Scene-text image collection format for multilingual OCR, visual translation, accessibility AI, document-in-the-wild recognition, and text-region detection.
Real-World Text and Signage Photos are built for teams evaluating how vision and OCR systems understand text captured in real environments. The format focuses on signage, menus, transit boards, retail labels, printed notices, product labels, public information displays, and other approved real-world text contexts.
The dataset is designed to support OCR and visual-language systems that need more than clean document scans.
Structured collection
Real-World Text and Signage Photos use defined scene-text capture tasks.
Metadata depth
The format supports alignment between image files, text-region boxes, transcription records, reading order, translation fields…
Delivery-ready package
Includes image quality, text visibility, transcription reviews, plus dataset manifest, CSV catalog, consent status metadata, QA…
Request sample access
Share your target languages, scene categories, transcription needs, annotation schema, and evaluation goals. We’ll reply with a technical scope for the collection.
Scene-text image files
Text-region IDs · Ground-truth transcription records
Language metadata · Script metadata · Scene category labels · Country or city metadata where scoped · Lighting condition metadata · Resolution metadata · Privacy review status
Optional bounding boxes · Optional reading order labels · Optional translation fields
Real-World Text and Signage Photos use defined scene-text capture tasks. Contributors capture approved public or non-sensitive text sources under rules for visibility, privacy boundaries, location context, and transcription requirements.
Image quality review · Text visibility review · Transcription review · Language/script review · Privacy review · Annotation consistency review · Delivery approval
Image files · Ground-truth transcription CSV · Text-region annotation files where scoped · Scene metadata CSV · Dataset manifest JSON · CSV file catalog · QA summary · Consent status metadata · Delivery notes
Product and packaging image sets with multi-angle capture, lighting metadata, package-side labels, category fields, resolution checks, QA status, and optional bounding-box or OCR annotations.
Everyday object image sets with object IDs, multi-view capture, material/category metadata, angle labels, resolution checks, QA fields, and optional bounding boxes for perception models.
Paired multimodal datasets linking video, audio, screen, text, or image-derived records through shared file IDs, cross-modal manifests, metadata schemas, and QA fields.
Fully scoped custom data collection with buyer-defined modality, schema, metadata fields, annotation layers, QA criteria, manifest structure, delivery package, and collection rules.
Ambient audio sets with scene labels, acoustic context metadata, speech screening, device fields, noise-level estimates, and QA status for acoustic scene and robustness evaluation.
Activity video sets captured across varied real-world environments, with task labels, environment metadata, camera position fields, lighting conditions, duration, and QA status for activity recognition and embodied AI evaluation.
