Annotator tools, filters and shortcuts
The annotator combines a canvas, a task-specific tool bar, the current tool’s options, an annotation list and the dataset navigator. Manual shapes are committed immediately; most smart-tool results stay pending until confirmed. Ordinary edits save when you submit or navigate, while Pending AI Review remains a separate review state.
The six working areas
- Top bar — back navigation, default label, Human/AI mode, Predictions, Help and account controls where available.
- Left toolbar — selection, drawing and smart tools, Undo/Redo, fit, visibility, canvas adjustments and keyboard reference.
- Options strip — settings for the active tool, item navigation and Submit.
- Canvas — the image or the current video frame and every visible annotation.
- Annotations panel — labels, attributes, ordering, visibility, locks, multi-selection and filters.
- Dataset panel and status bar — move between media and inspect file, zoom and canvas state.
Manual tools
| Tool | What it creates |
|---|---|
| Selection / Edit | Select, move, resize, rotate, lock, hide and reorder existing work. A rotation handle turns polygons and polylines, baking the angle into their points so every export format carries it |
| Bounding box | Axis-aligned rectangle |
| Rotated bounding box | Available in an existing rotated-detection project. The current Create project UI does not offer this subtype; this is not a tool offered in every detection project. |
| Circle | Ellipse/circle with adjustable radius |
| Polygon | Closed region placed point-by-point or drawn free-form |
| Polyline | Open path for line-like structures |
| Brush / Eraser | Pixel mask painting and cleanup |
| Keypoint | A pose instance based on the project skeleton |
Smart and assisted tools
The Text prompt dialog lets you select Labels to detect and edit their phrases. Choose a supported SAM + CLIP/SigLIP method or Grounding DINO for direct detection boxes where available. Review controls provide Reject, previous/next navigation and Accept; accepting a suggestion is a separate decision from running inference.
These propose shapes for you to accept, correct or reject rather than drawing them outright. Which ones appear depends on the project type and the platform’s capabilities.
| Tool | What it does |
|---|---|
| One-click mask (Segment Anything) | Click an object and get a mask from MobileSAM, SAM 2.1 or SAM 3 — an editable polygon by default, or the exact pixel mask when you choose Mask output |
| Click to refine (RITM) | Interactive click-to-segment: positive and negative clicks build the mask up |
| Text prompt (Grounding DINO) | Type a phrase and get boxes to review, in a detection project |
| Auto-pose (RTMPose) | Draw a box around a person and get a COCO-17 skeleton placed for you |
| Detection assistant | Box one example and a similarity routine proposes matching regions in the same image — no model, no download |
| Find by example (visual prompt) | Uses objects you already annotated as references and predicts similar ones, on the active segmentation engine |
| Box to mask (G) and Object coloring (W) | Assisted region tools that cut a shape out of a rough box or scribble |
| Prediction model | A curated Auto-annotation model or your own imported ONNX, set up per project, pre-labels the current image |
Smart-tool output is still editable
One-click segmentation and interactive segmentation compute a pixel mask. In an instance-segmentation project AnnotateIt traces that result into an editable polygon by default; choose Mask to retain the exact pixels for brush editing and RLE export. Detection projects convert the same result to a box. Text prompt, Auto-pose and prediction sources appear only where their project type and platform capability allow them.
Video projects add two more surfaces to the annotator: keyframed Video tracks that interpolate a shape between frames, and Ranges for labelling time intervals. The built-in Image editor, opened from the dataset menu, crops and adjusts the source image itself. Each has its own guide.
Annotation list, attributes and filters
Use the list to assign labels, fill per-object attributes, select several shapes, toggle visibility and locks, delete or change drawing order. A required attribute never blocks saving; a badge shows how many required values remain and Dataset Quality reports them across the dataset. Filters can combine label, attribute, size, score, source and shape conditions. Active filter chips mean the list and canvas may intentionally show less than the whole item.
Saving and review states
- Moving to another item saves ordinary annotation changes silently.
- Submit explicitly saves the current item; Undo and Redo operate on the current editing history.
- A smart-tool preview is accepted with the checkmark or Enter and discarded with X or Escape.
- Batch drafts marked Pending AI Review are excluded from exports, versions and ground-truth statistics until accepted.
Keyboard and touch
The keyboard icon in the toolbar is the canonical shortcut list for the running build. Common tool keys include V for Selection, B for Bounding box, C for Circle, P for Polygon, M for One-click segmentation and I for Click to refine; the dialog also lists navigation, save, undo and redo. Phone layouts hide the shortcut dialog, and touch support is tool-specific rather than a promise that every mouse gesture has a touch equivalent.