Research map / Image generation / Fast diffusion sampling
Unleashing Vecset Diffusion Model for Fast Shape Generation
Zeqiang Lai, Yunfei Zhao, Zibo Zhao, Haolin Liu, Fuyun Wang, Huiwen Shi, Xianghui Yang, Qingxiang Lin, Jingwei Huang, Yuhong Liu, Jie Jiang, Chunchao Guo 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 |
|---|---|---|---|
| add hierarchical volume decoding to the VAE decoder (ablation) | Hunyuan3D-2 VAE decoder · Shape reconstruction (3072 tokens) | decoding time (s): 22.33 -> 2.322 with V-IoU 96.11% unchanged | Improved |
| add efficient decoder design (narrower, smaller MLP ratio, fewer LayerNorms) (ablation) | Hunyuan3D-2 VAE decoder · Shape reconstruction (3072 tokens) | decoding time (s): 2.322 -> 0.731; V-IoU 96.11% -> 96.08% | Improved |
| add adaptive KV selection (ablation) | Hunyuan3D-2 VAE decoder · Shape reconstruction (3072 tokens) | decoding time (s): 0.731 -> 0.491; V-IoU 96.08% -> 95.55% | Improved |
| add adaptive KV selection (ablation) | Hunyuan3D-2 VAE decoder · Shape reconstruction (3072 tokens) | V-IoU: 96.08% -> 95.55% | Hurt |
| apply FlashVDM to Hunyuan3D-2 VAE (3072 tokens) | Hunyuan3D-2 · Shape reconstruction | decoding time (s): 22.33 -> 0.491 with V-IoU 96.11% -> 95.55% | Improved |
| apply FlashVDM distillation and fast decoder to Hunyuan3D-2 for generation | Hunyuan3D-2 · Image-to-3D generation | generation time (s): 34.85 -> 1.041; ULIP-I 0.1303 -> 0.1260 | Improved |
| add guidance distillation warm-up before consistency distillation (ablation) | Progressive Flow Distillation · 5-step shape generation (qualitative) | visual shape quality | Improved |
| add EMA update of the target model (ablation) | Progressive Flow Distillation · 5-step shape generation (qualitative) | visual shape quality | Improved |
| replace L2 loss with Huber loss (ablation) | Progressive Flow Distillation · 5-step shape generation (qualitative) | visual shape quality | Improved |
| add Phase 1 fine-tuning after Phase 5 pretraining (ablation) | Progressive Flow Distillation · 5-step shape generation (qualitative) | visual shape quality | Improved |
| add adversarial fine-tuning with latent-space discriminator on real 3D data (ablation) | Progressive Flow Distillation · 5-step shape generation (qualitative) | visual shape quality | Improved |
| add dilation and tSDF supervision to hierarchical volume decoding (ablation) | Hierarchical volume decoding · Shape reconstruction (qualitative) | mesh completeness | 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.
- Michelangelo — Michelangelo: Conditional 3D Shape Generation based on Shape-Image-Text Aligned Latent Representation (NeurIPS 2023) · better on 2 of 2 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
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.87)
- Scaling Diffusion Mamba with Bidirectional SSMs for Efficient 3D Shape Generation (AAAI 2025 · similarity 0.87)
- One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion (CVPR 2024 · similarity 0.86)
- Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders (CVPR 2025 · similarity 0.86)
- Make-A-Shape: a Ten-Million-scale 3D Shape Model (ICML 2024 · similarity 0.86)
- Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modeling (NeurIPS 2025 · similarity 0.86)
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.
