- Home
- Training data
- Images & Vision
Image training data for vision models.
Licensed images for vision-language models, document AI, and detection, delivered with captions, boxes, masks, and text transcription.
What we can supply
- Image and text pairs with dense captions
- Scene text and OCR imagery
- Charts, diagrams, and infographics
- Document and page images with layout
- Images for object detection and segmentation
- Product and catalog imagery
- Satellite and geospatial imagery
- De-identified medical imaging, annotated by clinicians
Labels and enrichment
- Dense captions
- Bounding boxes
- Instance masks
- Keypoints
- OCR text and reading order
- Attributes
- Chart data and Q&A pairs
- Expert findings
Formats and delivery
JPEG, PNG, or TIFF with JSON or COCO-style annotations. Delivered to AWS S3, Google Cloud Storage, Azure, or your own storage, with a datasheet and provenance records.
Common uses
- Vision-language pretraining
- Text-to-image models
- OCR and document AI
- Detection and segmentation
- Visual search
- Earth observation
Questions
Can you caption or label images we already have?
Yes. Send your images and we return them with captions, boxes, masks, or OCR, attached as JSON on every image.
Do images with people include releases?
We document the rights basis for every image, including releases where they are required, and can exclude images of people entirely.
How is medical imaging handled?
It is de-identified, annotated by clinicians, and licensed on terms that reflect the source institution's requirements.
Own data like this? License it through ModelCore