I really liked the overall results compared to other open source solutions I saw. But for some images I tested it really seems the zero123++ views are bad, compared to other options (such as sv3d).
So I wanna confirm in a high level how the integration can be done. Does you solution requires a specific quantity of images in specific angles? Or can we use the 21 frames sv3d generates?
Our sparse-view large recconstruction model was trained using free-viewpoint images, thus it can be integrated with arbitrary multiview diffusion model in theory.
I really liked the overall results compared to other open source solutions I saw. But for some images I tested it really seems the zero123++ views are bad, compared to other options (such as sv3d).
So I wanna confirm in a high level how the integration can be done. Does you solution requires a specific quantity of images in specific angles? Or can we use the 21 frames sv3d generates?