Object detection model output


I’ve trained a custom object detection model to detect and differentiate between two objects and it has been succesfully trained. However, when I open the model on Netron, it’s output layer contains some weird output.

What can I do to get the conventional bounding boxes and the class probability outputs from the model from either Edge Impulse or some external post processing tool?

Weird output is attached with this post
EI Model

Can you share the project ID?

@rjames do you notice any issues here with the output, or are you able to help with the bounding box query?



Sure here’s the project: airpods-mario - Dashboard - Edge Impulse


The output format of our YOLOv5 learn block is as follows:

(xcenter, ycenter, width, height, score, cls...)

where cls.. represents your class probabilities. In your case there’s 2 classes.

You can use your own processing tool or use our inferencing SDKs for end-to-end processing of input data, inferencing and post processing.

Thanks @rjames this was helpful! Also @Eoin

I’m facing an error with the same project: when I try to train a Yolov5 model again with a higher input resolution, it shows me the following error:

“DeadlineExceeded - this job took longer than is allowed. Enterprise users can change the job time limits on the Dashboard.
Job failed (see above)”

Never mind, I found there’s a 20 min time limit. Correct?

Hello @karkapur,

Yes, there is a 20 min limit per job on for our community tier.