Classical fits
Clear geometry, stable lighting, measurable features, tight takt with deterministic tools.
Knowledge · Architecture
classical Computer Vision vs AI inspection, classical first, AI when needed.
Rules, blobs and gauging win on explainability and low data hunger. AI earns its place with texture, cosmetic defects or natural variation that fixed thresholds cannot handle.
Choice
When which approach
Sedeco does not position AI as a default replacement for classical Computer Vision, but as a targeted upgrade when feasibility proves it.
Consider AI when
Variable texture, cosmetic defects, natural products, or patterns rules cannot capture.
Hybrid
Often: classical for presence/gauging + AI for cosmetic zones, do not force everything into one model.
Feasibility
Data and samples before training promises
Without representative good/bad samples and process context, AI is marketing, not a project.
-
01
Try classical
Optics and lighting first; many ‘AI problems’ are contrast problems.
-
02
Document failure
Where rules fail structurally (false rejects/accepts), AI becomes relevant.
-
03
Dataset needs
Variation across batches, lighting and product types, not only lab photos.
-
04
Integration
Include PLC/MES result, logging and change management.
Unsure about AI?
Send good/bad samples, we say whether classical holds or AI testing is warranted.