Home/Data/Image

Image data

Real-world images from MENA environments — retail, street, document, and mapping contexts — for computer vision, OCR, and visual recognition.

A model trained on generic stock imagery underperforms on the storefronts, road signage, packaging, and handwriting styles actually found in this region. We collect images in-context — a Riyadh grocery aisle, Cairo street signage, Arabic invoice layouts — rather than substituting Western equivalents, and we handle consent and PII (faces, plates, storefront names) as part of the collection protocol, not as an afterthought.

What a reviewer checks

Annotation accuracy against the task spec (bounding boxes, OCR transcription, category labels), image usability (blur, occlusion, lighting), and consent/redaction compliance before any image leaves our pipeline.

Worked example. A retail-AI buyer needed 50,000 shelf images from grocery stores across Riyadh for product-recognition training. Contributors photographed real shelves under a strict consent and redaction protocol; a review tier checked labeling accuracy and flagged any image with identifiable bystanders for redaction or rejection.