Custom vision - Impact on model when training for different hours

Antoine CHABERT 21 Reputation points
2021-06-08T09:52:52.69+00:00

Hi all,

I was not able to find any sort of information about difference in algorithms/layers of DNN when we train the model for different hours. For example: What is the approach if I train model for 2 hours, 10 hours or 24 hours.
In fact, what does it changes for the model if you train it for 1 hour or 24 hours?
Is there any guideline that says (for example): If you have 50 images per label (for a binary classification) you should train your model for at least 2h? And so on.

Thanks a lot for your help.
Antoine CHABERT.

Azure AI Custom Vision
Azure AI Custom Vision
An Azure artificial intelligence service and end-to-end platform for applying computer vision to specific domains.
220 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,396 questions
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. romungi-MSFT 42,286 Reputation points Microsoft Employee
    2021-06-08T13:19:33.557+00:00

    @Antoine CHABERT Based on my experience with the service some of the details regarding like algorithms, training cycles, iterations used in the backend of the service are not made public so there is no documentation on the internals of the service.

    With respect to training time the guidance on allocation of time is based on quality of images and dataset. Based on training budget the settings are decided for training that is a best fit. There is no specific relationship between the number of training images and training hours.

    0 comments