Research map / Image generation / Text-to-image diffusion
Mesh-Pro: Asynchronous Advantage-guided Ranking Preference Optimization for Artist-style Quadrilateral Mesh Generation
Zhen Zhou, Jian Liu, Biwen Lei, Jing Xu, Haohan Weng, Yiling Zhu, Zhuo Chen, Junfeng Fan, Yunkai Ma, Dazhao Du, Song Guo, Fengshui Jing and 1 more
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 |
|---|---|---|---|
| use ARPO instead of GRPO (ablation) | Async RL post-training · Async RL algorithm comparison | quad ratio: 68% -> 81% | Improved |
| use ARPO instead of DPO (ablation) | Async RL post-training · Async RL algorithm comparison | quad ratio: 75% -> 81% | Improved |
| use GRPO instead of pretrained model (no RL) (ablation) | Mesh-Pro · Async RL algorithm comparison | user study score: 0.4 -> 1.3 | Improved |
| add advantage guidance to ranking preference (ablation) | ARPO (group 4) · Async ARPO component ablation | user study score: 2.4 -> 2.9 | Improved |
| increase group size from 2 to 4 (ablation) | ARPO · Async ARPO component ablation | user study score: 2.1 -> 2.4 | Improved |
| asynchronous online instead of synchronous RL (ablation) | RL post-training · training time | speedup: 3.75x faster | Improved |
| asynchronous online instead of offline RL (ablation) | ARPO · Async ARPO component ablation | CD/HD/BR/QR/US | Improved |
| add pre-start stage (ablation) | async ARPO · Async ARPO component ablation | stability / user study: 2.7 -> 3.5 | Improved |
| diagonal-aware tokenizer instead of previous tokenizer (ablation) | Mesh-Pro · Tokenizer comparison (no RL) | broken ratio: 38% -> 24% | Improved |
| apply async ARPO to previous tokenizer (ablation) | QuadGPT-style tokenizer · Tokenizer comparison | broken ratio: 38% -> 46% | Hurt |
| apply async ARPO to diagonal-aware tokenizer (ablation) | Mesh-Pro · Tokenizer comparison | broken ratio: 24% -> 22% | Improved |
| increase ARPO training data 400 to 1000 meshes (ablation) | async ARPO · Data scale in async ARPO | broken ratio: 22.35% -> 21.83% | Improved |
Closest papers by meaning
- Mesh-RFT: Enhancing Mesh Generation via Fine-grained Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction (NeurIPS 2025 · similarity 0.87)
- EdgeRunner: Auto-regressive Auto-encoder for Artistic Mesh Generation (ICLR 2025 · similarity 0.86)
- MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers (ICLR 2025 · similarity 0.85)
- MeshArt: Generating Articulated Meshes with Structure-Guided Transformers (CVPR 2025 · similarity 0.85)
- Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards (NeurIPS 2025 · similarity 0.84)
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.
