Custom Vision Training Client logo

Custom Vision Training Client

Custom Vision Training Client

Custom Vision Training API programmatically trains image-classification / object-detection models — create projects, upload labeled images, trigger training, publish iterations.

Visit site ↗Health checked 14h ago
Use it when

Fully programmatic — script the entire training flow

Watch for

No SLA on training time (typically minutes to tens of minutes)

First check

Create a Custom Vision Training resource in Azure for endpoint + training key. POST /customvision/v3.4-preview/training/projects to create, upload images, then POST /train to trigger training.

Auth
CORS
No
HTTPS
Yes
Signup
?
Latency
53 ms
Protocol
REST
Pricing

Uptime · 30-day window

Probes: 30Uptime: 100%Avg latency: 150ms
01

About this API

Custom Vision Training API is the programmatic counterpart of the visual training UI. Compared to the portal, the Training API's value is automation — e.g. a data pipeline that collects new labeled samples from production daily and auto-retrains and publishes a new version. The API covers project management, tag management, image+label upload, training trigger, iteration management, and publish operations. Supports classification (multi-label and single-label) and object detection (with bounding boxes). After training, inference goes through Prediction API.

02

What you can build

  • 1Auto-retrain models in CI/CD when new labeled data arrives
  • 2Bulk upload training data
  • 3A/B test multiple model iterations
  • 4Integrate training into ML pipelines
03

Strengths & limitations

Strengths

  • Fully programmatic — script the entire training flow
  • Multiple iterations per project for A/B comparison
  • Supports transfer learning (fine-tune from a general model)

Limitations

  • No SLA on training time (typically minutes to tens of minutes)
  • For complex tasks, Azure Machine Learning is a better fit
04

Getting started

Create a Custom Vision Training resource in Azure for endpoint + training key. POST /customvision/v3.4-preview/training/projects to create, upload images, then POST /train to trigger training.

05

FAQ

How do I use the model after training?+

Call publish iteration to push a specific iteration to the Prediction endpoint, then invoke via Prediction API.

How many iterations per project?+

Up to 20 iterations retained per project; you must delete older ones to add new ones.

06

Technical details

CORS: NoHTTPS: YesSignup: ?Open source: No
Auth type
unknown
Pricing
unknown
Protocols
REST
SDKs
python, javascript, csharp, java, go
Response time
53 ms
Last health check
6/26/2026, 6:23:46 AM
07

Tags

08

More from Microsoft