Hello everyone,
I’ve successfully trained Edge Impulse to recognize two different objects (F1 score = 1), and the object detection works well with these. However, I’ve noticed an issue: when I hold additional objects that look similar to the ones I trained, the system mistakenly identifies them as the trained objects.
Does anyone have tips on improving the model’s accuracy to avoid these false identifications, especially with similar-looking objects? Thanks in advance for any advice!
You need to introduce a third category “background” (name it however you want), and include random pictures where the two objects are not occuring.
Thank you for the answer. I tried it out, but the F1 score is now low, because I took pictures of many different Objects and classified them as one. Do I have to take more pictures or how do I solve this issue?
I would take more random pictures to the “wrongObject” class. If you connect your mobile to Edge Impulse, it’s easy to take tens or hundreds of random pictures in a very short time. I usually shoot whatever happens to be around me, obviously ensuring no persons or sensitive material are being photographed.
