Dataset for Evaluating Sentinel-2 Super-Resolution Algorithms for Automated Building Delineation

Evaluating Sentinel-2 Super-Resolution Algorithms for Automated Building Delineation

This dataset is associated with the Master's Thesis "Evaluating Sentinel-2 Super-Resolution Algorithms for Automated Building Delineation" and includes all relevant datasets that were created to facilitate experiments conducted. The thesis included the evaluation of SR algorithms on the downstream task of building delineation on the example of Austria. To achieve this, several datasets had to be accessed and created, which are featured in this repository. Further information regarding the process involved, code repositories, and the published thesis are accessible under the GitHub repository: https://github.com/Zerhigh/Evaluating_Sentinel-2_Super-Resolution_Algorithms_for_Automated_Building_Delineation

Structure & Processing Details

All image files are processed similarly:  

Remote sensing images are saved as geotiffs with provided spatial transformation parameters. When using these images, retain their spatial attributes. 

Images are processed and annotated with STAC metadata, with each folder containing its own collection.

The following datasets are available:

main datasets:

hr_masks: 2.5m resolution cadastral masks with building footprints

hr_orthophoto: 2.5m resolution orthophotos of Austria

lr_s2: 10m resolution Sentinel-2 images of Austria (temporally and spatially aligned with the other data sources)

image_samples: samples dataset representing the structure of this data repository

building_delineation_inference: building delineation masks extracted from super-resolved or interpolated Sentinel-2 and orthophoto images

metric_results: results from the conducted experiments on presented metrics

stratification_tables: train/validation/test splits for different dataset configurations

super_resolved: super-resolved Sentinel-2 images (from lr_s2) output from all used SR models

tracasa_evaluation: dataset to achieve evaluation on a small subset for proprietary SR models

thesis_figures: figures and plots featured in the written thesis

This dataset contains only the image data and results, code repositories are available on the linked GitHub repository.