Custom Image Analysis model - mistakenly identifies products

Question

Hi folks, I really need some help and guidance.

I'm training a custom image analysis model for a product. I'm following these guidelines - https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/how-to/model-customization?tabs=studio
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/how-to/shelf-model-customization

Unfortunately after several models that I trained and different configuration of the datasets, I do not see any improvements in the model's output.

The one dataset consists of 41 images and other one of 19. I also trained a model with pictures of the product itself. I've trained a model for the budget of 1 hour, and last night I trained a model with a budget of 7 hours.

However, when I overlay the output results over an image I'm getting absolutely the same results - attaching the pictures. The model is trained on products with green arrows but it mistakenly identifies shelf products with white packaging as the same.

Answer 1

Hello @nikola , Thanks for using Microsoft Q&A Platform.

Try to increase the size of your dataset, so that model will learn the features that distinguish between different products.

You can also try adding more quality image datasets to improve the accuracy of your model.

If you still face issue, for the deeper investigation on this issue please raise a support ticket in the Azure portal.

Regards,

Vasavi