🗺️ Major TOM: Expandable Datasets for Earth Observation

A standard for curating large-scale (Terabyte-scale) EO datasets.

This repo currently provides some basic functionality and examples for interacting with Major TOM datasets. This will grow and change as more datasets are created.

🎞 Official Image Datasets

Dataset	Modality	Number of Patches	Sensing Type	Comments
Core-S2L2A	Sentinel-2 Level 2A	2,245,886	Multi-Spectral	General-Purpose Global (about 23 TB)
Core-S2L1C	Sentinel-2 Level 1C	2,245,886	Multi-Spectral	General-Purpose Global (about 23 TB)
Core-S1RTC	Sentinel-1 RTC	1,469,955	SAR	General-Purpose Global (about 16 TB)
Core-DEM	Copernicus DEM 30	1,837,843	Digital Surface Model (DSM)	General-Purpose Global (about 1 TB)

📊 Official Embedding Datasets

Dataset	Modality	Number of Embeddings	Sensing Type	Source Dataset	Source Model	Size
Core-S2L1C-SSL4EO	Sentinel-2 Level 1C	56,147,150	Multi-Spectral	Core-S2L1C	SSL4EO-ResNet50-DINO	252.9 GB
Core-S1RTC-SSL4EO	Sentinel-1 RTC	36,748,875	SAR	Core-S1RTC	SSL4EO-ResNet50-MOCO	332.5 GB
Core-S2RGB-DINOv2	Sentinel-2 Level 2A (RGB)	56,147,150	True Colour	Core-S2L2A	DINOv2	223.1 GB
Core-S2RGB-SigLIP	Sentinel-2 Level 2A (RGB)	20,212,974	True Colour	Core-S2L2A	SigLIP-SO400M-384	41.3 GB

🔭 Demo

You can view samples from the MajorTOM-Core dataset instantly in your browser here: https://huggingface.co/spaces/Major-TOM/MajorTOM-Core-Viewer

🌐 Generate uniform grid across the globe

In case 10 km grid is not appropriate, you can easily define Major TOM grids at different scales by using grid.py script to generate a uniform grid accross the globe:

dist = 100 # sets distance in km
grid = Grid(dist) # returns grid points coordinates

📌 Open Access Manuscript

This project has been outlined in https://arxiv.org/abs/2402.12095/.

Read Abstract

Deep learning models are increasingly data-hungry, requiring significant resources to collect and compile the datasets needed to train them, with Earth Observation (EO) models being no exception. However, the landscape of datasets in EO is relatively atomised, with interoperability made difficult by diverse formats and data structures. If ever larger datasets are to be built, and duplication of effort minimised, then a shared framework that allows users to combine and access multiple datasets is needed. Here, Major TOM (Terrestrial Observation Metaset) is proposed as this extensible framework. Primarily, it consists of a geographical indexing system based on a set of grid points and a metadata structure that allows multiple datasets with different sources to be merged. Besides the specification of Major TOM as a framework, this work also presents a large, open-access dataset, MajorTOM-Core, which covers the vast majority of the Earth's land surface. This dataset provides the community with both an immediately useful resource, as well as acting as a template for future additions to the Major TOM ecosystem.

If you found this useful for your research, please cite accordingly as:

@inproceedings{Major_TOM,
  title={Major TOM: Expandable Datasets for Earth Observation}, 
  author={Alistair Francis and Mikolaj Czerkawski},
  year={2024},
  eprint={2402.12095},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}

Powered by Φ-lab, European Space Agency (ESA) 🛰️

---

FAQ

Is Major TOM just another EO dataset?

Almost. Major TOM is not a dataset, but a project aiming to standardize some of the future EO datasets. As an example of what such a dataset could be like, MajorTOM-Core is released as a nearly global dataset of Sentinel-2 data.

Scroll up to the 📊 Available Datasets section of this file to see the list of current datasets.

Who is going to contribute to upcoming Major TOM datasets?

Anyone can contribute. The original authors of the Major TOM paper are already working on a few other datasets that will join the Major TOM initiative.

Can I join Major TOM organisation on HuggingFace?

Anyone can join the organisation with reading rights. In order to gain contributor rights, you will need to contact one of the admins and verify who you are and how you would like to contribute (you should be allowed to contribute with any dataset that follows Major TOM standard).