Computer Vision's Model Customization is a custom model training service that allows users like developers to easily train an image classification model (Multiclass only for now) or object detection model, with low-code experience and very little understanding of machine learning or computer vision required. The service is available in regions: West US 2, East US, West Europe.
This is a sample repository demonstrating how to train and predict a custom model with Cognitive Service for Vision, using Python. To get started, check out this tutorial in Python notebook.
Moreover, you can use your familiar Azure Machine Learning (AML) environment and Data Science Tools (AML Data Assets or Jobs, CLIv2 or MLFlow) for training and evaluating your custom Florence model following this documentation.
One of the scenarios we would like to introudce as a model customization scenario is Product Recognition. Computer Vision's product recognition service has been designed to be used in retail scenarios, where users would like to detect products, such as Consumer Packaged Goods (CPG), on a shelf. It comes with a set of APIs, a pre-built AI model, and a custom AI model that can be trained following the model customization guide above. You can try these out by following tutorials in Python notebooks:
Refer to export_cvs_data_to_blob_storage.ipynb for instructions or directly run export_cvs_data_to_blob_storage.py to export Custom Vision images and annotations to your own blob storage, which can be later used for model customization training.
Once data is exported, you can use it with Cognitive Service Vision Model Customization.
If you would like to explore more functionalities offered in Cognitive Service Vision, you can refer to this link for a quick start.
For frequently asked questions or quick troubleshooting, check out FAQ, including things like troubleshooting guides, quota information, etc.
For more documentation, check out: