SatImgNet

← Back to all benchmarks

Date: 2023-04-23

Name: SatImgNet

Domain: Climate & Earth Science

Focus: Satellite imagery classification

Task Types: Image classification

Metrics: Accuracy

Models: CLIP, BLIP, ALBEF

AI/ML Motif: Multimodal Reasoning

Resources

Benchmark: Visit

Datasets: SatImgNet on Hugging Face

Results: SatImgNet Leaderboard

Keywords

land-use zero-shot multi-task

Citation

• Jonathan Roberts, Kai Han, and Samuel Albanie. Satin: a multi-task metadataset for classifying satellite imagery using vision-language models. ICCV Workshop: Towards the Next Generation of Computer Vision Datasets, 3 2023. doi:10.48550/arXiv.2304.11619.

@article{roberts2023satin,
author = "Roberts, Jonathan and Han, Kai and Albanie, Samuel",
title = "Satin: A multi-task metadataset for classifying satellite imagery using vision-language models",
year = "2023",
month = "3",
journal = "ICCV Workshop: Towards the Next Generation of Computer Vision Datasets",
doi = "10.48550/arXiv.2304.11619"
}

Ratings

CategoryRating

Software

0.00

No scripts or environment information provided

Specification

4.00

Tasks (image classification across 27 satellite datasets) are clearly defined with multi-task and zero-shot framing; input/output structure is mostly standard but some task-specific nuances require interpretation.

Dataset

5.00

Hosted on Hugging Face, versioned, FAIR-compliant with rich metadata; covers many well-known remote sensing datasets unified under one metadataset, though documentation depth varies slightly across tasks.

Metrics

5.00

Accuracy of classification is an appropriate metric

Reference Solution

4.00

Baselines like CLIP, BLIP, ALBEF evaluated in the paper; no constraints specified

Documentation

5.00

Paper provides all required information

Average rating: 3.83/5

Radar plot

Edit: edit this entry