Skip to content
LexDataLexData
PlatformIndustriesCustomers
DocsThe Field GuideBlogWhy models drift
AboutCareersSecurityContact
Log inStart now

// MANUFACTURING

Parts Count Verification

Parts Count Verification. What it takes to run it.

Each part counted once as it crosses the belt, exact for a kit, a band with a floor for a case, on the camera already there.

Start with your footageBook a demo →All of manufacturing →

// 01 · The question

“How many of this part just passed, and does the count match the kit or the case?”

// 02 · Why it is hard

Touching or clumped parts are the labeling decision the count rests on: a box drawn tight around each one teaches the model to separate a clump, a box drawn around the clump teaches it that a clump is one part. A part half hidden behind another needs a written rule, boxed when its head or its edge shows, applied the same way by every labeler, or the model learns to flip a coin on every partial part.

// 03 · What the labels have to be

Bounding box + tracking

Decided before the first frame

A box alone gives a count of sightings, not of parts: the same bolt sits in frame for many samples as it crosses the belt. A track gives each part one identity so it is counted once, at the line it crosses. A kit or a case counted in a single still frame needs no track at all, since every part in it is visible at once; what that count needs instead is a box tight around every part, including the ones touching their neighbours.

This is the decision that is expensive to reverse. The geometry has to match what the answer contains, and finding out it does not means labelling the set a second time.

// 03a · What it finds

Every part in frame

a box on each bolt, screw or bottle, from the camera already over the belt or the bench

Each part counted once

a track that keeps one identity on a part from the edge of the frame to the line it crosses, so a part seen in many samples is counted in exactly one of them

The kit or the case against its own rule

the boxes in one frame totalled against the kit's exact spec or the case's own band with a floor, no track needed because every part is already visible

A box the model doubts

a clump of touching parts or one half in shadow, held for a person before the count ships

// 03b · What it checks

  1. 01

    The count against the kit's exact spec, an equality, or the case's own band with a floor, so an exact-count station is never asked to pass a part it only half saw and a band station is never stopped for the one bottle the camera could not quite see.

  2. 02

    The tracked count against a second source already on the line, a feeder wheel or a scale, with a severity and a cooldown, so a mismatch beyond the agreed margin pages the operator rather than every bag.

  3. 03

    The rule approved before it goes live, written as a sentence and tested against footage from this belt or bench before anyone trusts it on a live kit.

  4. 04

    A doubted frame, a clump, a part in shadow, held for a person rather than scored either way, the correction recorded against the frame and the frame kept for the next version.

// 03c · What the line provides

  • The camera already over the belt or the bench; no new camera, no scale, no gate.
  • Frames from this belt or bench, at its own speed and under its own light, not a part photographed clean on a bench under a lamp.
  • The kit spec or the case pack size in the plant's own words, and a way for the rule to read it the day it changes.
  • Someone who can look at a doubted frame, a clump or a part in shadow, the way the operator already decides on the bench.

// 04 · How we run it

Three parts of one loop, on this job.

LexAnnotate

Defect Annotation

Pixel-level defect annotation for surface scratches, cracks, dents, weld faults, and assembly misalignments. Multi-sensor alignment for multi-camera inspection systems.

LexInsight

Line Performance

Ask which shift or which line is producing the most of one defect class, and get an answer with the frames behind it, graded against the plant's own written standard.

LexAlert

Quality Alerts

Monitor defect detection accuracy across changing production conditions. Detect when lighting shifts, new materials, or equipment aging degrades model performance. Auto-retrain to maintain 99%+ accuracy.

How the loop fits together →

// 05 · What breaks it after launch

Concept drift

The spec changed

→Concept drift →

A new kit bill of materials or a new case pack size changes the right answer without changing a single pixel the camera sees. The model's count is still correct; the number the rule compares it against is what moved, and has to change the day the new spec reaches the bench, not whenever someone remembers to update it.

It is not the only one that can get this use case, it is the one that usually gets it first. All ten conditions.

// 06 · In manufacturing

Defect Detection at Scale

Full write-up ↓Every case study →

The Challenge

Vision models performed well in pilots but lost accuracy as factory conditions shifted.

Our Execution

Built pixel-level annotated datasets and multi-sensor alignment for reliable defect detection.

The Result

Maintained 99%+ defect detection accuracy across changing production environments. Reduced unplanned downtime.

99%+

Accuracy maintained in production

// 06a · Questions we get asked

Does this replace the scale or the tally sheet?

No. It gives a count from the frame, with the frame attached when the two disagree. The plant's own scale or tally sheet stays the reference the camera is checked against, the way a packer keeps weighing bags after the camera goes in.

Can it tell a kit's exact count from a case's band?

Yes, but only because the two are set up differently before any labeling. A kit that ships to a customer who counts is an equality; a case the packer cannot physically overfill is a band with a floor. Which one a station is gets decided first, not learned by the model on its own.

What about parts that are touching or stacked?

A box drawn tight around each one, even inside a clump, is the label that teaches the model to separate them. The clump is also the frame a person checks hardest, since a clump labeled as one part teaches the model to undercount every time.

What happens when the kit spec changes?

The pixels stay the same and the right answer moves. The count itself is still correct; the number the rule compares it against has to change the same day the new spec reaches the bench, which means the spec lives where the rule can read it rather than in a binder by the door.

Read next

Counting objects on a conveyor belt without counting any of them twice →Counting the screws in a kit, an exact count for a kit and a band for a case →Assembly verification: presence and orientation at a fixed station, the same completeness check at a different scale →What a new kit spec or case pack size does to a count that was never wrong →
Start with your footageBook a demo →

// 07 · Also in manufacturing

Surface Defect Detection→Weld Quality Inspection→Assembly Verification & Error Proofing→Package & Label Inspection→Date & Lot Code Inspection→Predictive Maintenance→Multi-Sensor Monitoring→Tablet Defect Inspection→Syringe Barrel Defect Detection→Foreign Material Detection→Glass Defect Detection→Semiconductor Wafer Inspection→Strip Surface Defect Inspection→Ceramic Defect Detection→Wood Defect Detection→Body-in-White Inspection→Contact Lens Defect Inspection→Coffee Bean Defect Inspection→All forty-one→
Surface Defect DetectionWeld Quality InspectionAssembly Verification & Error ProofingPackage & Label InspectionDate & Lot Code InspectionPredictive MaintenanceMulti-Sensor MonitoringTablet Defect InspectionSyringe Barrel Defect DetectionForeign Material DetectionGlass Defect DetectionSemiconductor Wafer InspectionStrip Surface Defect InspectionCeramic Defect DetectionWood Defect DetectionBody-in-White InspectionContact Lens Defect InspectionCoffee Bean Defect InspectionAll forty-one →

Send us a week of this footage. We’ll show you what comes back.

Start with your footageBook a demo →
LexData
LexData

Product

  • Platform
  • Industries
  • Use cases

Resources

  • Docs
  • The Field Guide
  • Blog
  • Why models drift

Industries

  • Energy & utilities
  • Oil & gas
  • Agriculture
  • Manufacturing
  • Insurance
  • Retail
  • Robotics

Company

  • About
  • Customers
  • Careers
  • Contact

Trust

  • Security
  • Privacy
  • Terms

Stay updated

What we learn running vision models in production.

See everything.
Miss nothing.

Stay updated

What we learn running vision models in production.

Terms of use & Privacy policy

© 2026 LexData Labs · All rights reserved