CNEN
Contact us

DATA ANNOTATION

Data annotation services

Give every label a basis.Make every batch reviewable.

Turn images, video, speech, text, 3D and multimodal data into training labels. Align classes and boundaries in a pilot, then deliver labels with versioned guidelines and review records.

Discuss annotation needsView annotation samples
Vision · Language · Spatial data · Associations
Vision · Language · Spatial data · Associations
Task definitionClasses · attributes · boundaries
Quality basisRules · review · traceability
DeliverablesLabels · reports · versions

ANNOTATION ACROSS DATA TYPES

BEFORE YOU STARTIf raw material is not ready, check files, indexes and pairings first. If classes are still unclear, use a pilot to establish the rules.

ANNOTATION SAMPLES

Choose the task. See inputs, labels and outputs.

Explore typical tasks, required inputs and output formats by data type.

IMAGEANNOTATION ACROSS DATA TYPES
Positions, classes and attributes support image understanding.
Positions, classes and attributes support image understanding.
  • Boundaries
  • Classes
  • Spatial relations
  • Attributes

IMAGE

Define information from objects to pixels.

Typical tasks
Object detection, semantic and instance segmentation, keypoints, OCR and attributes.
Inputs to prepare
Images, class dictionaries, boundary rules, transcription guidelines and edge cases.
Delivery formats
Boxes, regions, keypoints or text records; assess COCO, YOLO or custom JSON for the task.
Discuss this annotation type

DELIVERABLES

Annotations with the records needed to review them.

Geometry, semantics, timing and relationships require different fields. Confirm the structure in the pilot, then reconcile labels, versions and quality records for each batch.

Structured annotations

Geometry, transcripts, entity spans, event intervals or pair indexes organized for the agreed task.

Rules & class definitions

Class dictionaries, attribute rules, boundary examples and change history.

Quality & issue records

Review records, issue categories, rework status and batch acceptance notes.

Batch & version manifest

File indexes, count reconciliation, delivery versions and key relationships.

PROCESS & QUALITY

Align the standard. Then scale the delivery.

Each stage has a defined output. Sampling ratios, schedules and acceptance thresholds are agreed after project evaluation.

  1. Scope & rules
  2. Pilot & guideline review
  3. Production & review
  4. Acceptance & delivery

STAGE 01 / 04

Scope & rules

Define the model task, class boundaries, inputs and acceptance approach.

THE OUTPUTtask brief & draft rules
  1. Scope & rules
  2. Pilot & guideline review
  3. Production & review
  4. Acceptance & delivery

STAGE 02 / 04

Pilot & guideline review

Test the rules on a small batch and align difficult or disputed cases.

THE OUTPUTagreed samples & rule version
  1. Scope & rules
  2. Pilot & guideline review
  3. Production & review
  4. Acceptance & delivery

STAGE 03 / 04

Production & review

Annotate in batches, review in stages, record issues and complete rework.

THE OUTPUTlabels & review records
  1. Scope & rules
  2. Pilot & guideline review
  3. Production & review
  4. Acceptance & delivery

STAGE 04 / 04

Acceptance & delivery

Check structure, coverage and agreed quality items before packaging.

THE OUTPUTdataset & acceptance record
Every step leaves a record.

APPLICATION CASE · ROAD VISION

One image batch. Clear rules and review steps.

ANN-002 · Road vision

A road-vision team piloted 2D bounding boxes on 360 images, focusing on small objects, occlusion and truncation. Rule revisions and rechecks established consistent boundaries and issue handling.

Need
Define class and box rules for vehicles, pedestrians, cyclists and traffic signs.
Approach
Align boundaries in a small pilot, record issues and rechecks, and keep unresolved items visible.
Deliverables
Pilot annotations, rule versions, an issue log and review records.
View annotation caseExplore delivery & quality
Road-image object annotationROAD VISION / ANNOTATION
360Images
4Object classes
1Pilot batch

FAQ

Before the first conversation

01Can we run a pilot without a label schema?

Yes. Use representative samples to define classes, boundaries and exclusions, then test the guidelines in a pilot. Volume, timing and cost depend on task complexity.

02Can you work with our format or platform?

Share the format specification, platform requirements and de-identified samples. We validate field mappings and exports before confirming compatibility and access.

03How are timing and cost estimated?

Estimates depend on volume, task complexity, review requirements and delivery cadence. Changes to scope or guidelines require a revised estimate.

START WITH YOUR DATA

Define the first batch, together.

Share your data type, model objective and estimated volume to frame an initial pilot.

Discuss annotation needsExplore delivery & quality