This document provides a technical overview of the quality assurance workflow used for evaluation datasets prepared by Origin Data Lab.
The evaluation package follows the same operational standards, validation pipeline, metadata preparation, privacy processing, and delivery structure defined for the corresponding production scope.
Capture Consistency
- Camera configuration follows the defined collection profile.
- Resolution, frame rate, capture orientation, and field of view are standardized within each batch.
- Recording requirements are communicated before field collection begins.
Segment Integrity
- Source recordings are segmented using defined, repeatable processing rules.
- Segment ordering and temporal continuity are preserved where required by the delivery specification.
- Every delivered clip maintains a reference to its original source recording.
Metadata Validation
- Metadata is prepared for every delivered segment.
- Required fields are checked for presence, format, and consistency before packaging.
- Available timestamps, GPS signals, motion signals, capture attributes, quality flags, and delivery fields are validated according to the agreed project scope.
- Scene and environment tags describe observable conditions rather than inferred intent.
Quality Control
Delivered segments are reviewed against applicable quality criteria, including:
- Camera obstruction
- Severe motion blur
- Severe exposure problems
- Broken or corrupted frames
- File readability and structural integrity
Source Traceability
- Delivered clips remain linked to their originating source recordings.
- Clip IDs, source IDs, and metadata references remain consistent across the delivery package.
- Collection-session context is preserved where required for technical review or sequence reconstruction.
Privacy Processing
- Faces and license plates are anonymized when required by the collection and delivery scope.
- Privacy processing is reviewed for consistency across relevant frames.
- Access to original source footage is controlled and governed by the applicable agreement.
Delivery Package Validation
Before delivery, the package is checked for applicable technical components, including:
- Dataset manifest
- Clip-to-metadata consistency
- Documentation completeness
- File and folder structure
- Integrity records or checksums when included in the delivery specification
Evaluation and Production Alignment
- Evaluation data is prepared as a structurally complete subset of the agreed production delivery.
- The same applicable review, metadata, privacy, and packaging rules are used for evaluation and production.
- Evaluation samples are not manually altered outside the defined preparation and quality-control workflow.
Exact metadata fields, privacy requirements, quality thresholds, annotation formats, and delivery documents are defined for each project and confirmed before production.