Trouble Getting Accurate Results from Custom Object Detection Model

Hi everyone,
I’m working on a custom object detection project using Edge Impulse, targeting a low-powered microcontroller. I’ve followed the tutorial to collect and label images, trained the model, and deployed it successfully. However, the model seems to struggle with accuracy—it either misses objects or detects incorrect ones.

Here’s what I’ve tried so far:

  • Balanced dataset (approx. 300 images per class)
  • Good lighting conditions during data collection
  • Data augmentation enabled

Still, the results are inconsistent. Could this be an issue with image resolution, model choice, or something else?

Any suggestions would be greatly appreciated.

Hello @auroralane7754 first of all welcome to the Edge Impulse community!

Could you please confirm that your low-powered MCU is taking images with the same quality or light that you used to train the model? Actually did you use the same camera to train the model?

Let us know if you used the same camera for training the model.

Thanks