Textphotos — Text-Rich Images
Text across languages, scenes and surfaces — explore the images and inspect the words within.
A world of text, within reach
Rotate the sphere, narrow the language, scene or text amount, and open an image to inspect its words and regions.
Two layers, scoped separately
Image context and text-region annotations stay linked to the same image. Review depth and production coverage are scoped to your brief.
- Confirmed primary language, supported secondary languages and scene labels
- Text-region polygons and saved OCR transcripts
- Word and character counts, annotated-area ratio and image dimensions
- Per-image annotation and cross-check status
- Transcript and polygon adjudication to an agreed acceptance standard
- Language, scene and text-density balancing for a production batch
- Custom quality gates and downstream-task evaluation
Machine-readable contract: textphotos.schema.json
Inside this sample
Counts reflect the browsable sample. Individual cross-checks use independent readings and AI visual review; the status stays attached to each image.
| Metric | Value | Context | Source |
|---|---|---|---|
| Browsable sample | 2,000 | Actual images included in the sphere and its filters. | Published sample manifest |
| Language scope | 6 | English, Russian, Chinese, Arabic, Hindi and Spanish. Images with uncertain primary-language evidence are excluded from this sample. | Sample language policy |
| Individually cross-checked | 2,000 | Independent OCR readings plus an AI visual audit for every sample. Polygon boundaries are not pixel-perfect glyph masks. | Published sample manifest |
What we tell you before you ask
- Every sample has independent OCR readings and an AI visual audit. This review does not produce pixel-perfect glyph masks.
- Annotated area is the union of text-region polygons.
- Delivery volume, language balance and acceptance criteria are scoped per project.
Selected examples from the Textphotos image collection. Production sourcing and rights requirements are specified with the delivery brief.
License class: Rights scoped per delivery. Production provenance and rights requirements are agreed with the delivery scope. Methodology →
- Image files with per-image JSON annotations
- JSON / Parquet metadata index
- WebDataset shards
- S3 / GCS delivery
Loader snippets for HuggingFace datasets, WebDataset and
PyTorch are in the docs.
Three steps
- 01 Explore the public samples
Right on this page — schema, QA numbers and live viewer where cleared.
- 02 Request an eval batch
A private sample matched to your domain, under NDA, usually within 48 hours.
- 03 License the batch or scope the build
Take the current batch as it stands, or scope volume, annotation depth, license constraints and refresh cadence to your training recipe.