Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Generative AI and Creativity: A dialogue between machine learning researchers and creative professionals

An Object is Worth 64x64 Pixels: Generating 3D Object via Image Diffusion

Xingguang (Kevin) Yan · Han-Hung Lee · Ziyu Wan · Angel Chang

[ ] [ Project Page ]
Sat 14 Dec 1 p.m. PST — 2 p.m. PST

Abstract:

This paper presents "Object Images," a novel method that represents 3D shapes with UV-patch structures as 64x64 pixel images, encapsulating geometry and appearance details. By converting complex 3D shapes into a 2D format, we leverage image generation models like Diffusion Transformers to produce realistic 3D models with UV-patches. This patch-based format naturally describes local regions of the generated object, making editing easier and enhancing the creative design process.

Chat is not available.