Reference Glossary
Computer vision tagging
The underlying image-classification and object-detection technology — not the DAM feature built on top of it — that recognises visual content in an image well enough to produce a label for it.
Why it matters in a DAM
The distinction matters because a DAM's auto-tagging feature is only as good as the computer vision model underneath it, and that model's training data sets a hard ceiling on what it can recognise. A model trained on general photo datasets can label 'shoe' correctly but has no way of knowing that shoe is this season's flagship product unless someone has fine-tuned it on the brand's own catalogue images.
Common mistake
Teams assume computer vision tagging understands brand-specific context — product names, campaign names, internal codenames — when out-of-the-box models only recognise generic visual categories; anything brand-specific still needs a custom-trained model or a human pass.
Computer vision tagging is the technical layer, not the product feature — the image-classification and object-detection models that decide what’s in a picture. ‘Auto-tagging’ in a DAM is the workflow wrapped around that layer: taking the model’s raw output and turning it into tags on an asset record.
The reason to separate the two is that the model’s training data is the actual ceiling on what any auto-tagging feature can ever recognise. A model trained on broad, general-purpose photo datasets learns categories like ‘chair’, ‘car’, or ‘shoe’ reliably, because those categories are well represented in its training data. It has no basis for recognising a specific product SKU, a campaign name, or an internal codename — those concepts simply never appeared in what it learned from.
Closing that gap requires either fine-tuning a model on the organisation’s own labelled images — a real engineering project, not a settings toggle — or accepting that brand-specific identification stays a human tagging task indefinitely. Buying a DAM with ‘AI tagging’ does not, by itself, change which of those two is true.
Frequently asked
How is computer vision tagging different from AI auto-tagging as a DAM feature?
Computer vision tagging is the underlying technical layer — the image-classification and object-detection models — while auto-tagging is the DAM workflow wrapped around it that turns the model's raw output into tags on an asset record.
What sets the ceiling on what an auto-tagging feature can ever recognise?
The computer vision model's training data — a model trained on general photo datasets can label 'shoe' correctly but has no way of knowing that shoe is this season's flagship product unless it's been fine-tuned on the brand's own catalogue images.
Can computer vision tagging recognise brand-specific product names or campaign codenames out of the box?
No — out-of-the-box models only recognise generic visual categories; those brand-specific concepts never appeared in what the model learned from, so anything brand-specific still needs custom training or a human pass.
What are the two ways to close the gap between generic and brand-specific recognition?
Fine-tuning a model on the organisation's own labelled images — a real engineering project, not a settings toggle — or accepting that brand-specific identification stays a human tagging task indefinitely.
Does buying a DAM advertised with 'AI tagging' solve brand-specific recognition automatically?
No — buying such a DAM does not by itself change which of the two paths, custom training or human tagging, is required for brand-specific identification.
What kinds of categories does a general-purpose computer vision model reliably recognise?
Well-represented categories like 'chair,' 'car,' or 'shoe,' because those are well represented in the broad, general-purpose photo datasets these models are typically trained on.