IMAGEOBJECTS
Everyday Object Photo Sets
Object images across viewpoints and states, with persistent instance relationships where the task depends on them.
View datasetIMAGEPRODUCTS
Product and Packaging Image Sets
Product recognition often depends on relationships across images: the same item from several views, different package faces, changing backgrounds or visible text on the packaging. Product and Packaging Image Sets keep those relationships explicit instead of treating each image as an unrelated example.
Where packaging text matters, OCR-derived text and region information can be connected to the same product identity. This gives a model access to both visual appearance and the textual information printed on the object without separating them into disconnected datasets.
Technical structure
- Persistent product identifiers across multiple images or views
- View, package-side, background or condition metadata where relevant
- OCR-derived text and text-region geometry when useful to the task
- Optional logo, barcode or attribute annotations
Typical applications
Product recognition, instance retrieval, variant discrimination, visual search, packaging analysis and text extraction.
Delivery note
Product identity, view structure and annotation depth are defined around the visual distinctions the model needs to make.
Delivery
The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.