Feature Vs Model Accuracy

I am trying to use ESP32 Cam for classifying the images (Stop or Ok lables).
Project - https://studio.edgeimpulse.com/studio/79307

After capturing images ( QVGA 320 x 240 - Negative effect style images) given good distinguished features -

But when model is trained its accuracy is not up to the mark , what would be missing in this case ?

Any help would be highly appreciated :slight_smile:

@atyadav you have only 53 items, I will recommend to increase images and introduce 3rd label images as โ€œnoneโ€ images which contain images which are not โ€œokโ€ and โ€œstopโ€ images and will act as background subtractors or in simpler terms will increase accuracy in transfer learning.

I hope this helps.

@AnshumanFauzdar - Yes I also added third label , but that did not help. But have to add few more images.

@atyadav increasing images will definitely increase your accuracy and do model testing with test data and verify results!