Synthetic Data For Real Problems

Computer vision is everywhere! But teaching an algorithm to identify objects requires a lot of data and this is definitely the case when we think about GeoAI

But it is not enough to have a lot of data we also need data that is labeled

If we are looking for cars in images we need a lot of images of cars and we need to know which pixels are the car!

Of course, I am oversimplifying but I hope you get the idea,

Now imagine that you can automatically generate a large labeled data set of realistic images of cars based on the specifications of a specific sensor.

These data sets are often referred to as synthetic data or fake data and to help us understand more about this I have invited Chris Andrews (linkedin.com/in/chrisjandrews) from Rendered AI (rendered.ai) on the podcast.

Here are a few previous episodes you might find interesting

Computer Vision And GeoAI

mapscaping.com/podcast/computer-vision-and-geoai

In this episode, the discussion is aimed at an increased understanding of the differences between computer vision and the AI that is used in the Earth Observation world.

Labels Matter

mapscaping.com/podcast/labels-matter

What it takes to create labeled training data manually. If you are new to the idea of labeled data sets this is a good place to start.

Fake Satellite Imagery

mapscaping.com/podcast/fake-satellite-imagery

This is a good episode if you want to know more about Generative AI and Generative Adversarial Networks.

Also, check out this website thisxdoesnotexist.com to get an idea of where and how these Generative Adversarial Networks can be used. Look for a website called This City Does Not Exist thiscitydoesnotexist.com

On a silently similar note try uploading an image to bard.google.com … it's pretty interesting!