5.2.1. Image Classification

Image classification assigns labels to entire images. Custom Vision lets you train classifiers for your specific categories when pre-built services don't cover your domain.

Classification types:

• Multiclass: One label per image (mutually exclusive categories)
  • Example: "Cat" OR "Dog" OR "Bird"
• Multilabel: Multiple labels per image (non-exclusive tags)
  • Example: "Outdoor" AND "Sunny" AND "Beach"

Training requirements:

• Minimum 5 images per tag (15+ recommended)
• Balanced classes produce better models
• Include negative examples (what it's NOT)

Workflow: Upload images → Tag images → Train → Evaluate → Publish → Call prediction endpoint