Bounding box with Labeling method


Is there anyway to see what the detection model “sees” as the object when you use One Label Per Data Item labeling method to classify objects. I like the fact that the bounding box method shows where in the image the model is detecting the object, but would like something similar with One Label Per Data method.



Hello @ross6699,

Yes and no :slight_smile:
To achieve what you want, I would suggest that you use the data explorer and keep you images unlabeled. You can then set the label based on their distance to the cluster. This feature is called assisted labeling: Data explorer - Edge Impulse Documentation

It’s not quite the same as the labeling queue but it can hopefully help you.



Hello Louis,

Thanks for the quick reply.

That’s not quite what I was trying to achieve. The issue I have is that I cannot be sure that what the model is seeing is what I intended it to learn. To explain. I have an image of a plastic clip located in a clip housing on a part. I thought I was teaching the model the presence of the clip, but the model was learning the clip housing features. If I had a bounding box showing me that the model was looking at features I was not aware of it would have helped me in trying to teach the model about the clip only.

I hope that makes sense.



@ross6699 There may be other ways to do this but one way is to use Python. You can use this code to run your Model using Python and save what FOMO was seeing. I found this very useful to see how FOMO was locking in on un-wanted objects. See code lines 71, 72 and 73 to see how they are saving the FOMOed image (eh, a FOMOed image?) and possibly try and adapt to your device code saving to an SD card. If you figure this out please post your code.