--- title: "Video Data Collection and Evaluation | YPAI" url: https://ypai.ai/video-data/ description: "Managed video collection for models that need movement and context: in-cabin and dashcam, human motion, scenario capture. A provenance record per delivery." source: "src/copy/routes (route /video-data/)" --- # A frame tells you where. A sequence tells you why. > Managed video collection for models that need movement and context: in-cabin and dashcam, human motion, scenario capture. A provenance record per delivery. We collect video against the task your model has to solve, with specified coverage, quality control and documented rights. A model trained on frames learns where things are. The change between frames is where the decision lives, and it is what this collection is specified to capture. YPAI designs and runs managed video collection programmes: scenario, cohort, environment and capture mode specified before the first take, reviewed against that specification, and delivered with the rights and provenance record for every recording. The counter shows how many frames this step contains. Show dropped frames plays the step at 10 fps, the loss the platform checks for on every recording. ## One requirement becomes one accountable video operation Model task, scenario, cohort, environment and acceptance criteria are written before capture. The same specification is what the footage is reviewed against, and what the delivery record documents. ### Specification - The model task sets scenario, cohort, environments, capture mode and acceptance criteria, before anyone records. - **Coverage**: Cohort, environment, device and condition, agreed before capture - **Acceptance**: Criteria written before work starts ### Capture - Footage recorded under the specified conditions - Identity-verified contributors record under the specified conditions; rate, resolution and sync are fixed in the specification. - **Frame rate**: Selected for motion speed and the downstream task - **Synchronisation**: Multi-view, timecode or hardware sync where the task requires it ### Quality gates - Each recording accepted, or returned with a reason - Automated checks, then human review against the specification; a rejected item returns with its reason. - **Review**: Native reviewer on sampled work - **Verdict**: Accepted, or returned with a reason ### Delivery - The footage delivered with its record - Accepted footage leaves with its metadata, rights and provenance record, acceptance report, checksums and manifest. - **Record**: Provenance per recording, consent per contributor and purpose - **Package**: Checksums and a versioned delivery manifest Capture specifications, calibration and reshoot terms are defined in the statement of work. ## Start from the movement your model is missing Eight programmes YPAI runs, and three adjacent kinds of work with their own pages. Each programme names the movement it captures; open it for the problem, the delivery and the annotation it hands off to. ### [Collection at scale](https://ypai.ai/video-data-collection/) - The footage your model needs does not exist, or misses the conditions it will meet. - Scenario-based, quota-managed capture across countries, environments and devices, remote or moderated, with reshoot workflows and delivery manifests. A coverage gap becomes recording instructions. - Coverage is specified per cell before volume is counted. - The same scene under every condition it will meet. - [Object tracking and temporal events](https://ypai.ai/annotation/video-annotation-services/) ### In-cabin and dashcam - Driver, occupant and road-facing models need European cabins and roads under real conditions. - Driver monitoring, occupant monitoring, cabin activity and road-facing video, captured across the drivers, vehicles, lighting and weather the deployment will meet. Voice in the cabin is its own programme. - Coverage of cabin and road that matches the deployment, with geographic scope defined against the intended use. - A glance away from the road is captured as a sequence of frames. - [Keypoints and sensor fusion](https://ypai.ai/annotation/video-annotation-services/) ### Human motion, gesture and pose - The model needs movement, interaction or several views of the same action. - Full-body and upper-body motion, gestures, locomotion and action sequences, with multi-view synchronisation where the task requires it. Pose and keypoint labels are annotation work and have their own page. - Synchronised human-action video collected for the temporal task. - Each action is recorded from its start to its end. - [Keypoints and temporal segmentation](https://ypai.ai/annotation/video-annotation-services/) ### Video integrity and synthetic media - Detection models go stale as generation methods move, and benchmarks hide their failure modes. - Manipulated-media datasets, synthetic-media evaluation, authenticity review and human-perception studies, on provenance-aware footage. Handling and purpose limits are agreed per engagement. - Real footage with its record, so the authentic half of the benchmark is documented. - Manipulation often shows between frames before it shows in a single frame. - [Evaluation and failure analysis](https://ypai.ai/annotation/video-annotation-services/) ### On-camera speech and conversational video - Visible speakers, audio-video sync and lip-sync-sensitive data, recorded as one programme. - Solo, paired and multi-speaker recording with isolated speaker tracks, transcription and diarisation, demographic and language quotas. Speaker recruitment and the consent record live on the speech hub. - Isolated per-speaker tracks and audio-video sync fixed in the specification. - Lip movement and sound are checked for synchronisation. - [Speech-video alignment](https://ypai.ai/annotation/audio-speech-annotation-services/) ### Licensed video datasets - The deadline does not allow custom collection, or a rights review failed on the set you have. - Existing inventory identified, sourced through partners, sampled and technically reviewed, rights checked, licensed, packaged and delivered. License what exists and collect only what is missing. - Chain of title and permitted use are checked before a sample is shared. - Every licensed clip is delivered with its provenance. - [Annotation documented per set](https://ypai.ai/annotation/video-annotation-services/) ### Indoor activity and consumer behaviour - Home, workplace and in-store models need everyday activity under the conditions people actually live in. - Human activity, consumer behaviour and work-task video in indoor and outdoor environments, device and environment controlled, remote or moderated, with consent per contributor and purpose. - Footage from opt-in households and workplaces. - A fall, a hand-over, a purchase: each is a change across frames. - [Temporal events and tracking](https://ypai.ai/annotation/video-annotation-services/) ### Face, identity and talking-head video - Facial movement, expression, gaze and head pose across identities, sessions and lighting. - Per-project consent and rights scope (model release, permitted use, reuse) shape how this delivery is set up, with file-to-participant traceability throughout. - Repeated-session identity and speech-linked facial movement, traceable to the participant file. - Expression, gaze and head pose are recorded as movement across frames. - [Keypoints and evaluation](https://ypai.ai/annotation/video-annotation-services/) - Adjacent work with its own page ### [Speech and audio collection](https://ypai.ai/speech-data/) - The task is the voice, with no camera in the loop. - Speech and audio programmes have their own hub: recording, transcription, language coverage and the consent record for speakers. ### [Egocentric and physical AI video](https://ypai.ai/physical-ai-data/) - First-person video, hand-object interaction and task demonstrations for robot learning. - Workplace tasks, tool use, grasp, place and assembly, with synchronised device and sensor data. The physical AI hub owns the episode structure. ### [Video annotation and evaluation](https://ypai.ai/annotation/video-annotation-services/) - Tracks, temporal events, action labels and keypoints on footage you already hold. - Object tracking, temporal events, action labels, keypoints, interactions and speech-video alignment, with generated-video review and failure analysis. - Gap-fill against footage you already own Working with a scenario not listed here? Describe the deployment and the condition it runs under, and YPAI designs the capture protocol. ## Four axes decide whether a sequence survives deployment You name the range your deployment has to hold. The programme is built to span it, and every recording lands with its coverage in its record. - The mount, the sensor and the rate are fixed in the specification, because a model trained on one capture mode degrades on another. - Frame rate and resolution selected for the task - The street, the cabin, the room. Light, weather and glass are specified, not hoped for. - Quota-managed cohorts, identity-verified, with the rights for every visible person recorded. - Age range, gender identity, skin tone - The change between frames is the edge case. It is captured on purpose, at the rate the decision needs. - The same scene under every condition it will meet - Fast motion, frame rate selected for it Coverage is agreed before capture begins and delivered as part of the record. Where a required range cannot be covered, YPAI says so during scoping rather than after delivery. The reading in each tab is taken from the frame itself: sharpness as the variance of the Laplacian, and the share of pixels under 25 IRE. The platform takes the same readings on every recording it accepts. ## Every recording carries its own evidence Provenance-aware datasets for teams building or defending against synthetic media. The record travels with the file. ### Source evidence per recording - Capture metadata and a checksum for every delivered asset, with lineage from source to derivative. ### Manipulated and synthetic media - Manipulated-media datasets, synthetic-media evaluation and authenticity review carry their own handling and purpose limits, agreed per engagement. ### Human-perception studies - Reviewer panels judge authenticity and artefacts against the same specification the footage was collected to. ## We built the system behind the data. YPAI runs its own data collection and assurance platform, built for GDPR obligations, across speech, video, image, text, human feedback, agents, robotics and other multimodal programmes. Customer specifications and agreements become executable controls across sourcing, collection, technical validation, consent, human review, rights, dataset assembly and delivery. Every accepted asset is linked to the evidence for why it passed. The chain, for one video asset - **Specification**: Your specification and the signed agreements become the project the platform runs against - **Source**: Contributor and device qualified before the first take - **Collection**: Conditions controlled at capture, in the browser, on device or on site - **Quality**: Real-time edge control, then central automated validation, then qualified human review - **Agreements**: Consent and agreement state bound to the asset, statement by statement - **Rights**: Permitted use enforced by the platform, per asset - **Dataset**: Constraint-based assembly against the contract - **Delivery**: Signed delivery or controlled model access - **Verification**: Data Assurance Record per asset, independently verifiable What the platform measures on every video recording - Dropped and duplicated frames, bitrate and codec - Focus, exposure, motion artefacts and continuity - Participant visibility, scene compliance and liveness - Camera provenance, facial and biometric rights - **Identity**: Asset, modality and schema version - **Contract**: The requirements it was collected against, and their coverage - **Origin**: Source, device, application, environment and time of creation - **Quality**: Rule-based, statistical, model-based and cross-modal checks, then a human verdict - **Agreements**: Applicable statements, their hashes and signatures - **Privacy**: Processing purpose, legal state and data categories; consent per visible person Built by YPAI. Built for GDPR. Documented asset by asset. ## The record your legal and security review will ask for The operation produces the governance artefacts as it runs. Depth and format follow your risk profile and are agreed during scoping. The review asks · the record answers Where a person is in the frame, the consent record travels with the recording. - **Legal basis**: What is the legal basis for each recording?. Consent, legitimate interest or contractual necessity, documented per recording - **Consent**: Can you show consent, per contributor, and what happens on withdrawal?. A traceable rights and consent record linked to each contributor, recording and delivered asset where participant consent is the applicable basis, with withdrawal status; withdrawal runs through a workflow with an audit trail - **Rights**: What may the footage be used for, and by whom?. Permitted use, release and licence scope defined per engagement; reuse by explicit written grant - **Provenance**: Where did each recording come from?. Provenance per recording, capture metadata and checksum, lineage from source to derivative - **Residency**: Where does the footage reside, and under whose jurisdiction?. EEA infrastructure by default, Norwegian jurisdiction; EEA-only sourcing on request - **Retention**: What happens to the footage when the engagement ends?. Erasure within 30 days of contract end; retention and closeout defined in the engagement - **Agreements**: What governs the engagement on paper?. DPA signed, SCCs available for cross-border transfer Each recording is documented with its legal basis: consent, legitimate interest or contractual necessity. Where participant consent is the applicable basis, a traceable rights and consent record links each contributor to their recordings and delivered assets, with withdrawal status. Withdrawal runs through a workflow with an audit trail. Permitted use, release and licence scope are defined per engagement. Any reuse needs an explicit written grant. Every recording carries its provenance, capture metadata and a checksum, with lineage from the source to each derivative. Footage resides on EEA infrastructure by default, under Norwegian jurisdiction. EEA-only sourcing is available on request. Footage is erased within 30 days of contract end. Retention and closeout are defined in the engagement. A signed DPA governs the engagement, and SCCs are available for cross-border transfer. ## Bring the footage problem. Leave with a capture plan and an acceptance proposal. Technical assessment inside one EU business day. ### Describe the system - Model task, target scenario, cohort or geography, scale and timeline. ### Feasibility read - Capture plan, coverage matrix, rights scope and acceptance proposal. ### Pilot - A scoped delivery against the agreed specification, reviewed before production. ### Production - Full programme with ongoing review, versioned delivery and support. We will define the video programme required for your deployment context and constraints. If YPAI is not the right fit, we will say so directly. specified · reviewed · delivered with its record - [Read how pilots work](https://ypai.ai/pilots/)