Research map / Image generation / Text-to-image diffusion
Steering 3D Generations: Preference Alignment via Direct Reward and Preference Optimization
Yuanhang Wang, Lizhe Qi, Wenqiang Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT on curated watertight data (ablation) | Hunyuan3D-2 · Toys4k | Chamfer Distance: 0.0079 -> 0.0045 | Improved |
| add SFT on curated watertight data (ablation) | Hunyuan3D-2 · Toys4k | F-Score: 73.79 -> 76.28 | Improved |
| add DPO preference stage after SFT (ablation) | Baseline+SFT · Toys4k | F-Score: 76.28 -> 76.77 | Improved |
| add DPO preference stage after SFT (ablation) | Baseline+SFT · Toys4k | Chamfer Distance: 0.0045 -> 0.0048 | Hurt |
| add DRO reward optimization after SFT (ablation) | Baseline+SFT · Toys4k | Chamfer Distance: 0.0045 -> 0.0036 | Improved |
| add DRO reward optimization after SFT (ablation) | Baseline+SFT · Toys4k | F-Score: 76.28 -> 77.97 | Improved |
| use DRO instead of DPO (ablation) | Baseline+SFT · Toys4k | Chamfer Distance: 0.0048 -> 0.0036 | Improved |
| use DRO instead of DPO (ablation) | Baseline+SFT · Toys4k | F-Score: 76.77 -> 77.97 | Improved |
| DRO fine-tuned Hunyuan3D-2 versus untuned Hunyuan3D-2 and other generators | Hunyuan3D-2 · Toys4k | Chamfer Distance: 0.0089 -> 0.0036 | Improved |
| DRO-tuned model compared against Clay in a human study | Hunyuan3D-2 · User study | vote share: 0.6 of the votes | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 3DTopia-XL — 3DTopia-XL: Scaling High-quality 3D Asset Generation via Primitive Diffusion (CVPR 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DreamAlign: Dynamic Text-to-3D Optimization with Human Preference Alignment (AAAI 2025 · similarity 0.90)
- Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards (NeurIPS 2025 · similarity 0.89)
- MVReward: Better Aligning and Evaluating Multi-View Diffusion Models with Human Preferences (AAAI 2025 · similarity 0.88)
- Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences (ICML 2025 · similarity 0.87)
- DreamDPO: Aligning Text-to-3D Generation with Human Preferences via Direct Preference Optimization (ICML 2025 · similarity 0.87)
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
