If you have never done a graduate course in Satellite Remote Sensing there are a lot of techniques that you use on satellite imagery that the LLM guides you through without knowing it upfront. Additionally having a way to have the imagery loaded with the LLM in a single interface and apply the LLMs suggestions isn't a trivial engineering problem. We have done that too.
Thanks mate! Yes we are working now with remote sensing people to have more targeted end-to-end use cases. So flooding is one another would be environmental impact of mining. Can you email me on [email protected] would love to incorporate more feedback
Thanks for the feedback! Yep you are right - actually in the intro video[1] i talk about hallucinations too, as we are very alpha. The good thing is there's a panel on the right that lets you overwrite the AI's selection or you can even ask the LLM to change it. I've update the params manually so if you go to the demo now you will see another color map - Virids :-)
[1] https://www.youtube.com/watch?v=ikEDbnUVfYQ
After spending years building remote sensing analysis products I decided it's time to bring LLMs in the mix and let them see the imagery along with context of spectral bands to do the heavy lifting!
ah yes we have been testing other embedding models but not google's. I'll try this too. Its interesting most of them are doing land cover classes which is kinda solved already. We are also testing mixing agenic workflows with smaller directed prompts for users to provide the classes. Incidentally we are Berlin based. We should grab a coffee :)
Thanks! This live demo uses metadata and stats only. Right now we are testing ViTs and Foundation Models as well. But quality of results from EO FMs haven't been worth the inference cost so far. Early days though. Also starting to fine tune models for specific downstream tasks ourselves.
We've been working on this challenge in the satellite domain with https://earthgpt.app. It’s a subset of what Fei-Fei is describing, but comes with its own unique issues like handling multi-resolution sensors and imagery with hundreds of spectral bands. Think of it as computer vision, but in n-dimensions.
Happy to answer questions if you're curious. PS. still in early beta, so please be gentle!
I am working on https://geobase.app/ which is a platform for geospatial full-stack developers.
We have created workflows that a specific to the geospatial, mapping and GIS industry use cases. This is currently in private beta but going live in a few weeks. It is built on top of supabase's self-hosted stack.
Apologies for the shameless plug but do checkout https://geobase.app We built it with such use cases in mind and solves the scaling issues with geo data driven apps. We are in private beta but happy to give you early access if you hit us up.