Skip to content
Book a scoping call →

Work

Judge the labels,
not the pitch.

A production project first: 23,551 railway frames labelled and reviewed by our team. Then seven task demos in the classes, conventions and export formats we use on client projects.

Case study · Rail infrastructure

Railway overhead-line detection

Cabin-camera video from five Indian routes, labelled so a pantograph’s height can be measured straight from a detection. Every frame was proposed by a model, then corrected and reviewed by hand.

Labelled railway frame, Mumbai: 14 boxes in one frame: masts, arms and the pantograph against direct sun.

Mumbai14 boxes in one frame: masts, arms and the pantograph against direct sun.

  1. Proposal: A pretrained detector proposes boxes inside CVAT.

    01Proposal

    A pretrained detector proposes boxes inside CVAT.

  2. Correction: A proposal that runs too tall is caught by the labeller.

    02Correction

    A proposal that runs too tall is caught by the labeller.

  3. Review: The fixed box after a second-pass reviewer signs it off.

    03Review

    The fixed box after a second-pass reviewer signs it off.

  4. Hardest class: Contact points are ~1% of frame width, placed at high zoom.

    04Hardest class

    Contact points are ~1% of frame width, placed at high zoom.

Real project frames from our public showcase (CC BY-NC-SA 4.0). The full 8 GB dataset stays private · View on Kaggle ↗

More task types · synthetic demos, drag to compare

BUILDING · 72 VEGETATION WATER
Labelled
Raw
‹ ›

Demonstration scenes are synthetic and generated for this site. No client data is shown. Drag the divider to compare.

Every delivery

Data is half the
deliverable.

  • Labelled dataIn your schema and file structure, validated against it before delivery.
  • Batch QA reportPer-class metrics against the gold set, rework rate and agreement scores.
  • Issue logAmbiguous items flagged with a question, never silently guessed.
  • Guideline changelogEvery rule change versioned, so you can trace any label to the rule behind it.

Batch report

PRJ-ADAS · Batch 14

Accepted

Frames

2,400

Objects

33,033

Mean IoU

0.931

Rework

2.1%

ClassObjectsIoURecallRework
car 18,412 0.947 0.991 1.2%
truck 2,106 0.931 0.986 2.1%
pedestrian 6,930 0.912 0.978 3.4%
cyclist 1,204 0.904 0.972 3.9%
traffic_sign 4,381 0.928 0.989 1.8%
Gold sample 5% · blindGuideline v1.6IAA κ 0.84
Example report · illustrative figures

Next step

Scope your project with
an annotation lead.

A 30-minute call with the person who would run your project. We review a sample of your data, recommend an approach and send a fixed-price proposal within two business days.

Book a scoping call → or email services@nextribes.com