I am working on an object detection project using Edge Impulse.
I trained a model to detect a specific product, but I have some problems during testing:
When the object is detected, the model shows a circle around it, not a square.
The model sometimes detects the object even when there is no object present.
Sometimes, it detects multiple objects, even though there is only one object.
I want to understand: What is causing these problems? How can I fix them?
Thank you for your help!
Can you please share some screenshots of your dataset size and class balance. FOMO outputs circles not squares so you are likely using FOMO. There can be cases where objects that are close together appear as the same with the architecture of FOMO being for smaller classes of device the grid size can be large with higher res images.
Screenshots of your impulse configuration and your dataset details plus the training results so folks can give you a better answer.
Please also review this document it can help with understanding how to improve model performance:
