Lesson 2: building an image classifier that can recognize “not this type of image”

I’m working through the first few fast ai courses and I’m building my own classifier for my pet cat that I want to deploy at home. I’m having trouble understanding how to build a image classifier that detects any pictures that aren’t of my cat, not just pictures of another class that I upload (dog, bird).

By then I mean, I need to train on different classes to tell me it’s class A instead of a general “not cat” type. I wonder if theres a way to say, this model is only 30% confident that this picture your pet cat, so we can reject it.

Just getting started so still trying to wrap my head around fast ai/ deep learning.

Thanks in advance!

1 Like

The Is It A Bird? notebook does essentially this, right? Seems like a good starting point.

No actually this is the same notebook that I’m using already, it only shows the probability that the image is a bird or a forest. But if you were to say give it a frog as an impact, the sum of percentage of bird and forest would still be 100%, thus there’s no other category.

For anyone else looking, found the answer in this lesson here! (Thanks Zach)