Classification
Classification, three ways
Single-label for clean categories, multi-label when media carries several attributes, hierarchical when your taxonomy has structure — Animal → Mammal → Dog.
Pick the project type that fits your taxonomy
Single-label
One image, one class, fast keyboard tagging. The quickest way through a sorting task.
Multi-label
Tag every attribute that applies — an image can be "outdoor", "night" and "rainy" at once.
Hierarchical
Organize labels as a tree and tag at the level you need.
Zero-shot help
Classify without any training. CLIP is built in; switch the zero-shot engine to SigLIP 2 on the Models page for stronger, multilingual matching. Downloaded once, then local.
Sort a backlog in one pass
Semantic search ranks the whole dataset against a description, and a batch run can apply the label to every match — left as proposals you confirm, never written behind your back.
The three Classification variants accept images, native video and Camera media. Tag a whole media item, or apply labels to a start-to-end frame range in video, then export. Single-label image datasets can export as YOLO folder-per-class. Multi-label, hierarchical, anomaly and video classification use Datumaro when annotations and schema must survive, or a plain ZIP for media only.
Common questions
Can one project mix classification types?
No — choose single-label, multi-label or hierarchical when creating the project, and the tagging workflow is built around that choice.
Can a model pre-sort my images?
Local options include CLIP or SigLIP 2 suggestions, semantic-search batch pre-labeling and an imported ONNX classifier (experimental). You can also classify the current image through the optional online ChatGPT mode, then use the standard Accept / Reject controls.
Does tagging work offline?
Yes. Native builds work fully offline; in the web app, annotation runs locally once the app and the models you use are cached.
Open AnnotateIt in your browser
Create your first local project in the browser. Optional ChatGPT or Claude annotation needs your own AI connection.
Questions before you start?Contact support →