the relative distance.
Hi, thank you for releasing this checkpoint!
Instead of directly using the model with threestudio, I wanted to test the novel view synthesis performance of this checkpoint.
I thought that the camera model remained the same as the original zero123 for this checkpoint.
However, when I tested on the dinosaur image, I got a weird result when I only changed the azimuth angle:When using the zero123-xl checkpoint, things work as expected:
Is this an expected behavior? If so, how should I change my camera condition?
Best,
Ryu
You're welcome! Yes is an expected Behavior since : Model Details
Developed by: Stability AI
Model type: latent diffusion model.
Finetuned from model: lambdalabs/sd-image-variations-diffusers
License: StabilityAI Non-Commercial Research Community License. If you want to use this model for your commercial products or purposes, please contact us here to learn more.
Training Dataset
We use renders from the Objaverse dataset, utilizing our enhanced rendering method
Training Infrastructure
Hardware: Stable Zero123 was trained on the Stability AI cluster on a single node with 8 A100 80GBs GPUs.
Code Base: We use our modified version of the original zero123 repository.
Misuse, Malicious Use, and Out-of-Scope Use
The model should not be used to intentionally create or disseminate images that create hostile or alienating environments for people. This includes generating images that people would foreseeably find disturbing, distressing, or offensive; or content that propagates historical or current stereotypes. So you can tried to Training him