Pith. sign in

REVIEW 4 cited by

DreamReward: Text-to-3D Generation with Human Preference

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2403.14613 v1 pith:DJEL2H4H submitted 2024-03-21 cs.CV cs.CLcs.LG

Junliang Ye , Fangfu Liu , Qixiu Li , Zhengyi Wang , Yikai Wang , Xinzhou Wang , Yueqi Duan , Jun Zhu This is my paper · ORCID
classification cs.CVcs.CLcs.LG
keywords humantext-to-3ddreamrewardfeedbackmodelspreferenceresultscomparisons
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

3D content creation from text prompts has shown remarkable success recently. However, current text-to-3D methods often generate 3D results that do not align well with human preferences. In this paper, we present a comprehensive framework, coined DreamReward, to learn and improve text-to-3D models from human preference feedback. To begin with, we collect 25k expert comparisons based on a systematic annotation pipeline including rating and ranking. Then, we build Reward3D -- the first general-purpose text-to-3D human preference reward model to effectively encode human preferences. Building upon the 3D reward model, we finally perform theoretical analysis and present the Reward3D Feedback Learning (DreamFL), a direct tuning algorithm to optimize the multi-view diffusion models with a redefined scorer. Grounded by theoretical proof and extensive experiment comparisons, our DreamReward successfully generates high-fidelity and 3D consistent results with significant boosts in prompt alignment with human intention. Our results demonstrate the great potential for learning from human feedback to improve text-to-3D models.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization

    cs.CV 2025-05 conditional novelty 7.0 of 10

    ADHMR aligns diffusion-based human mesh recovery with a learned scorer and direct preference optimization, improving accuracy and in-the-wild robustness without human preference labels.

  2. Edit in 2D, Verify in 3D: Reinforcement Learning for Multi-view Consistent Scene Editing

    cs.CV 2026-03 conditional novelty 6.0 of 10

    RL3DEdit fine-tunes FLUX-Kontext with GRPO using VGGT confidence and pose rewards to produce multi-view consistent 3D scene edits in a single pass.

  3. MVReward: Better Aligning and Evaluating Multi-View Diffusion Models with Human Preferences

    cs.CV 2024-12 conditional novelty 6.0 of 10

    MVReward, trained on 16,000 human comparisons, matches human ranking of seven multi-view generation methods perfectly (Spearman 1.00) and MVP fine-tuning improves human preference for Wonder3D and Era3D.

  4. AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers

    cs.CV 2024-11 conditional novelty 6.0 of 10

    AC3D improves camera control in video diffusion transformers by conditioning only early denoising steps and the first 8 of 32 blocks, and by adding 20K static-camera dynamic videos to training.

Pith tools