Packaging manufacturing · A small factory

Small factory: computer vision assists the inspector

Labels and appearance are checked by eye; lighting varies, defects are rare and attention fades by the end of a shift.

Situation

A packaging manufacturer checks labels and appearance by hand. Lighting changes, defects are rare, and repetitive viewing is tiring by the end of a shift. The goal is repeatable inspection and faster detection of suspicious items.

What changes

A stable camera and lighting, a set of labelled examples, then a model that looks for agreed visible defects. The interface highlights the suspicious area.

Where people stay in control

The model first runs alongside full human inspection. Manual viewing can be reduced only after quality has been checked and only for a suitable, non-critical process; thresholds and failure procedures are agreed in advance.

What changes for the team

The inspector sees where to look, keeps the decision, and disputed examples are reviewed together.

How the example is calculated

Hours of visual inspection (possible mode after validation)
Before
33.3 h
After
15 h

Preliminary estimate · batch of 6,000 items

6,000 × 20 seconds = 33.3 h; 20% of items × 30 seconds = 10 h + 5 h of spot checks and upkeep = 15 h

Assumptions

6,000 items; 20 seconds per item now; after validation 20% of items go to manual review at 30 seconds — an assumption, not an achieved model performance; 5 hours of checks and upkeep. The parallel pilot saves no time.

What to check in a pilot

Missed defects by type, false alarms, changes in lighting and materials. This scenario may not suit critical products.

Explore a similar situation

The map starts from the same kind of problem. None of this story’s volumes or results are assumed for your business.