Notice: Microsoft has announced the planned retirement of Azure AI Custom Vision. The service will be fully supported for existing customers until September 25, 2028; after that, calls to the service will fail. For new projects, Microsoft recommends Azure Machine Learning AutoML, generative approaches via Foundry models, or Azure Content Understanding in Foundry Tools (preview), depending on the use case.
What is Azure AI Custom Vision?
Azure AI Custom Vision is an image recognition service for training custom image classification and object detection models without deep ML expertise. You upload images, label them with your own tags, and train a model that recognizes your specific objects. The service is part of Foundry Tools and is optimized to produce initial results quickly with comparatively few training images.
Custom Vision offers two capabilities: image classification assigns one or more labels to an image. Object detection additionally localizes the position of detected objects in the image via coordinates. The service is optimized for clearly distinguishable visual differences; according to Microsoft, it is not optimal for very subtle distinctions (for example, minor cracks or dents).
Since Microsoft has announced the retirement of the service, new projects should evaluate the recommended alternatives, though existing implementations can continue to run until 2028.
Core Features
- Image classification for single-label and multi-label scenarios
- Object detection with localization of detected objects
- Training algorithms with domain variants optimized for certain subject types (e.g., landmarks, retail items)
- Usage via web portal (customvision.ai), REST API, or SDKs
- Export of trained models for offline use
Typical Use Cases
Manufacturing companies use Custom Vision for prototypes of automated quality control, for example to detect obvious defects or missing components.
Retailers use object detection for early shelf-analysis prototypes to recognize products or coarse out-of-stock situations from camera images.
Medical technology companies occasionally use the service for early prototypes of screening tools, before developing production-grade solutions with specialized or generative models.
Benefits
- Fast entry point without deep ML expertise required
- Training in a short time with comparatively few images
- Web portal for easy labeling and testing without code
- Export options for offline and edge scenarios
Integration with innFactory
As a Microsoft Solutions Partner, innFactory supports you with Azure AI Custom Vision as well as migration planning to Azure Machine Learning AutoML, Foundry models, or Azure Content Understanding: use case analysis, data strategy, model training, and migration strategy.
Contact us for a non-binding consultation on Azure AI Custom Vision and its successor solutions.
Typical Use Cases
Frequently Asked Questions
Is Azure AI Custom Vision still available?
Yes, the service continues to run and Microsoft provides full support until September 25, 2028. Microsoft has announced its planned retirement and recommends transitioning to alternatives during this window.
How many images do I need for training?
As a rule of thumb, Microsoft recommends around 50 images per label for good prototyping results. For production-grade models with high variance (lighting, angles, background), significantly more images per class are usually advisable.
What is the difference between Classification and Object Detection?
Classification assigns one or more labels to an entire image. Object Detection additionally localizes where in the image the detected labels appear, including coordinates. The choice depends on your use case.
What alternatives does Microsoft recommend for new projects?
For classic ML approaches, Microsoft recommends Azure Machine Learning with AutoML for image classification and object detection. For generative approaches, Foundry models and Azure Content Understanding in Foundry Tools (preview) are available, which is suitable for image classification workflows, among other things.
Can I export trained models?
Yes, models can be exported for offline use, for example for deployment on edge devices. See the official documentation for details on supported export formats.
Note: All product information on this page has been compiled with care, but is provided without guarantee and may be outdated or incomplete. Cloud services evolve rapidly — features, pricing, SLAs, and availability change frequently. Authoritative and up-to-date information can only be found on the official product page of Azure (official documentation). This page does not represent an offer by Azure.
