mlsanityData to 7 Oct 2026

Mesh-Pro: Asynchronous Advantage-guided Ranking Preference Optimization for Artist-style Quadrilateral Mesh Generation

arXiv 2603.00526 ↗First posted 28 Feb 2026cs.CVcited by 2

On the map: LLM alignment / RLHF / preference optimisation → Diffusion and multimodal model alignment

Abstract (arXiv)

Reinforcement learning (RL) has demonstrated remarkable success in text and image generation, yet its potential in 3D generation remains largely unexplored. Existing attempts typically rely on offline direct preference optimization (DPO) method, which suffers from low training efficiency and limited generalization. In this work, we aim to enhance both the training efficiency and generation quality of RL in 3D mesh generation. Specifically, (1) we design the first asynchronous online RL framework tailored for 3D mesh generation post-training efficiency improvement, which is 3.75× faster than synchronous RL. (2) We propose Advantage-guided Ranking Preference Optimization (ARPO), a novel RL algorithm that achieves a better trade-off between training efficiency and generalization than current RL algorithms designed for 3D mesh generation, such as DPO and group relative policy optimization (GRPO). (3) Based on asynchronous ARPO, we propose Mesh-Pro, which additionally introduces a novel diagonal-aware mixed triangular-quadrilateral tokenization for mesh representation and a ray-based reward for geometric integrity. Mesh-Pro achieves state-of-the-art performance on artistic and dense meshes.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Diffusion and multimodal model alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 52 of its 64 references that are in the map. First posted on arXiv 28 Feb 2026.

Builds on

52 of 64 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Mesh-RFT: Enhancing Mesh Generation via Fine-grained Reinforcement Fine-Tuning22 May
  2. 2025DeepMesh: Auto-Regressive Artist-mesh Creation with Reinforcement Learning19 Mar

Other references · 50

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  4. 2015ShapeNet: An Information-Rich 3D Model Repositorycited by 2,880
  5. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 1,980
  6. 2022Objaverse: A Universe of Annotated 3D Objectscited by 1,522
  7. 2023Objaverse-XL: A Universe of 10M+ 3D Objectscited by 767
  8. 2024Structured 3D Latents for Scalable and Versatile 3D Generationcited by 742
  9. 2025Flow-GRPO: Training Flow Matching Models via Online RLcited by 605
  10. 20233DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Modelscited by 449
  11. 2024CLAY: A Controllable Large-scale Generative Model for Creating High-quality 3D Assetscited by 444
  12. 2025Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generationcited by 342
38 more
  1. 20203D-FUTURE: 3D Furniture shape with TextUREcited by 306
  2. 2020PolyGen: An Autoregressive Generative Model of 3D Meshescited by 267
  3. 2023MeshGPT: Generating Triangle Meshes with Decoder-Only Transformerscited by 254
  4. 2025AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoningcited by 234
  5. 2023Michelangelo: Conditional 3D Shape Generation based on Shape-Image-Text Aligned Latent Representationcited by 225
  6. 2024Direct3D: Scalable Image-to-3D Generation via 3D Latent Diffusion Transformercited by 210
  7. 2025TripoSG: High-Fidelity 3D Shape Synthesis using Large-Scale Rectified Flow Modelscited by 204
  8. 2023Guiding Pretraining in Reinforcement Learning with Large Language Modelscited by 185
  9. 2024MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformerscited by 140
  10. 2025Hunyuan3D 2.5: Towards High-Fidelity 3D Assets Generation with Ultimate Detailscited by 120
  11. 2021Using Shape to Categorize: Low-Shot Learning with an Explicit Shape Biascited by 115
  12. 2025Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attentioncited by 103
  13. 2024MeshAnything V2: Artist-Created Mesh Generation With Adjacent Mesh Tokenizationcited by 102
  14. 2024MeshXL: Neural Coordinate Field for Generative 3D Foundation Modelscited by 100
  15. 2021Hierarchical Transformers Are More Efficient Language Modelscited by 99
  16. 2025Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridgingcited by 95
  17. 2024Meshtron: High-Fidelity, Artist-Like 3D Mesh Generation at Scalecited by 87
  18. 2024Noise Contrastive Alignment of Language Models with Explicit Rewardscited by 85
  19. 2024EdgeRunner: Auto-regressive Auto-encoder for Artistic Mesh Generationcited by 84
  20. 2025Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modelingcited by 84
  21. 2025SparseFlex: High-Resolution and Arbitrary-Topology 3D Shape Modelingcited by 76
  22. 2024Scaling Mesh Generation via Compressive Tokenizationcited by 71
  23. 2024MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction Modelcited by 55
  24. 2024PivotMesh: Generic 3D Mesh Generation via Pivot Vertices Guidancecited by 47
  25. 2025Ultra3D: Efficient and High-Fidelity 3D Generation with Part Attentioncited by 43
  26. 2025VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Spacecited by 40
  27. 2025TreeMeshGPT: Artistic Mesh Generation with Autoregressive Tree Sequencingcited by 38
  28. 2024Insights into Alignment: Evaluating DPO and its Variants Across Multiple Taskscited by 35
  29. 2025NANO3D: A Training-Free Approach for Efficient 3D Editing Without Maskscited by 29
  30. 2025Unleashing Vecset Diffusion Model for Fast Shape Generationcited by 25
  31. 2025FastMesh: Efficient Artistic Mesh Generation via Component Decouplingcited by 17
  32. 2025QuadGPT: Native Quadrilateral Mesh Generation with Autoregressive Modelscited by 11
  33. 2025CrossGen: Learning and Generating Cross Fields for Quad Meshingcited by 10
  34. 2024NeurCross: A Neural Approach to Computing Cross Fields for Quad Mesh Generationcited by 8
  35. 2025MeshMosaic: Scaling Artist Mesh Generation via Local-to-Global Assemblycited by 8
  36. 2025FreeMesh: Boosting Mesh Generation with Coordinates Mergingcited by 6
  37. 2025XSpecMesh: Quality-Preserving Auto-Regressive Mesh Generation Acceleration via Multi-Head Speculative Decodingcited by 5
  38. 2025Point2Quad: Generating Quad Meshes from Point Clouds via Face Predictioncited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenizationcited by 2
  2. 2026Flow Matching Reinforcement for 3D Mesh Generation via Dynamic Homing Optimizationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Are We Ready for RL in Text-to-3D Generation? A Progressive Investigation
  2. 2026Flow3D-OPD: Multi-Teacher On-Policy Distillation for 3D Geometry Generation with Flow-Matching Diffusion Transformer
  3. 2025DreamCS: Geometry-Aware Text-to-3D Generation with Unpaired 3D Reward Supervision
  4. 2025End-to-End Fine-Tuning of 3D Texture Generation using Differentiable Rewards
  5. 2025Mesh RAG: Retrieval Augmentation for Autoregressive Mesh Generation
  6. 2025MeshCraft: Exploring Efficient and Controllable Mesh Generation with Flow-based DiTs
  7. 2026Judging to Improve: A De-biased VLM-as-3D-Judge Protocol for Single-Image 3D Generation
  8. 2025DanceGRPO: Unleashing GRPO on Visual Generation