Saved in:
Bibliographic Details
Main Authors: Mercier, Antoine, Nakhli, Ramin, Reddy, Mahesh, Yasarla, Rajeev, Cai, Hong, Porikli, Fatih, Berger, Guillaume
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2401.07727
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911758424211456
author Mercier, Antoine
Nakhli, Ramin
Reddy, Mahesh
Yasarla, Rajeev
Cai, Hong
Porikli, Fatih
Berger, Guillaume
author_facet Mercier, Antoine
Nakhli, Ramin
Reddy, Mahesh
Yasarla, Rajeev
Cai, Hong
Porikli, Fatih
Berger, Guillaume
contents Despite the latest remarkable advances in generative modeling, efficient generation of high-quality 3D assets from textual prompts remains a difficult task. A key challenge lies in data scarcity: the most extensive 3D datasets encompass merely millions of assets, while their 2D counterparts contain billions of text-image pairs. To address this, we propose a novel approach which harnesses the power of large, pretrained 2D diffusion models. More specifically, our approach, HexaGen3D, fine-tunes a pretrained text-to-image model to jointly predict 6 orthographic projections and the corresponding latent triplane. We then decode these latents to generate a textured mesh. HexaGen3D does not require per-sample optimization, and can infer high-quality and diverse objects from textual prompts in 7 seconds, offering significantly better quality-to-latency trade-offs when comparing to existing approaches. Furthermore, HexaGen3D demonstrates strong generalization to new objects or compositions.
format Preprint
id arxiv_https___arxiv_org_abs_2401_07727
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle HexaGen3D: StableDiffusion is just one step away from Fast and Diverse Text-to-3D Generation
Mercier, Antoine
Nakhli, Ramin
Reddy, Mahesh
Yasarla, Rajeev
Cai, Hong
Porikli, Fatih
Berger, Guillaume
Computer Vision and Pattern Recognition
Despite the latest remarkable advances in generative modeling, efficient generation of high-quality 3D assets from textual prompts remains a difficult task. A key challenge lies in data scarcity: the most extensive 3D datasets encompass merely millions of assets, while their 2D counterparts contain billions of text-image pairs. To address this, we propose a novel approach which harnesses the power of large, pretrained 2D diffusion models. More specifically, our approach, HexaGen3D, fine-tunes a pretrained text-to-image model to jointly predict 6 orthographic projections and the corresponding latent triplane. We then decode these latents to generate a textured mesh. HexaGen3D does not require per-sample optimization, and can infer high-quality and diverse objects from textual prompts in 7 seconds, offering significantly better quality-to-latency trade-offs when comparing to existing approaches. Furthermore, HexaGen3D demonstrates strong generalization to new objects or compositions.
title HexaGen3D: StableDiffusion is just one step away from Fast and Diverse Text-to-3D Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2401.07727