Introduction

High-quality maps are foundational to humanitarian response, climate adaptation, public health, and inclusive development. Yet many regions particularly across Low and Middle Income Countries (LMICs) remain under-mapped, and the local communities that need this data most are too often left out of the GeoAI revolution: absent from the datasets models are trained on, and disconnected from the tools that could put GeoAI to work in their own contexts.

HOT is not setting out to solve this alone. Through our fAIr initiative, our role is that of a catalyst bringing together the people building GeoAI models with the open mapping communities ready to use, validate, and improve them. fAIr is the connective tissue: an open, community-driven platform where GeoAI models meet real-world mapping needs, without requiring users to be AI/ML engineers.

To bring this vision to life, we need GeoAI models and we need the GeoAI community to help build it with us.

This is an open invitation to collaborate without an end date

The Vision Behind the Call

GeoAI models are being developed in research labs, universities, and companies all over the world. At the same time, open mapping communities in LMICs are eager to generate the data they need for sanitation, health, agriculture, environmental monitoring, and disaster.

Communities mapping their own neighborhoods don't have the technical pathway to benefit from them. And without local feedback, models don't improve where they're needed most.

Who Can Participate

We welcome contributions from across the GeoAI ecosystem, including:

• Universities and academic research groups
• Research centers and non-profit labs
• Businesses and start-ups developing GeoAI solutions
• Independent GeoAI researchers and open source contributors

If you are building GeoAI models that can help map our world, this call is for you.

GeoAI Models We Are Looking For

We are seeking open source GeoAI models across the following types and not limited to them.

All models are expected to perform prediction tasks on high resolution satellite + aerial imagery ( < 1 meter ground resolution )

Model expectations

The GeoAI models are expected to have:

• Clear open source code for
  • Pre processing : fAIr provides your GeoAI model with a list of images and GeoJSON with list of Geo features and their attributes
  • Inference : the model can be deployed and run on new imagery out of the box.
  • Fine-tuning : the model architecture and training pipeline support further training on new geospatial data.
  • Post Processing Convert your raw model output to decoded predictions (labels / masks / boxes) that can be consumed by fAIr
• Open source model weights
• Can be described or following using STAC-MLM https://github.com/stac-extensions/mlm
• Matches the scope as of the contributing documentation https://hotosm.github.io/fAIr-models/contributing/model/

Eligibility & Evaluation

To be accepted, GeoAI models must:

1. Be open source, with a license that allows use by the open mapping community.
2. Include a model card describing architecture, training data, intended use, limitations, and performance metrics.
3. Pass the tests described in the fAIr model contribution guide: https://hotosm.github.io/fAIr-models/contributing/model/
4. Pull Request in fAIr-model repository is approved and merged

What We Offer

Funding

USD 3,000 per accepted GeoAI model as a contribution toward your work.

Access to Open, Human-Verified Geospatial Datasets

For GeoAI models that support fine-tuning, contributors will gain access to open source geospatial datasets that are collaboratively collected and human-verified in partnership with the MapSwipe community. These datasets are tied to the model's task and become available once the model is adopted in fAIr and used for fine-tuning within fAIr.

Your model can keep improving on data that has been validated by volunteers on the ground including from geographies that are typically under-represented in GeoAI training sets.

Human Feedback Loop

Your GeoAI model will receive human feedback on its predictions from mappers using fAIr a uniquely valuable signal for understanding real-world performance in LMICs.

Visibility

Accepted GeoAI models will be showcased to the global open mapping community, with full attribution to you and your institution.

How to Participate

1. Submit your proposal including the following to fair@hotosm.org using the subject line: [GeoAI Research];
   • Your model card/details
   • Which of the GeoAI model type you are contributing; ref: the Model Scope above
   • Whether the model supports inference, fine-tuning, or both
   • A link to the model repository and evaluation dataset; if any
   • A short description of your team or institution/organization

1. fAIr team will review your proposal and reply to you; the team preserve the right to reply to only selected proposals in case of high volume of proposals/submissions;
2. If selected; collaboration agreement is signed as of the next section

For any questions about this open call, the platform, or the collaboration process, please reach out to the same address: fair@hotosm.org with subject line [Question: GeoAI Research]

Collaboration & Deliverables

Selected contributors will enter into a collaboration agreement with HOT covering:

• Funding
• Licensing and attribution
• Adaption of the GeoAI model within the fAIr platform as a delivery

Let's collaborate.

The HOT fAIr Team

fair@hotosm.org hotosm.github.io/fAIr-models