Best Dataset Size for Object Detection Model and AI Labeling in Edge Impulse

Help
,
#1

Hello Edge Impulse community,

I’m currently working on an object detection model to detect person, car, truck, fire, and smoke. I would like to ask for your guidance on two key aspects of this project:

  1. Dataset Size: What is the ideal dataset size for training an object detection model to recognize the aforementioned objects? Should I aim for a specific number of images per class, or is there a general rule of thumb for dataset size when working with Edge Impulse?
  2. Labeling Process: I noticed that Edge Impulse offers an AI-based labeling option. Can anyone share their experience with this feature, especially for object detection? How accurate is it, and are there any best practices to follow when using this tool to label objects like people, vehicles, or fire?

Any advice or experiences you can share would be highly appreciated! Thanks in advance for your help.

#2

Hi Anis, Not sure on point 1. I have about 233 total images across 4 different objects. Point 2 I have recently been trying to use the AI labelling function for object detection of wheelie bins. In order to use any of the GPT-4o functions you need an Open AI account and to have paid for credits in order for the function to work otherwise the labelling returns an error. Playing round with the thresholds can help and I think things like shadows and reflections can affect object detection. Anyway, best of luck.