sky2ground

Sky2Ground: A Benchmark for Site Modeling under Varying Altitude

Zengyan Wang, Sirshapan Mitra, Rajat Modi, Grace Lim, Yogesh Rawat
CVPR 2026
[arXiv] | [Dataset] | [Code]

🖼 Dataset Preview

Our dataset bridges the gap between synthetic environments and real-world captures. Below are samples of the multi-view perspectives provided.

🌐 Synthetic Dataset (GIF Samples)

Generated environments featuring a full 5-view suite.

Satellite | Aerial View 1 | Aerial View 2 | Aerial View 3 | Street View

---

📸 Real-World Dataset (Static Images)

Authentic captures for domain validation.

Satellite | Aerial View 1 | Aerial View 2 | Street View 1 | Street View 2 | Street View 3

Note: Real-world samples are provided as high-resolution static images, while synthetic samples include dynamic transitions (GIFs) to demonstrate environmental variance.

🚀 Access the Dataset

The dataset will be available on the following platforms for ease of use in machine learning workflows:

Platform	Link	Recommended For
Hugging Face	🤗 Under Construction
Kaggle	📁 Under Construction

---

🛠 Project Progress

• Synthetic Images
• Real Images
• Benchmark

Citation

If you find our work to be useful in your research, please consider citing our paper:

@misc{wang2026sky2groundbenchmarksitemodeling,
      title={Sky2Ground: A Benchmark for Site Modeling under Varying Altitude}, 
      author={Zengyan Wang and Sirshapan Mitra and Rajat Modi and Grace Lim and Yogesh Rawat},
      year={2026},
      eprint={2603.13740},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.13740}, 
}