Research map / 3D vision / 3D generation
SceneDreamer: Unbounded 3D Scene Generation From 2D Image Collections
Zhaoxi Chen, Guangcong Wang, Ziwei Liu
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 BEV scene encoder (ablation) | SceneDreamer · scene generation | FID: 80.62 -> 76.73 | Improved |
| add BEV scene encoder (ablation) | SceneDreamer · scene generation | camera error (CE): 0.044 -> 0.021 | Improved |
| add semantic field to BEV input (ablation) | SceneDreamer · scene generation | FID: 83.20 -> 76.73 | Improved |
| add height field to BEV input (ablation) | SceneDreamer · scene generation | FID: 79.80 -> 76.73 | Improved |
| add semantic field to BEV input (ablation) | SceneDreamer · scene generation | depth error: 0.401 -> 0.277 | Improved |
| use BEV instead of semantic voxels (ablation) | SceneDreamer · memory budget 40GB | max supported LOD: 512x512x256 -> 2048x2048x256 | Improved |
| use generative hash grid instead of pure MLP (ablation) | SceneDreamer · scene generation | FID: 88.76 -> 76.73 | Improved |
| use generative hash grid instead of MLP with positional encoding (ablation) | SceneDreamer · scene generation | FID: 83.14 -> 76.73 | Improved |
| use generative hash grid instead of HyperNet (ablation) | SceneDreamer · scene generation | FID: 117.64 -> 76.73 | Improved |
| use generative hash grid instead of tri-plane (ablation) | SceneDreamer · scene generation | FID: 103.86 -> 76.73 | Improved |
| add rejection sampling of cameras instead of random sampling (ablation) | SceneDreamer · scene generation | FID: 102.60 -> 76.73 | Improved |
| add rejection sampling instead of area sampling (ablation) | SceneDreamer · scene generation | FID: 96.90 -> 76.73 | Improved |
Later papers that compared with this method
2 later accepted papers report SceneDreamer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Compositional Generative Model of Unbounded 4D Cities (TPAMI 2025 · better on 4 of 4 shared results)
- CityDreamer: Compositional Generative Model of Unbounded 3D Cities (CVPR 2024 · better on 4 of 4 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
- 3D-SceneDreamer: Text-Driven 3D-Consistent Scene Generation (CVPR 2024 · similarity 0.91)
- CityDreamer: Compositional Generative Model of Unbounded 3D Cities (CVPR 2024 · similarity 0.90)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.88)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.88)
- Real3D: The Curious Case of Neural Scene Degeneration (AAAI 2024 · similarity 0.88)
- SemCity: Semantic Scene Generation with Triplane Diffusion (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.
