Research map / 3D vision / 3D generation
Hyper-3DG: Text-to-3D Gaussian Generation via Hypergraph
Donglin Di, Jiahui Yang, Chaofan Luo, Zhou Xue, Wei Ren Chen, Xun Yang, Yue Gao
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 HGRefiner hypergraph refinement on top of LucidDreamer pipeline | Hyper-3DG · User study | user rating: 3.6 -> 4.1 | Improved |
| use ISM loss instead of SDS or VSD (ablation) | Hyper-3DG · text-to-3D prompts (lion, bus) | visual quality (qualitative) | Improved |
| use K-Means instead of DBSCAN or GMM for 3DGS-Patchify (ablation) | Hyper-3DG · text-to-3D prompts (strawberry, headphones, fire truck) | visual quality (qualitative) | Improved |
| set K_pat near 80 instead of extreme values 1 or 200 (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): K_pat 1 and 200 artifacts; 80 repeated artifacts | Mixed |
| use mid-range KNN K_spa and K_lat (13-33) (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): 13-33 | Improved |
| replace graph convolution (GCN) with hypergraph convolution (ablation) | Hyper-3DG · text-to-3D prompts (superman, panda, Packard car) | visual quality (qualitative) | Improved |
| increase warm-up steps N0 from 0 to 1,000 (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): 0 -> 1,000 | Improved |
| increase refine steps N1 from 10 to 70 (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): 10 -> 70 | Improved |
| increase refine steps N1 from 70 to 100 (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): 70 -> 100 | Hurt |
| use Point-E instead of Shap-E as 3D initializer (ablation) | Hyper-3DG · text-to-3D prompts (eagle, corgi, armored dog) | visual quality (qualitative) | Improved |
| use DINO instead of other 2D feature extractors (ResNet, ResNeXt, ViT, Swin-T) (ablation) | Hyper-3DG · text-to-3D prompts (panda, basketball) | visual quality (qualitative) | No clear effect |
| raise number of random render views CM from 2-3 to 4-7 (ablation) | Hyper-3DG · Packard car prompt | visual quality (qualitative): 2-3 -> 4-7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LucidDreamer — LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching (CVPR 2024) · better on 1 of 1 shared result
- DreamGaussian — DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation (ICLR 2024) · better on 1 of 1 shared result
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
- Text-to-3D using Gaussian Splatting (CVPR 2024 · similarity 0.88)
- GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models (CVPR 2024 · similarity 0.87)
- Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation (ICLR 2025 · similarity 0.87)
- DIRECT-3D: Learning Direct Text-to-3D Generation on Massive Noisy 3D Data (CVPR 2024 · similarity 0.87)
- HyperSDFusion: Bridging Hierarchical Structures in Language and Geometry for Enhanced 3D Text2Shape Generation (CVPR 2024 · similarity 0.87)
- Textured Gaussians for Enhanced 3D Scene Appearance Modeling (CVPR 2025 · 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.
