customvision.ai (Azure AI Custom Vision)

Custom Vision is built around a deliberately simple loop. You upload labeled images — or use the service's assisted tagging to label them quickly — train a model against those examples, then review precision and recall scores on a test set before publishing. Once published, the model is available as a prediction endpoint that can be called over HTTP with a simple REST request, returning predicted tags and confidence scores for new images.

The service supports two problem shapes: multiclass classification, where each image gets exactly one label, and multilabel classification, where an image can carry several tags at once. For developers who need to run inference outside the cloud, models can be exported for offline scenarios.

Custom Vision is part of the broader Microsoft Cognitive Services family and is accessed with a Microsoft account. Because it has been folded into the newer Azure AI services lineup, new users are generally directed toward Azure AI Vision resources for provisioning keys and endpoints, while the customvision.ai portal remains the training and iteration surface. Pricing follows Azure's consumption-based Cognitive Services model, with a free tier for experimentation.