Catherine Nakalembe

With those images, fields are misrepresented, so there are places where there are no crops but are labeled as crops.

There are places where there should be crops and people that are completely missing.

When you go and try to assess something for the ministry and you use that as input, you're basically feeding them garbage.

You want to make sure that what you're feeding in is really good, and that actually requires work.

To train a model to understand that complexity, you need a lot of examples.

You have to go on the ground.

What we did is we used GoPros.

You wear a GoPro as you're driving on a motorcycle, or you can do it in the car.

And as you drive, we take pictures.

Basically, Google Street View, not for streets, but for crops, so the camera is actually facing towards the field.

and then adapt basically what would be face detection.

But instead of detecting faces, cats, dogs, et cetera, we modified it to detect maize, beans, cassava.

We covered all of Western Kenya in two weeks with just two teams.

collected over five million images, a lot of them with volunteers, everyday motor taxi drivers, students.

This would allow us to build a more complex model that can learn from all these different examples from all the different contexts.

There was a flood in Kenya in 2024.

It happened really, really rapidly.

And, you know, the entire country was affected pretty much.

I got an email from the Ministry of Agriculture asking to do an assessment using satellite data to look at where floods happened, where were crops and give an estimate of what the total area of cropland that's been affected was.

And then what the ministry does with the information is they make their response programs, which is where do we need to go to provide seeds so people can replant, as an example of an action that was taken.