Publication Type

Conference Proceeding Article

Version

acceptedVersion

Publication Date

4-2026

Abstract

While text-to-3D generation has attracted growing interest, existing methods often struggle to produce 3D assets that align well with human preferences. Current preference alignment techniques for 3D content typically rely on hardly-collected preference-paired multi-view 2D images to train 2D reward models, when then guide 3D generation — leading to geometric artifacts, such as the Janus face problem and geometric incompleteness, due to their inherent 2D bias. To address these limitations, we construct 3D-MeshPref, the first large-scale unpaired 3D preference dataset, featuring diverse 3D meshes annotated by a large language model and refined by human evaluators. We then develop RewardCS, the first reward model trained directly on unpaired 3D-MeshPref data using a novel CauchySchwarz divergence objective, enabling effective learning of human-aligned 3D geometric preferences without requiring paired comparisons. Building on this, we propose DreamCS, a unified framework that integrates RewardCS into text-to3D pipelines — enhancing both implicit and explicit 3D generation with human preference feedback. Extensive experiments show DreamCS outperforms prior methods, producing 3D assets that are geometrically faithful and human-preferred.

Discipline

Artificial Intelligence and Robotics | Graphics and Human Computer Interfaces

Research Areas

Intelligent Systems and Optimization

Areas of Excellence

Digital transformation

Publication

Proceedings of the Fourteenth International Conference on Learning Representations (ICLR 2026), Rio de Janeiro, Brazil, April 23-27

First Page

1

Last Page

46

Publisher

ICLR

City or Country

Rio de Janeiro, Brazil

Additional URL

https://openreview.net/forum?id=PoU33ZdtCP

Download

Share

COinS