Image Classification using 3 different classes

Model training accuracy is 0%
After model training, my samples only point to one data set ( the bottle); however, they do not create points for the other classes. I’m not sure if my pictures are correlated to the accuracy

Project ID:
626059

Context/Use case:
I’m working on a project involving the object classification of different floating trash in outdoor water ( bag, bottle, can), and I’m simulating the water with an open box filled with water for image classification realism

Steps Taken:

  1. Placed objects in water took around 60 photos with phone for training in multiple angles/ perspective/object arrangments
  2. took 20 more for the test set and created impulse following the lamp/coffee edge impulse guide video
  3. Generated features and arrived at only 1 data set (bottles)
  4. tried going through with retraining and testing ( failed)

Expected Outcome:
I want to achieve at least 80% accuracy with my object detection of trash that can be found in different environments and arrangements in the water

Actual Outcome:
both data sets failed with no object detection

Environment:

  • Platform: my mobile phone and the website ( beginner programming)
  • Build Environment Details: N/A
  • OS Version: Windows 10, 64bit
  • Custom Blocks / Impulse Configuration: I used 320x 320 for image, image processing block, image detection learning block ( coffee lamp video guide)
    Logs/Attachments:
    [Include any logs or screenshots that may help in diagnosing the issue]

Hello @HamzaSalam4,

I quickly checked your project and it seems that you are using the FOMO architecture for object detection. However, your object takes a large portion of your image. FOMO is best for smaller objects that takes roughly 1/8th of the image.

I would suggest to try the MobileNetv2 SSD architecture, you can change that directly in the Object Detection page, under the choose model button.
See: Object detection | Edge Impulse Documentation for more informations.

Let me know if that provides better results.

Best,

Louis