Hello.
I have setup an Object Detection Impulse with 6 features images @96x96 px. I am using an iPhone 8 with the “Run This Model” test from the dashboard using the QR code. There are times when a particular feature (Let’s call it “Feature-A”) is not on screen and a specific “non-feature” object is detected as Feature-A. If both objects are on the screen at the same time, there is no issue and Feature-A is detected @ 0.98-1.0 accuracy. I read through https://docs.edgeimpulse.com/docs/tips-and-tricks/increasing-model-performance to see if there was any advice. Would this be from lack of data? Also, is there a way to add negative data to the detection, i.e.: register pics with the tag “if anything is detected in this pic, nothing should be detected” or “if this particular object is encountered, it does not match any of the registered features so do not detect”? I am not encountering the issue with any of the other registered features. Please let me know if you have any advice. Thanks.
Hi @MLawless
We will need to see some screenshots and examples to comment more than the guidance we have in the docs.
Best
Eoin