Research map / 3D vision / 3D generation
DreamScene: 3D Gaussian-based End-to-end Text-to-3D Scene Generation
Haoran Li, Yuli Tian, Kun Lan, Yong Liao, Lin Wang, Pan Hui, Pengyuan Zhou
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 DreamScene (planning, FPS, three-step camera sampling) instead of Text2Room | DreamScene · Scene user study | consistency: 2.57 -> 4.24 | Improved |
| use DreamScene instead of Set-the-Scene | DreamScene · Scene user study | quality: 2.45 -> 3.92 | Improved |
| use DreamScene instead of Set-the-Scene on generation time | DreamScene · Scene generation time | hours: 1.5 -> 1.5 | No clear effect |
| use multi-timestep sampling (FPS) instead of 3DGS+DreamTime | 3D Gaussian text-to-3D · R-Precision ViT-L/14 | CLIP R-Precision: 34% -> 71.9% | Improved |
| use multi-timestep sampling (FPS) instead of 3DGS+DreamTime | 3D Gaussian text-to-3D · R-Precision ViT-BigG/14 | CLIP R-Precision: 33.3% -> 70.6% | Improved |
| add annealing of the time window T (ablation) | DreamScene · R-Precision ViT-BigG/14 | CLIP R-Precision: 68.6% -> 70.6% | Improved |
| add annealing of the time window T (ablation) | DreamScene · R-Precision ViT-L/14 | CLIP R-Precision: 71.9% -> 71.9% | No clear effect |
| use random sampling within each interval instead of fixed-step sampling in MTS (ablation) | MTS · object generation (qualitative) | generation quality | Improved |
| use linearly decreasing time window instead of fixed max step 1000 or DreamTime schedule (ablation) | MTS · object generation (qualitative) | generation quality | Improved |
| add 3D Gaussian filtering in reconstruction task (ablation) | 3D Gaussian reconstruction · NeRF-360 | memory: 73.9% memory reduction with slightly blurred image | Mixed |
| add 3D Gaussian filtering in generation task (ablation) | DreamScene FPS · object generation | memory: 66.1% reduction, no significant loss of quality | Improved |
| use three-step camera sampling instead of random sampling (ablation) | DreamScene environment generation · scene generation (qualitative) | scene consistency | Improved |
Closest papers by meaning
- ArtiScene: Language-Driven Artistic 3D Scene Generation Through Image Intermediary (CVPR 2025 · similarity 0.90)
- GALA3D: Towards Text-to-3D Complex Scene Generation via Layout-guided Generative Gaussian Splatting (ICML 2024 · similarity 0.89)
- 3D-SceneDreamer: Text-Driven 3D-Consistent Scene Generation (CVPR 2024 · similarity 0.89)
- DreamControl: Control-Based Text-to-3D Generation with 3D Self-Prior (CVPR 2024 · similarity 0.88)
- 3DitScene: Editing Any Scene via Language-guided Disentangled Gaussian Splatting (ICLR 2025 · similarity 0.88)
- Towards Text-guided 3D Scene Composition (CVPR 2024 · similarity 0.88)
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.
