Data Licensing · Image data

120 million labelled property images

Real property photographs, each paired with a structured JSON record — room, element, material, condition and defects. Purpose-built training data for computer vision and AI.

Explore the labelled image data

Real records from the sample — each image paired with its exact JSON. Click any tile to see the full record, including every defect labelled.

Structured labels on every image

Labels applied at the moment of capture by trained inspectors, through closed picklists across a 42-field taxonomy.

Room types

Kitchen, Bathroom, Bedroom, Living room, Hallway, Stairs, Exterior…

Elements

Walls, floors, ceilings, doors, windows, fixtures, appliances…

Condition ratings

Brand new, As new, Good, Clean, Fair, Aged, Poor…

Materials & finishes

Painted plaster, laminate, tiles, fitted carpet, UPVC, granite…

Defects

Cracked plaster, chips, worn walkways, marks, discolouration, damaged sealant…

Property attributes

Archetype, property type, region, inspection type…

Why train on this data

Image + companion JSON

Every photograph ships with a self-contained JSON record — room, element, material, colour, finish, condition and every defect — applied at capture by trained inspectors through closed picklists.

Ground truth for vision models

Consistent structure and captioning make it usable for training and evaluation out of the box: condition detection, material and defect classification, element and fixture recognition, archetype models.

Privacy-reviewed

Every image is individually reviewed — GPS stripped, no addresses, house numbers, plates, people or readable documents; location expressed as region only.

Clean, licensable rights

First-party imagery captured through our own app and stored on our infrastructure, with clear ownership — defensible provenance for training-data buyers.

Request the image sample

423 labelled images, each with its full JSON record, for evaluation. Released after this short form so we know who we're sharing the data with — no anonymous downloads.

Looking for structured records for insurance or analytics instead? See the property-data product.

Get the sample dataset

Tell us a little about your use case.

Optional

Optional — regions, property types, fields or volumes

Image data FAQs

How big is the image dataset?

120 million labelled photographs of UK residential property, drawn from around 1.9 million inspection reports across 906,000 properties, and growing continuously. Every image carries a structured JSON record across a 42-field taxonomy.

How are the images labelled?

Labels are applied at the point of capture by trained inventory professionals using closed picklists inside the miProgram app — not free text, not crowd-sourced. Each image's JSON gives the room, element, material, colour, finish, condition rating and any defects.

What format is delivered?

JPEG images with a matching JSON file per image (the sample uses exactly this structure). Delivery is flexible — flat files, an API/feed, secure data sharing or a marketplace listing — with custom slices by region, property type or label set.

Is it GDPR-compliant?

Core inspections are of vacant properties containing only landlord-owned assets. Every image is individually reviewed, GPS metadata stripped, and images with any identifying detail excluded; location is region-level only. No tenant personal data is supplied.

Can I evaluate a sample first?

Yes — request access and you'll get a labelled sample of 423 images, each with its full JSON record, for evaluation and research.

Evaluate the data today

Request a labelled sample and we'll help you scope the right dataset for your models.

Request access

Prefer to talk? data-licensing@miprogram.co.uk