Skip to content

AutoGeoAI4Sci/GeoSR-Bench

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 

Repository files navigation

GeoSR-Bench

Dataset and code for the paper:

Beyond Visual Fidelity: Benchmarking Super-Resolution Models for Large-Scale Remote Sensing Imagery via Downstream Task Integration [arXiv]

GeoSR-Bench is a large-scale benchmark for evaluating remote sensing super-resolution models beyond conventional image fidelity metrics, such as PSNR and SSIM. GeoSR-Bench evaluates whether super-resolved images improve downstream Earth observation tasks.

The complete GeoSR-Bench dataset is available on Hugging Face: https://huggingface.co/datasets/ai-spatial/GeoSR-Bench

Benchmark Tasks

GeoSR-Bench covers two cross-platform super-resolution settings:

  • MODIS → Landsat-8
  • Sentinel-2 → NAIP

It also includes multiple downstream task datasets for evaluating the practical utility of super-resolution outputs, such as:

  • Land cover classification
  • Road and building mapping
  • Crop mapping
  • Water mapping
  • Gross primary productivity estimation
  • Canopy height estimation

Repository Status

The trained models will be released soon.

About

Dataset and code for the paper "Beyond Visual Fidelity: Benchmarking Super-Resolution Models for Large-Scale Remote Sensing Imagery via Downstream Task Integration"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 92.0%
  • Jupyter Notebook 7.7%
  • Cuda 0.2%
  • C++ 0.1%
  • Dockerfile 0.0%
  • MATLAB 0.0%