Awesome-Remote-Sensing-Foundation-Models

[brunosan]

This page is a really nice curated collection of EO foundational models, which Clay is too. So it makes sense to relate the work linked here to us, and see how we can improve our work to be most useful.

The first thing to me is precisely that there are no standard ways to compare how good a EO foundational model is. Similar benchmarks for text and images do exist and function as the standard. But not for EO. There are several attempts at that, but none has taken over the standard everyone (providers or users) could use to quickly compare performance.

They are mostly very recent, and the pace of new ones is indeed increasing.

Without valuing the merit of these, which I hope to find time to digest here with time, I believe Clay is (or will be once we release v0) unique in offering all these:

• Open
  • Open source. The code is fully open source, without limits. Apache or MIT.
  • Open data. The model uses fully open data, which greatly simplifies deploying the model, when available without uncertainty or license payment.
  • Open model. The model weights are openly available and released with a license that has no limits.
• Deployment ready
  • With a peer-review paper and/or documentation to explain and evaluate the work done, scope, challengues, ...
  • There is documentation and examples to fully replicate, adapt and/or finetune the model fo particular needs.
  • There is a team dedicated to developing, supporting and promoting an open ecosystem around it.
• By a dependable Non-profit
  • Legally and functionally committed to public benefit.