Vision Extraction Reference

Reference documentation for the Vision Extraction system, including all status indicators, metrics, configuration variables, and output formats.

System Status Indicators

The Vision Extraction page displays the current configuration:

Status

Enabled or Disabled

Shows whether vision extraction is currently active

Vision Model

Model name (e.g., qwen2-vl)

The Ollama model being used for image analysis

Ollama Server

Server URL

The Ollama instance processing images

Image Index Statistics

Metric	Description
Properties	Total number of properties in the database
With Images	Properties that have at least one photo
Total Images	Total number of property photos across all listings
Descriptions	Number of images with AI-generated descriptions
Visual Embeddings	Number of images with visual similarity embeddings
Coverage	Percentage of images that have been indexed (Descriptions / Total Images × 100)

Coverage Calculation

Coverage % = (Images with Descriptions / Total Images) × 100

Target: 100% coverage for complete searchability

Job Status States

Jobs progress through these states:

Status	Badge Color	Description
pending	Gray	Job queued but not started
running	Blue	Currently processing images
completed	Green	Successfully finished
failed	Red	Encountered critical error
cancelled	Yellow	Stopped by user action

Job Options Reference

Option	Type	Default	Description
Limit	Integer	`null` (all)	Maximum number of properties to process
Skip visual embeddings	Boolean	`false`	Generate descriptions only, skip image similarity embeddings

Feature Distribution Metrics

Top Features

Ranked list of extracted feature tags with counts:

Field	Type	Description
Feature	String	Tag name (e.g., “hardwood floors”)
Type	String	Category (material, view, style, appliance)
Count	Integer	Number of images with this feature

Room Types

Classification breakdown showing photo distribution:

Kitchen
Living Room
Bedroom
Bathroom
Exterior
Dining Room
Office
Garage
Other

Extraction History Columns

Column	Data Type	Description
Status	String	Job completion state (completed, failed, cancelled)
Started	DateTime	Timestamp when job began (ISO 8601)
Duration	Time	Total processing time (HH:MM:SS)
Properties	Integer	Number of properties processed
Images	String	Format: “successful / failed” (e.g., “1,234 / 5”)
Tags	Integer	Total feature tags created across all images
Triggered By	String	User or system identifier that started the job

Configuration Variables

Environment variables controlling vision extraction:

Variable	Type	Default	Description
`OLLAMA_BASE_URL`	URL	-	URL of your Ollama server (required)
`VISION_MODEL`	String	`qwen2-vl`	Model name to use for image analysis
`VISION_CONCURRENCY`	Integer	`4`	Number of parallel image processing tasks

Example Configuration

OLLAMA_BASE_URL=https://ollama.supported.systems
VISION_MODEL=qwen2-vl
VISION_CONCURRENCY=8

Output Data Formats

Image Description

{
  "image_id": "property-123-photo-1",
  "description": "Modern kitchen featuring granite countertops, stainless steel appliances, and pendant lighting over a large island",
  "room_type": "kitchen",
  "features": [
    "granite countertops",
    "stainless steel appliances",
    "pendant lighting",
    "kitchen island"
  ],
  "generated_at": "2024-12-28T10:30:00Z"
}

Visual Embedding

{
  "image_id": "property-123-photo-1",
  "embedding": [0.023, -0.451, 0.782, ...],  // 768-dimensional vector
  "embedding_model": "qwen2-vl",
  "generated_at": "2024-12-28T10:30:00Z"
}

Rate Limits

Default processing constraints:

Limit Type	Default Value	Adjustable Via
Concurrent Requests	4	`VISION_CONCURRENCY`
Max Batch Size	Unlimited	Job “Limit” option
Timeout per Image	30 seconds	Server configuration

Extract Property Features — Step-by-step extraction guide
Vision Concepts — How the system works