Custom Vision Model

Question

To Whom It May Concern

Hello I am student from Department of Computer Engineering, Faculty of Engineering, Mahidol University from Thailand. I have proposed my thesis project with an implementation of Microsoft Azure in the Service of Custom Vision. Could you please provide me about the detail of the Machine learning model behind the Custom Vision in both Classification and Object Detection. What is the algorithms , Is it a Deep Learning or just normal machine learning model? and another question is about Advance Training, how it is different from quick training?

Thank you for take the time reading my question.

Sincerely.

Answer

Hi, according to the document, the Custom Vision service uses machine learning algorithms (I assume deep learning) to analyze images. Further details aren't available publicly and I haven't found any white paper regarding the service. However, I'm following up internally with the team, and will provide more details if available. Thanks!

Updates: Custom Vision uses different deep learning models in the system optimized for different purposes such as accuracy, training and inference speed, memory cost, etc. Quick training just trains with a few default configs that cost less for training, which is good for POC, i.e., reasonable performance given low cost. Advanced training will utilize the AutoML system in the backend, which gives better training performance, but costs more training time.

Share via

Custom Vision Model

1 answer

Your answer