Custom Vision is an Azure service for training purpose-built image classifiers with labeled examples. It suits narrow visual jobs, such as recognizing distinctive book-cover styles that generic image analysis may not capture. section 7.3 section 7.3.2
A practical Custom Vision workflow looks like this:
- Provision a Custom Vision Training account with the Azure CLI, choosing a resource group, region, and service tier. section 7.3.2
- Create a Vision Studio project using Classification and the Compact domain for a faster, lighter model. section 7.3.2
- Upload representative images and apply labels such as CircuitBoardCover, RetroFuturismCover, or MinimalistCover. section 7.3.2
- Train the model, then use the prediction endpoint and key to classify new images through REST. section 7.3.2
Book-cover tagging
A neon-lined hacker manifesto can be sent to the trained endpoint and tagged as FuturisticTechCover. Those predictions support genre browsing and visual-style recommendations, replacing repetitive manual catalog tagging with a repeatable pipeline. section 7.3.2 section 7.6
C:/> az cognitiveservices account create --name banningBooksCustomVision --resource-group BanningRG --kind CustomVision.Training --sku S0 --location westus --yes
The Azure CLI listing shows the resource-creation command that starts a Custom Vision workflow. section 7.3.2
C:/> az rest --method post
--url "https://banningBooksCustomVision.cognitiveservices.azure.com/
customvision/v3.0/Prediction/<your-project-id>/classify/
iterations/Iteration1/url"
--headers "Prediction-Key:<your-prediction-key>"
"Content-Type=application/json"
--body "{\"url\":\"https://myblob.blob.core.windows.net/
covers/newcover.jpg\"}"
The REST listing shows how a stored image URL is submitted to a trained prediction iteration. section 7.3.2
