emasquil/shadow-eo

Dataset Card for S-EO: A Large-Scale Dataset for Geometry-Aware Shadow Detection in Remote Sensing Applications

Project page

We introduce the S-EO dataset: a large-scale, high-resolution dataset designed to advance geometry-aware shadow detection. Collected from diverse public-domain sources, including challenge datasets and government providers such as USGS, our dataset comprises 702 georeferenced tiles across the USA, each covering 500 × 500 meters. Each tile includes multi-date, multi-angle WorldView-3 pansharpened RGB images, panchromatic images, and a ground-truth DSM of the area obtained from LiDAR scans. For each image, we provide a shadow mask derived from geometry and sun position, a vegetation mask based on the NDVI index, and a bundle-adjusted RPC model. With approximately 20,000 images, the S-EO dataset establishes a new public resource for shadow detection in remote sensing imagery and its applications to 3D reconstruction. To demonstrate the dataset’s impact, we train and evaluate a shadow detector, showcasing its ability to generalize even to aerial images. Finally, we extend EO-NeRF — a state-of-the-art NeRF approach for satellite imagery — to leverage our shadow predictions for improved 3D reconstructions.

---

Project developed between Universidad de la República, Digital Sense, Eurecat, AMIAD, and ENS Paris-Saclay, Centre Borelli. Accepted at the CVPR EarthVision Workshop 2025.

---

Dataset Details

The dataset includes the following components:

• curated_aois_v3.csv: A list of Areas of Interest (AOIs) that constitute the dataset.
• DSMs (Max and Min): Digital Surface Models aggregated using maximum and minimum methods, respectively.
• MSI Crops: Multispectral image crops containing Red, Green, Blue, and NIR bands, provided in raw form prior to radiometric correction (uint16).
• Panchromatic Crops: High-resolution single-band images before top-of-atmosphere correction (uint16).
• Pansharpened Crops: Color-corrected and histogram-equalized pansharpened images (uint8), used as input for training the shadow detection model.
• RPCs: Bundle-adjusted Rational Polynomial Coefficients provided in JSON format.
• Shadow Maps (Max and Min): Shadow annotations derived from the corresponding DSMs, including uncertainty and vegetation masks.

Each resource is distributed as an independent tar.gz archive. For large files (e.g., pansharpened crops), archives are split into multiple parts due to size limitations.

Source Data

This dataset is derived from the IARPA’s CORE3D program data

Annotation process

We use a shadow simulation algorithm to automatically generate shadow masks at scale from existing data. For each image in the dataset, the shadow mask is computed using the sun position at the time of capture, the aligned DSM for the area of interest, and the image camera models provided by the RPCs.

Citation

BibTeX:

@inproceedings{masquil2025shadoweo,
  title={S-EO: A Large-Scale Dataset for Geometry-Aware Shadow Detection in Remote Sensing Applications},
  author={Masquil, El\'{i}as and Mar\'{i}, Roger and Ehret, Thibaud and Meinhardt-Llopis, Enric and Mus\'{e}, Pablo and Facciolo, Gabriele},
  booktitle={Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
  year={2025}
}

Dataset Card Authors

Elías Masquil

Dataset Card Contact

eliasmasquil[at]gmail[dot]com