IMAGE / VIDEO / LIDAR / MEDICAL

Pixel-accurate segmentation.Your taxonomy. Your evidence pack.

Semantic, instance, panoptic, 3D, and video segmentation for regulated CV teams. EEA-resident annotation, EU AI Act Article 10 evidence, kappa-gated production. SAM-assisted where it helps, human-reviewed where it counts.

  • EU AI Act Article 10
  • 40,000+ contributors
  • EEA-resident
Scope a Data Project See sample annotations
road vehicle vegetation building
SCHEMA Cityscapes 19-class

Schematic. Production runs against your raw imagery in your chosen taxonomy.

PROCUREMENT READINESS

Compliance posture for segmentation training data.

Article 10 enforcement begins 2 August 2026. YPAI ships every segmentation engagement with the artefacts a regulated buyer needs in their file.

Per-delivery artefact pack

Annotation guideline (versioned)
Gold set with documented injection rate
Per-class IoU report
Boundary F1 / HD95 for medical work
Records of processing (Article 30)
Signed DPA + sub-processor list

EU AI Act Article 10

Data governance for high-risk CV systems. Annotation provenance, bias-examination notes, class taxonomy documentation.

GDPR Articles 7, 28, 30

Per-contributor consent. Processor agreement. Records of processing. 30-day erasure SLA. Sub-processor change notifications.

EEA-resident processing

Norwegian company. EEA contributor network. EEA infrastructure. Outside US CLOUD Act reach. ISO 21448 SOTIF data-governance aligned.

Request a Procurement Readiness Brief →

We map the evidence package to your data, risk class, and deployment environment.

PRIMITIVE TYPES

Five segmentation primitives.

Which one you procure depends on what your model has to know. The double-spend failure mode is ordering semantic and discovering you needed instance.

Semantic

per-pixel class

One class per pixel; no instance separation. Stuff classes (road, sky, vegetation) and thing classes (vehicle, pedestrian) treated identically.

USED FOR Road scene parsing, organ regions, land-use classification

Instance

per-pixel class + instance ID

Each object gets a unique ID alongside its class. Counting becomes possible: 12 pedestrians, 4 cells, 3 defects per image.

USED FOR Counting, tracking, per-object measurement

Panoptic

semantic + instance unified

Stuff classes get per-pixel labels. Thing classes get per-pixel labels plus instance IDs. One unified mask, two semantic regimes.

USED FOR Full scene understanding, autonomous driving stack

3D / point cloud

per-point class

Each LiDAR point or voxel labelled with a class. Used in automotive (street scenes) and volumetric medical (3D CT, MRI organs).

USED FOR LiDAR perception, 3D medical imaging

Video / temporal

frame-wise + tracked

Per-frame masks with consistent instance IDs across time. Object tracking, surgical phase, action recognition, motion analytics.

USED FOR Video annotation, surgical workflow, sports analytics

WHERE SEGMENTATION IS PROCURED

Five verticals where pixel-class matters.

Different industries, different ontologies, same delivery contract: EEA-resident annotation, kappa-gated production, Article 30 records on delivery.

Automotive ADAS

Cityscapes-aligned taxonomy. Pedestrian, cyclist, vehicle, lane, traffic-sign. ISO 21448 SOTIF data-governance aligned.

CityscapesALKSSOTIF

Medical imaging

Brain MRI tumour (BraTS protocol), kidney CT (KiTS), histopathology nuclei. DICOM SEG output, Dice + HD95 reporting.

BraTSDICOMHD95

Satellite + aerial

Land-use classification, road extraction, building footprints, deforestation. SpaceNet schemas, GeoJSON delivery.

SpaceNetGeoJSONfwIoU

Agriculture

Crop-vs-weed, fruit yield, canopy gap, disease zones. Multispectral capable. Drone + ground imagery.

MultispectralDroneYield

Industrial + robotics

Surface defect (scratch / dent / contamination), pick-and-place region masks, LiDAR fusion for warehouse / drone / construction. Quality gate masks.

DefectLiDARRobotics
Discuss your domain

Or describe the domain and YPAI will scope ontology, kappa target, and delivery format.

HOW WE LABEL

Every project clears the same six gates.

Calibration before production. Documented mask-IoU thresholds. 100% human QA. The artefacts you need for your Article 10 file.

01

Schema design

Class taxonomy locked with your team. Aligned to public benchmark (Cityscapes, COCO-Stuff, BraTS, custom) or built fresh. Versioned.

Deliverable: Versioned schema spec

02

Annotation guideline

Edge cases enumerated per class. Boundary policy documented (snap-to-edge tolerance, occlusion handling, ambiguous class adjudication).

Deliverable: Annotation guideline document

03

Calibration round

Pilot batch on shared subset. Mask IoU between annotators computed. Disagreement patterns drive guideline refinement before scale.

Deliverable: Calibration mask-IoU report

04

IAA gate

Production starts only when calibration clears Landis-Koch substantial agreement (mask IoU 0.75+, Dice 0.85+ medical) on the schema.

Deliverable: Gate-pass attestation

05

Production labelling

SAM-assisted where domain-shift permits (natural images). Human-only on medical, aerial, 3D point cloud, and microscopy where SAM under-performs.

Deliverable: Labelled batches

06

QA + adjudication + delivery

Every batch reviewed. Gold-set items injected at 5-10% rate. Per-annotator Dice gate. Delivery ships with per-class IoU, boundary F1 / HD95 for medical, Article 30 records.

Deliverable: Final delivery pack with metrics report

Six gates. One trail of evidence. Every delivery.

Start the scoping process →

WHAT WE DELIVER

Hover any scene to see the annotation.

Schematic previews. Production work is delivered against your raw imagery in your chosen taxonomy.

HOVER reveal annotation

Automotive ADAS

19 classes (Cityscapes taxonomy)

HOVER reveal annotation

Medical imaging (brain MRI)

BraTS protocol: tumor core, edema, necrosis

HOVER reveal annotation

Satellite + aerial

Land use: building, road, vegetation, water

HOVER reveal annotation

Agriculture

Crop / weed / soil / canopy gap

HOVER reveal annotation

Industrial vision

Surface defects: scratch, dent, contamination

HOVER reveal annotation

Panoptic (full scene)

Stuff + things + instance IDs

PUBLIC BENCHMARK COVERAGE

Taxonomy-aligned with the corpora your model trained on.

Public datasets are useful as taxonomies, baselines, and audit references. Most are research-licensed; production work runs against your own data under your engagement DPA. License status is part of the procurement record.

Cityscapes

Research only

Automotive street scenes

Samples
5,000 fine + 20,000 coarse
Schema
PNG mask, 19 eval classes
License
Non-commercial research license
Taxonomy reference

Mapillary Vistas

Research only

Automotive, global cities

Samples
25,000 images
Schema
124 / 152 classes
License
Research / academic only
Taxonomy reference

BDD100K

Research only

US driving

Samples
10,000 seg images
Schema
COCO-style
License
Berkeley DeepDrive license
Taxonomy reference

A2D2

Commercial

Automotive (Audi)

Samples
41,277 frames
Schema
38 classes, PNG + JSON
License
CC BY-ND 4.0
Evaluation reference

COCO-Stuff

Commercial

Stuff + things general

Samples
164,000 images
Schema
171 classes, JSON
License
CC BY 4.0
Evaluation reference

ADE20K

Commercial

Scene parsing

Samples
25,000+ images
Schema
150 classes, PNG
License
BSD-3-Clause variant
Evaluation reference

BraTS

Research only

Brain tumour MRI

Samples
2,000+ MRI volumes
Schema
NIfTI, 4 classes
License
Challenge license
Domain reference

KiTS

Research only

Kidney tumour CT

Samples
500+ CT volumes
Schema
NIfTI
License
Challenge license
Domain reference

LIDC-IDRI

Commercial

Lung nodules

Samples
1,018 chest CT
Schema
DICOM + XML
License
NIH public
Domain reference

SA-1B

Commercial

General segmentation

Samples
11M images, 1.1B masks
Schema
COCO mask
License
Apache 2.0
Domain reference

SemanticKITTI

Research only

LiDAR point cloud

Samples
43,552 scans
Schema
25 classes per point
License
CC BY-NC-SA 4.0
Taxonomy reference

DAVIS

Research only

Video object segmentation

Samples
150 video sequences
Schema
Per-frame mask
License
Research license
Evaluation reference

License status reflects publicly stated terms at the dataset source. Verify per engagement before any commercial training use.

WHAT YOU RECEIVE

Every delivery ships with the artefact pack your Article 10 file needs.

The records a regulated buyer expects with every segmentation engagement. No upgrade tier, no separate request.

Annotation guideline and boundary policy.

Versioned guideline with edge cases enumerated, examples per class, glossary of domain terms, and boundary tolerance rules. Updated as adjudication surfaces new patterns; every version preserved for audit.

Gold set with injection rate.

Held-out gold set with 5 to 10 percent undisclosed injection during production. Per-annotator Dice gate. Calibration before queue entry, blind re-runs on failure. Quarterly refresh.

Per-class IoU and boundary metrics.

Per-class IoU report on every delivery. Boundary F1 and HD95 for medical, fwIoU for satellite. Confusion matrix and top failure modes by frequency. Panoptic Quality for panoptic work.

Mask IoU inter-annotator report.

Pairwise mask IoU between annotators on the calibration subset. Landis-Koch substantial threshold target. Per-class breakdown and per-annotator over-time trend.

Records of processing, DPA, and sub-processor list.

Article 30 records of processing, signed Article 28 DPA, lawful-basis documentation, 30-day erasure SLA, and full sub-processor list with Article 28(2) change notifications. All included with every engagement.

START A PROJECT

Brief us. We reply within one business day.

Short brief now, deeper scoping in the reply.

Capability lanes (NER, RLHF, etc.), languages, volume, regulatory context.

QUESTIONS BUYERS ACTUALLY ASK

Frequently asked questions

EEA-resident. Norwegian company, EEA contributor network, EEA infrastructure. 30-day GDPR Article 17 erasure SLA. Outside US CLOUD Act reach. Member-state region negotiable per engagement.

Yes. Per-engagement self-hosted CVAT or Label Studio project, dedicated access lists, no shared annotator pools across engagements. Single-tenant infrastructure with documented separation.

GDPR Article 28 processor obligations, EU AI Act Article 10 data governance, EEA-residency, DPA-by-default, and single-tenant isolation. Security artefacts package available on request.

No. Customer-owned work product. No reuse, no resale, no model-training rights retained. Sub-processor list disclosed at engagement start with Article 28(2) change notifications.

Opt-in only for medical pathology, defect, surveillance footage, and content with potential distress. Screening, rotation off sensitive batches, no-penalty opt-out. Specifics documented in the project brief.

GDPR-Native EU AI Act Article 10 EEA Operations Consent Evidence

Brief us on your segmentation project.

One business day reply. NDA on request. DPA included.

All 5 primitives Article 30 records per delivery
Or connect on LinkedIn →

Your information is never shared. We respond with the next scoping step.