Fields of The World (FTW)

Originally the core of the initiative, the FTW Benchmark Dataset is a global collection of labeled agricultural field boundaries, designed to benchmark model performance and support open research. It aggregates and harmonizes a number of open datasets into 1.6 million parcel boundaries and over 70,000 samples covering diverse agricultural landscapes across 4 continents and 24 countries.

You can explore and download the dataset from Source Cooperative.

What's included in the FTW Benchmark Dataset?

• Publicly available, labeled field boundaries from around the world
• Uniform data formatting and metadata schema
• A maintained subset of training/validation/test data
• Tools to evaluate model performance on shared benchmarks
• Contribution guidelines for submitting new datasets

Reference implementations of machine learning models trained on the FTW Benchmark Dataset. These models offer starting points and performance baselines for researchers and practitioners.

A growing suite of tools to support end-to-end workflows:

Data Access Tools

CLI to download and work with the benchmark dataset (FTW CLI, available in the ftw-baselines repository)

Inference Tools

CLI to run field boundary detection on new imagery (FTW CLI, available in the ftw-baselines repository)

Formatting Tools

Resources to convert raw data into FTW-compliant format (e.g., fiboa to FTW train/test split) using the fiboa CLI

QGIS Plugin

A plugin to browse and interact with field data within QGIS. Currently in alpha (ftw-qgis-plugin)

The goal for both the tools and the web app is to make it easy to run a number of field boundary models, not just FTW baseline models or FTW benchmark data derived ones.

A browser-based interface to easily run FTW models in any area of the world without running any code. The web app is powered by an API and deployed to AWS for global accessibility.

Improvements to the web app are under progress. The web app is powered by an API (see ftw-inference-api) and deployed to AWS via ftw-api-deployment.

Large-scale outputs of model-inferred field boundaries (e.g., country-wide field maps). While no full-country maps have been published yet, efforts are currently underway in Kenya, Vietnam, Laos, Colombia, and other countries/regions.

These outputs will be versioned and openly published to support external evaluation, planning, and downstream applications. As model quality improves, we aim to scale toward a global field boundary dataset.

Field Boundaries for Agriculture (fiboa)

The Field Boundaries for Agriculture (fiboa) project is focused on making field boundary data openly available in a unified format on a global scale.

Curated public datasets of field boundaries from government sources and hand-labeling campaigns, that may serve as inputs to the benchmark dataset. These are converted to the fiboa specification. Fiboa data can be found at https://source.coop/fiboa.

The data found in the Source Cooperative kerner-lab organization are in the fiboa format.

In the future, FTW datasets may be sourced from the fiboa datasets prior to FTW subsetting and harmonization.