Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

DreamCS: Geometry-Aware Text-to-3D Generation with Unpaired 3D Reward Supervision

About

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 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 Cauchy-Schwarz 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-to-3D pipelines -- enhancing both implicit and explicit 3D generation with human preference feedback. Extensive experiments show DreamCS outperforms prior methods, producing 3D assets that are both geometrically faithful and human-preferred. Code and models will be released publicly.

Xiandong Zou, Ruihao Xia, Hongsong Wang, Pan Zhou• 2025

Related benchmarks

TaskDatasetResultRank
Text-to-3D GenerationGPTEval3D 110 prompts
CP0.23
20
Text-to-3D GenerationGPTEval3D 30 evaluation prompts
CP0.3
10
Text-to-3D GenerationGPTEval3D 60 prompts
Proportion41
10
Text-to-3D GenerationGPTEval3D User Study 60 total prompts (30 representative prompts)
T-A Score3.41
5
Text-to-3D GenerationGPTEval3D 110 prompts
Text-Asset Alignment (T-A)3.61
5
Text-to-3D GenerationGPTEval3D 60 prompts (evaluation)
CP Score0.3
3
Showing 6 of 6 rows

Other info

Follow for update