Elevating 3D Models: High-Quality Texture and Geometry Refinement from a Low-Quality Model

Nuri Ryu1, Jiyun Won1, Jooeun Son1, Minsu Gong1, Joo-Haeng Lee2, Sunghyun Cho1
1 POSTECH, 2 Pebblous
SIGGRAPH 2025
arXiv Paper Code

Elevate3D enhances low-quality 3D models, transforming them into high-quality assets through iterative texture and geometry refinement. At its core, HFS-SDEdit generatively refines textures while preserving the input’s identity, leveraging high-frequency guidance.

Teaser Image

3D refinement examples from a degraded real-world scan (a) and a state-of-the-art image-to-3D generative model (c). Elevate3D effectively refines both texture and geometry while preserving their alignment, as shown in (b) and (d).

Abstract

High-quality 3D assets are essential for various applications in computer graphics and 3D vision but remain scarce due to significant acquisition costs. To address this shortage, we introduce Elevate3D, a novel framework that transforms readily accessible low-quality 3D assets into higher quality. At the core of Elevate3D is HFS-SDEdit, a specialized texture enhancement method that significantly improves texture quality while preserving the appearance and geometry while fixing its degradations. Furthermore, Elevate3D operates in a view-by-view manner, alternating between texture and geometry refinement. Unlike previous methods that have largely overlooked geometry refinement, our framework leverages geometric cues from images refined with HFS-SDEdit by employing state-of-the-art monocular geometry predictors. This approach ensures detailed and accurate geometry that aligns seamlessly with the enhanced texture. Elevate3D outperforms recent competitors by achieving state-of-the-art quality in 3D model refinement, effectively addressing the scarcity of high-quality open-source 3D assets.

Key Contributions

  • We propose Elevate3D which refines 3D models by alternating between texture and geometry refinement, creating high-quality assets with well-aligned features.
  • We introduce HFS-SDEdit for texture refinement, which mitigates the quality-fidelity trade-off of SDEdit leveraging high-frequency guidance.
  • HFS-SDEdit enables the diffusion model to freely generate high-quality low-frequency features, shifting the image generation away from the input's low-quality domain.
  • Simultaneously, HFS-SDEdit injects high-frequency components from the input to preserve its identity and crucial details without inheriting low-quality artifacts.

HFS-SDEdit

HFS-SDEdit tackles the quality-fidelity trade-off in SDEdit. SDEdit can remove the low-quality domain information only when a large amount of noise $\epsilon$ is added to the the low-quality reference image $\mathbf{z}_r$ (c). By initiating the denoising process from the noisy latent $\mathbf{z}_{t_h}$ a high-quality image is generated at the cost of fidelity (b). Adding a small mount of noise and starting from the the noisy latent $\mathbf{z}_{t_l}$ maintains the low-quality domain information, resulting in only minor refinements (d). In contrast, HFS-SDEdit incorporates high-frequency feature injection-based guidance, allowing for high-fidelity generation even when starting the denoising process from $\mathbf{z}_{t_h}$ (a).

HFS-SDEdit 1
HFS-SDEdit 2

Elevate3D

Leveraging HFS-SDEdit, we introduce Elevate3D, a framework that refines 3D models by iteratively alternating texture and geometry refinement for high-quality assets with well-aligned texture and geometry.

Elevate3D

Visual Comparison

Right Image
Left Image

3D Refinement Results (GSO)

►
RGB
RGB
Normal
Normal

3D Refinement Results (TRELLIS)

►
RGB
RGB
Normal
Normal

BibTeX

@inproceedings{10.1145/3721238.3730701,
author = {Ryu, Nuri and Won, Jiyun and Son, Jooeun and Gong, Minsu and Lee, Joo-Haeng and Cho, Sunghyun},
title = {Elevating 3D Models: High-Quality Texture and Geometry Refinement from a Low-Quality Model},
year = {2025},
isbn = {9798400715402},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3721238.3730701},
doi = {10.1145/3721238.3730701},
booktitle = {ACM SIGGRAPH 2025 Conference Papers},
articleno = {},
numpages = {12},
keywords = {3D model refinement, texture enhancement, geometry refinement, computer graphics, 3D vision},
location = {Vancouver, BC, Canada},
series = {SIGGRAPH '25}
}