GAIA — Satellite Image Dataset

GAIA is a dataset of annotated satellite images covering the five spheres of Earth (atmosphere, biosphere, etc.). The data is enriched with geographic metadata, tags, and captions, making this corpus relevant for environmental research, global monitoring, and AI model training.

Download dataset

Size

36,000+ satellite images (JSON + .png reconstructions via img2dataset)

Licence

MIT

Description

‍

GAIA is a dataset of web-scraped satellite images, enriched with metadata such as location, satellite used, shooting methods, resolutions, and text legends. The dataset is organized by environmental themes according to the five terrestrial spheres: atmosphere, hydrosphere, biosphere, geosphere and cryosphere.

‍

What is this dataset for?

‍

Training multimodal image/text models for the analysis of natural scenes
Analyzing environmental phenomena on a global scale (e.g. deforestation, marine pollution)
Serve as a basis for geospatial computer vision and intelligent mapping projects

‍

Can it be enriched or improved?

‍

Yes, the dataset is designed to be extensible. It is possible to add new image/text pairs, to enrich the metadata with climate data, or to convert the files into other formats (e.g. COCO). It can also be adapted for object detection, segmentation, or temporal analysis.

‍

🔎 In summary

Criterion	Evaluation
🧩 Ease of use	⭐⭐⭐✩✩ (Medium – requires reconstruction with img2dataset)
🧼 Need for cleaning	⭐⭐⭐⭐⭐ (Low – well-structured but unfiltered data)
🏷️ Annotation richness	⭐⭐⭐⭐⭐ (High – localization, tags, captions, resolutions)
📜 Commercial license	✅ Yes (MIT)
👨‍💻 Beginner friendly	⚠️ Moderate – requires command-line tools
🔁 Fine-tuning ready	🎨 Yes, for vision or multimodal tasks
🌍 Cultural diversity	🌎 Very good – planetary coverage with diverse contexts