Abstract
Recent advancements in open-world 3D object generation have been remarkable, with image-to-3D methods offering superior fine-grained control over their text-to-3D counterparts. However, most existing models fall short in simultaneously providing rapid generation speeds and high fidelity to input images - two features essential for practical applications. In this paper, we present One-2-3-45++, an innovative method that transforms a single image into a detailed 3D textured mesh in approximately one minute. Our approach aims to fully harness the extensive knowledge embedded in 2D diffusion models and priors from valuable yet limited 3D data. This is achieved by initially fine-tuning a 2D diffusion model for consistent multi-view image generation, followed by elevating these images to 3D with the aid of multi-view conditioned 3D native diffusion models. Extensive experimental evaluations demonstrate that our method can produce high-quality, diverse 3D assets that closely mirror the original input image.
Project Page: https://sudo-ai-3d.github.io/One2345plus_page/
Paper: https://arxiv.org/pdf/2311.07885.pdf
I don’t know about immediate applicability for generating images, but I think that in the long run, models are going to have to be based on 3d representations. That’s what humans do, after all.
Doing 2d is enormously limiting – you not making effective use of a lot of training data, since you can’t combine much knowledge about an object from different orientations.