About this service
Torq Technology delivers custom egocentric data collection captured from a first-person point of view in real-world environments. Programs are designed for computer vision, robotics, wearable AI, augmented reality, human activity understanding, hand-object interaction modelling and embodied AI applications. Each engagement is scoped around the intended model task, target environments, participant cohorts, devices, camera placement, recording duration, action taxonomy and annotation requirements. Collection can cover daily living activities, workplace procedures, assembly and maintenance tasks, navigation, retail interactions, mobility scenarios and buyer-defined workflows. Capture configurations can include head-mounted or body-worn cameras, synchronized microphones, inertial sensors and contextual metadata. Projects may include scripted and naturalistic sessions, multi-view reference cameras, object inventories, task prompts, calibration sequences and controlled environmental variation across locations, lighting conditions and participant demographics. Optional annotation supports temporal action segments, object tracks, hand-object interactions, gaze or attention proxies where supported by the device, trajectories, scene labels, task outcomes and event metadata. Dataset structure, sampling strategy, annotation ontology and acceptance thresholds are agreed before collection begins.
Collection methods
Data types
Industries
Geographic coverage
Languages
Delivery formats
Typical deliverables
Delivery can include calibrated first-person video and image sequences; synchronized audio and IMU or device sensor streams where included; session, device, participant and environment metadata; task prompts and completion outcomes; frame-level or temporal annotations; object and hand-object interaction labels; bounding boxes, tracks, trajectories and action segments where scoped; train, validation and test splits; dataset manifests in JSON, CSV or Parquet; annotation guidelines; capture protocol; consent-status records; file checksums; and a final collection and QA report. Folder structure, naming conventions, sampling rates, codecs, resolution, annotation schema and delivery packaging can be aligned with the buyer model-training and evaluation pipeline.
Quality assurance
Quality assurance combines automated checks and trained human review. Controls can include device and camera calibration, capture-protocol compliance, video resolution and frame-rate checks, focus and exposure checks, audio intelligibility, sensor synchronization and drift checks, missing or corrupted file detection, duplicate-session detection, metadata completeness, task-distribution and cohort-quota validation, and file-to-manifest reconciliation. Annotation QA can include reviewer training, gold tasks, double annotation, inter-annotator agreement, boundary and label consistency checks, targeted adjudication and statistically defined sampling. Failed sessions or annotations can be rejected, recollected or corrected against buyer-approved acceptance criteria before delivery.
Compliance and privacy
Collections are designed around documented participant consent, purpose limitation and data minimization. Project controls can include participant pseudonymization, separation of identity records from dataset identifiers, restricted access, encrypted transfer, agreed retention and deletion schedules, and buyer-defined geographic restrictions. Where required, review can include face, license-plate, screen, document and other personal-information redaction. Final consent language, usage rights, privacy controls and security requirements are agreed for each project and jurisdiction before capture begins.