Multimodal Paired Capture Sets
Linked audio, video, screen, image or text records with explicit relationships between modalities.
View datasetSome projects already have a clear data contract, internal schema or model objective that does not fit a standard collection format. Buyer-Scoped Custom Collection starts from that requirement rather than asking the buyer to adapt it to a predefined template.
The useful unit of observation, source media, inputs, targets, metadata and annotations can be defined around the existing workflow. Depending on the project, the delivery can also incorporate temporal structure, multimodal relationships, provenance fields or other technical elements needed to make the data usable inside the buyer's pipeline.
Existing internal schemas, combined-modality projects, evaluation-specific datasets, extensions to buyer-provided material and unusual record structures.
Nothing beyond the agreed specification is implied by this collection. The format is defined from the buyer requirement outward.
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.