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

// After launch

Something is now in the way. Here is what it does.

Vegetation, new plant or stored material sits between the camera and the asset, and part of the view is quietly gone.

From the floor

A tree grew. Pallets were stacked in a new place. A temporary structure went up and stayed. The camera still returns a good image of whatever it can see, and nothing reports that a portion of the scene has been unavailable for a month.

Which failure is it

Covariate shift

The visible input changed. Worth separating from other covariate cases because the fix is often physical rather than statistical: no amount of retraining recovers an asset that is genuinely no longer in shot.

Four modes, and they need different responses: covariate, concept, prior, and a pipeline event wearing drift’s coat.

What moves first

Accuracy is the last thing to tell you.

Detections fall for the assets in one region of the frame while the rest of the frame is unaffected. Spatial rather than global, which is what distinguishes it from lighting or weather.

Measuring accuracy directly needs fresh labels, so teams lean on proxies between audits. By the time a quarterly audit shows a drop, the system has been wrong for a quarter. The cheap signals are confidence per class, detection rate per camera against its own history, and how often a reviewer overrides the model.

What to do

Check the frame before you retrain. If the asset is occluded, that is a maintenance ticket, not a labelling job. Retraining around a blocked view teaches the model to stop expecting the thing.

What it costs to ignore

The most misdiagnosed condition here. Teams retrain repeatedly against a view problem and conclude the model cannot do the task.

Where this shows up

Energy & utilities →Agriculture →Oil & gas →

What it gets first

The jobs where this one is the usual cause.

Site Safety & PPE ComplianceAutomated compliance checks for field crews in hazardous zones.Energy & utilitiesYield EstimationAutomated fruit/plant counting for accurate harvest forecasting.AgricultureStore Safety & ValidationDetect spills, blocked exits, and unauthorized access.RetailWarehouse Spatial MappingDynamic updating of facility maps for AMR fleets.RoboticsDynamic Environment TrackingMonitoring changes in physical spaces over time.Robotics

The others

A camera movedThe season turnedThe spec changedA new site came onlineAll ten →

We watch for this on your feeds, and retrain when it lands.

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