Edge impulse trained model shows 100% prediction accuracy but no image is being shown

OpenMV Boards
1

oyster classification
oyster classification1366×768 201 KB

Greetings,

I have run through the edge impulse regimen for training an image classification and used 48x48 image size. Everything seemed to go smoothly except when I run the provided script.

It is showing 100% prediction accuracy on any image even when the lens cap is on. Why is this?

I trained the model on 100 different images of a single object I attained from an image scraper. Will the edge impulse training only work on images provided by the h7 plus cam itself?

Any ideas whats happening here?

Many thanks,

Matthew

2

Do you have more than 1 class? If you only have one class… then it will map everything to that one class.

The network is simply binning things into different buckets. 1 class means 1 output. If you have two classes then it can make a decision. But, realistically, you should always have the target classes you want and then a background class.

3

Yes I only have one class. Okay that makes sense. Thank you!