How to see images that are not recognized by the detector model?

AI Discussions
1

We can create and train an Object Detection model based on Training Custom Object Detector — TensorFlow 2 Object Detection API tutorial documentation

There is also a file for evaluating the trained mode ==> Training Custom Object Detector — TensorFlow 2 Object Detection API tutorial documentation

But how can I view images in the test dataset that were not recognized by the detection/pictures on which objects were not detected?

1 Like
2

I’m not familiar with that course, but the general method is that you have the labels (the Y values) and the predictions of the model (the $\hat{Y} values). The incorrect predictions are the cases in which Y \neq \hat{Y}, so that expression in python gives you a Boolean vector that you can use to enumerate the incorrect entries:

images[labels != predictions]

Adjusted for the variables names in your implementation, of course. And some fiddling may be required prior to that statement in the multiclass case: you’d want both labels and predictions in categorical form, not “one hot” form to make that work.

1 Like
3

Thank you, paulinpaloalto

But I was thinking… it is possible that the “Object detection API” already has methods for visualizing incorrectly recognized images

1 Like
4

You may be right. As I mentioned in my original reply, I am not familiar with that course or that assignment. You’ll need to read the documentation that is available in that course and assignment and see what you find.

1 Like