CVPR2025
Hash3D: Training-free Acceleration for 3D Generation
Xingyi Yang, Songhua Liu, Xinchao Wang
Abstract
An old car overgrown by vines and weeds An orc forging a hammer on an anvil A gummy bear playing the saxophone A marble bust of a fox head A porcelain dragon Dream-Gaussian Gaussian-Dreamer 1.5× 4. #× + Hash3D + Hash3D Figure 1. Examples by applying our Hash3D on Gaussian-Dreamer [63] and Dream-Gaussian [53]. We accelerate Gaussian-Dreamer by 1.5→ and Dream-Gaussian by 4→ with comparable visual quality.