Research map / 3D vision / 3D generation
Skyfall-GS: Synthesizing Immersive 3D Urban Scenes from Satellite Imagery
Jie-Ying Lee, Yi-Ruei Liu, Shr-Ruei Tsai, Wei-Cheng Chang, Chung-Ho Wu, Jiewen Chan, Zhenjun Zhao, Chieh Hubert Lin, Yu-Lun 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 appearance modeling to the reconstruction stage (ablation) | Skyfall-GS reconstruction · JAX_068 | FID_CLIP: Failed -> 41.90 | Improved |
| add opacity regularization (ablation) | Skyfall-GS reconstruction · JAX_068 | FID_CLIP: 41.90 -> 39.95 | Improved |
| add pseudo-camera depth supervision (ablation) | Skyfall-GS reconstruction · JAX_068 | FID_CLIP: 39.95 -> 38.01 | Improved |
| add pseudo-camera depth supervision (ablation) | Skyfall-GS reconstruction · JAX_068 (DFC2019 LiDAR DSM) | MAE (m): 2.980 -> 2.250 | Improved |
| add opacity regularization (ablation) | Skyfall-GS reconstruction · JAX_068 (DFC2019 LiDAR DSM) | MAE (m): 3.542 -> 2.980 | Improved |
| use two refined samples per view instead of one (ablation) | Skyfall-GS synthesis · JAX_068 | FID_CLIP: 34.11 -> 28.35 | Improved |
| increase samples per view from 2 to 5 (ablation) | Skyfall-GS synthesis · JAX_068 | CMMD: 2.875 -> 2.677 (FID 28.35 -> 29.17, time 6.37 -> 9.80 h) | Mixed |
| use high-to-low curriculum instead of random elevation sampling (ablation) | Skyfall-GS synthesis · JAX_068 | FID_CLIP: 33.79 -> 28.35 | Improved |
| use high-to-low instead of reversed (low-to-high) curriculum (ablation) | Skyfall-GS synthesis · JAX_068 | FID_CLIP: 53.03 -> 28.35 | Improved |
| use context-aware prompts instead of context-free prompts (ablation) | Skyfall-GS synthesis · JAX_068 | FID_CLIP: 30.78 -> 28.35 | No clear effect |
| use FlowEdit-based refinement instead of SDEdit (ablation) | Skyfall-GS synthesis · JAX_068 | FID_CLIP: 64.74 -> 28.35 | Improved |
| add diffusion-based IDU refinement on top of satellite reconstruction (vs Mip-Splatting) | Skyfall-GS · DFC2019 | FID_CLIP: 86.72 -> 27.03 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Mip-Splatting — Mip-Splatting: Alias-free 3D Gaussian Splatting (CVPR 2024) · better on 5 of 6 shared results
- CoR-GS — CoR-GS: Sparse-View 3D Gaussian Splatting via Co-Regularization (ECCV 2024) · better on 4 of 4 shared results
- CityDreamer — CityDreamer: Compositional Generative Model of Unbounded 3D Cities (CVPR 2024) · 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
- Sat2City: 3D City Generation from A Single Satellite Image with Cascaded Latent Diffusion (ICCV 2025 · similarity 0.87)
- Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes (CVPR 2025 · similarity 0.87)
- EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis (CVPR 2025 · similarity 0.85)
- CymbaDiff: Structured Spatial Diffusion for Sketch-based 3D Semantic Urban Scene Generation (NeurIPS 2025 · similarity 0.85)
- Dynamic 3D Gaussian Fields for Urban Areas (NeurIPS 2024 · similarity 0.84)
- MS-GS: Multi-Appearance Sparse-View 3D Gaussian Splatting in the Wild (NeurIPS 2025 · similarity 0.84)
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.
