metadata
license: apache-2.0
task_categories:
- image-to-3d
The datasets for Skyfall-GS
Project Page | Paper | GitHub
Skyfall-GS is a hybrid framework that synthesizes immersive, city-block scale 3D urban scenes by combining satellite reconstruction with diffusion refinement. This repository contains the JAX and NYC datasets used for training and evaluation.
Dataset Structure
According to the official GitHub documentation, the datasets should be organized in the data/ directory as follows:
data/
├── datasets_JAX/
│ ├── JAX_004
│ ├── JAX_068
│ └── ...
└── datasets_NYC/
├── NYC_004
├── NYC_010
└── ...
Each individual scene directory (e.g., JAX_068) contains the following structure:
your_dataset/
├── images/ # RGB images
│ ├── image_001.png
│ ├── image_002.png
│ └── ...
├── masks/ # Binary masks for valid pixels (optional)
│ ├── *.npy # NumPy format
│ ├── *.png # PNG format
│ └── ...
├── transforms_train.json # Training camera parameters
├── transforms_test.json # Testing camera parameters (optional)
└── points3D.txt # 3D point cloud
Citation
If you find this work or the datasets useful, please consider citing:
@article{lee2025SkyfallGS,
title = {{Skyfall-GS}: Synthesizing Immersive {3D} Urban Scenes from Satellite Imagery},
author = {Jie-Ying Lee and Yi-Ruei Liu and Shr-Ruei Tsai and Wei-Cheng Chang and Chung-Ho Wu and Jiewen Chan and Zhenjun Zhao and Chieh Hubert Lin and Yu-Lun Liu},
journal = {arXiv preprint},
year = {2025},
eprint = {2510.15869},
archivePrefix = {arXiv}
}