Image Annotation API Reference¶

API documentation for the Image Annotation service.

Overview¶

The Image Annotation API provides endpoints for annotating images using Vision-Language Models (VLMs).

REST API Endpoints¶

POST /api/annotate¶

Annotate an image using a VLM.

Request:

{
  "image_path": "path/to/image.png",
  "model": "gpt-4-vision",
  "prompt": "Describe this image for HED annotation"
}

Response:

{
  "description": "A person riding a bicycle...",
  "objects": ["person", "bicycle", "street"],
  "hed_annotation": "Sensory-event, Visual-presentation, ..."
}

GET /api/annotations/{image_id}¶

Retrieve stored annotations for an image.

GET /api/models¶

List available VLM models.

Services¶

VLM Service¶

The VLM service handles communication with vision-language models:

• Ollama: Local models via Ollama
• OpenAI: GPT-4 Vision
• Anthropic: Claude Vision

Annotation Storage¶

Annotations are stored in JSON format in the annotations/ directory.

Configuration¶

Environment variables:

Python API¶

from image_annotation.services import VLMService

# Initialize service
service = VLMService(model="gpt-4-vision")

# Annotate image
result = await service.annotate("path/to/image.png")
print(result.description)
print(result.hed_annotation)