Skip to content

Use cases

Any task where a person could point at the thing in an image and draw around it is a Model Bench use case. Most run on the Standard edition; acceleration and measurement need Pro. Typical marks applications the product supports that are not yet a named customer reference.

The example screenshots show how each task is labelled in Model Bench. Each class gets its own colour. The images are drawn illustrations made for this manual, not customer images, and they show labels, not model results.

Use case What the model finds Annotation Edition Status
Surface defects on metal, plastic, painted parts Scratches, dents, cracks, chips, paint runs Box or polygon Standard Typical
Weld and joint inspection Missing, porous or misplaced welds; adhesive bead gaps Box or polygon Standard Typical
Assembly completeness Missing screws, clips, connectors, gaskets, labels Box Standard Typical
Electronics and PCB Missing or misplaced components, visible solder bridges Box Standard; Pro for line speed Typical
Castings and moulded parts Flash, short shots, surface porosity, burrs Box or polygon Standard Typical
Glass, bottles and closures Cap missing or not seated, chipped rim, foreign object Box Standard Typical
Pharma and medical packaging Empty or double blister cavities, broken tablets Box Standard Typical
Print and label quality Label missing, skewed, torn or wrong variant Rotated box or polygon Standard Typical
Textiles, wood and sheet goods Knots, holes, stains, tears Box or polygon Standard Typical
Food and produce grading Bruises, blemishes, foreign material, grade classes Box or polygon Standard Typical

Each defect type becomes a class. The model learns it from labelled examples, and InSiSo Visual Detector publishes the result to the Data Layer for the PLC to accept or reject the part.

Project dashboard of an inspection project: image sets in folders, with the labelling status under each thumbnail

Examples

Surface defects. Scratches and an edge chip are traced as polygons, dents as boxes. Classes: scratch, dent, chip.

Metal plate with three scratches and an edge chip outlined as polygons and two dents boxed

Assembly completeness. Every screw position gets a box, labelled present or missing, so the model learns both. Classes: screw, screw_missing, connector.

Housing with four screws boxed green, two empty screw holes boxed red, and the connector boxed blue

Electronics and PCB. Components and empty pads are boxed. Classes: ic, smd_part, part_missing, capacitor.

Circuit board with the IC, capacitors and SMD parts boxed, and two empty pads boxed red as part_missing

Bottles and closures. One box per cap position. Classes: cap_ok, cap_missing, cap_not_seated.

Five bottles: three caps boxed as cap_ok, one bottle without a cap and one tilted cap marked as defects

Pharma blister packs. One box per cavity. Classes: tablet, cavity_empty, tablet_broken.

Blister pack of 18 cavities: tablets boxed green, two empty cavities red, one broken tablet orange

Print and label quality. A label is traced with its four corners, so a skewed label is drawn at its real angle. A carton without a label is boxed as a whole. Classes: label_ok, label_skewed, carton_no_label.

Four cartons: two labels marked OK, one skewed label outlined at its angle, and one carton without a label boxed red

Food and produce grading. Each fruit gets a box for its grade, and the bruise itself is traced as a polygon. Classes: apple_ok, apple_bruised, bruise.

Six apples boxed by grade, with the bruises on two of them traced as polygons

Use case What the model finds Annotation Edition Status
Parcel detection and orientation on a conveyor Each parcel, with its four face corners Rotated box + package corners Standard (CPU); Pro on Hailo Measured on hardware
Pallet and carton recognition Cartons and pallets; missing or misplaced cartons in a stack Box Standard Typical
Package dimensioning (2D) Length and breadth of a box face in mm, with a squareness check Rotated box or 4-corner polygon Pro Typical
Package dimensioning (3D) Length, breadth and height from a Helios2 ToF point cloud Outline on the intensity image Pro Typical
Shipping label presence Label present and on the right face Box Standard Typical
Counting on a conveyor Objects per frame, per class Box Standard Typical

Detection, 2D or 3D dimensioning, barcode and photo combine into a parcel profile. See Parcel dimensioning and profiling.

A rotated box gives the position and orientation of each part on the belt. With Pro, the same box is measured in millimetres from the camera scale, and the result goes to CSV or Excel.

A rotated box on a bolt, measured as length by breadth in millimetres

Example

Parcel orientation. Each parcel is outlined as a rotated box, at whatever angle it lies on the belt. The four corners give its position and orientation, and with Pro its length and breadth. Class: parcel.

Five parcels lying at different angles on a conveyor, each outlined with a rotated box

Industry Use case Edition
Automotive Part presence, connector and clip seating, label checks Standard
Battery manufacturing Visible cell, tab and busbar defects; part presence Standard
Agriculture Leaf, fruit and produce sorting; foreign-material detection Standard; Pro on Hailo
Robot guidance Position and orientation of parts for pick-and-place, from rotated boxes and corners Standard; Pro for speed
Machine builders (OEMs) Vision delivered inside a ctrlX machine; the end customer retrains from real production frames through the ctrlX capture loop Pro recommended
Area monitoring Person or forklift in a zone, as an advisory signal only Standard