Research map / 3D vision / 3D generation
Extend3D: Town-Scale 3D Generation
Seungwoo Yoon, Jinmo Kim, Jaesik Park
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 initialization from depth point cloud (ablation) | Extend3D patch-wise flow · appearance | LPIPS: 0.606 -> 0.425 | Improved |
| add initialization from depth point cloud (ablation) | Extend3D patch-wise flow · geometry | Chamfer distance: 0.0348 -> 0.0083 | Improved |
| add sparse structure optimization (ablation) | Extend3D · geometry | F-score: 0.693 -> 0.708 | Improved |
| add SLat rendering optimization (ablation) | Extend3D · appearance | LPIPS: 0.400 -> 0.240 | Improved |
| add SLat rendering optimization (ablation) | Extend3D · geometry | Chamfer distance: 0.0078 -> 0.0086 | Hurt |
| increase division factor from 2 to 4 (ablation) | Extend3D · appearance | LPIPS: 0.251 -> 0.240 | Improved |
| increase division factor from 4 to 8 (ablation) | Extend3D · appearance | PSNR: 20.4 -> 20.5 | Improved |
| use under-noising (t_noise 0.6, t_start 0.8) instead of standard SDEdit 0.6/0.6 (ablation) | Extend3D initialization · geometry | F-score: 0.657 -> 0.680 | Improved |
| use under-noising instead of standard SDEdit 0.8/0.8 (ablation) | Extend3D initialization · appearance | LPIPS: 0.550 -> 0.387 | Improved |
| add one iterative under-noised SDEdit step (ablation) | Extend3D · geometry | F-score: 0.647 -> 0.681 | Improved |
| increase SDEdit iterations from 2 to 3 (ablation) | Extend3D · geometry | F-score: 0.688 -> 0.677 | Hurt |
| replace SynCity with Extend3D | Extend3D · patch consistency | Intra-LPIPS: 0.631 -> 0.571 | 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.
- SynCity — SynCity: Training-Free Generation of 3D Worlds (ICCV 2025) · 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
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.89)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.89)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.89)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.88)
- Structured 3D Latents for Scalable and Versatile 3D Generation (CVPR 2025 · similarity 0.88)
- Denoising Diffusion via Image-Based Rendering (ICLR 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.
