Release notes

yaAnnotator now speaks seven more languages. Plus: a crash on macOS 27 fixed and optimizations to make large image libraries much faster to browse.

Headline features

• Seven new languages. The interface is fully localized into Simplified Chinese, Japanese, German, French, Korean, Spanish, and Italian.

Other improvements

• Browsing large libraries is substantially faster. Every row in the image list was rebuilding its context menu from the whole library during layout, so returning from the editor to a big project could stall for seconds, stutter, and drop clicks. Single-column view gains the most.
• Always a way back. New View ▸ Show Sidebar (⌃⌘S) and View ▸ Back to Images (⌘[). A collapsed sidebar, or a toolbar hidden while editing, could previously leave no visible way out of an image.
• The editor canvas has a minimum width, the same one the image browser uses.
• Validated against macOS 27 Golden Gate with more macOS 27 specific features to come. But we'll cross that bridge when we come to it ...

Bug fixes

• Fixed a crash when opening an image on macOS 27, caused by toolbar layout as the toolbar switched between the browser and the editor.
• The window no longer resizes itself when you leave the editor.
• The app now gazes enviously at the iPhone Duo. Sorry about that. Maybe if you keep it busy and distracted...

Help is still English only, and the translations have not met every native speaker yet. If something in the interface reads wrong in your language, please file a bug report. I did my best.

Two updates' worth in one release (1.1 never shipped; I just kept going): Smart Select pixel masks, on-canvas proposal review, a stronger built-in detector, batch detection, video-to-frames import, large dataset performance improvements, and near-duplicate cleanup.

Headline features

• Smart Select masks. Smart Select now creates pixel-accurate masks, not just outlines. Refine a mask by adding or removing regions with a click, or re-run segmentation from a box when you want a different result. (Mask creation is Pro; viewing, converting, importing, and exporting masks is free.)
• Review proposals on the canvas. Accept or reject model proposals without leaving the image: click to accept one, drag a marquee to select several, or accept them all at once. An inline confidence slider shows proposals appear and drop live, so you can find the right cutoff by eye.
• A stronger built-in detector. Detect Multiple now runs RF-DETR Nano instead of YOLOv3: better proposals across the same 80 common classes, still fully on-device.
• Batch detection. Select any number of images in the list or grid and run any Detect command over all of them, with proposals accepted automatically.
• Detect Subjects. Automatically outline the foreground instances of any class as polygons, so you start from a real segmentation instead of a blank canvas.
• Import video as frames. Point yaAnnotator at a video and pull frames into a folder you choose, ready to label. (Pro)
• Performance improvements. Large dataset performance improvements across the board, especially when you import and use masks.
• Near-duplicate cleanup. Find groups of near-identical images and remove the extras in one action, keeping one per group.

Other improvements

• Annotation layers. Show or hide boxes, polygons, and masks independently, and convert between them.
• Masks import and export. Round-trip masks through COCO RLE and PNG mask sidecars, with no data loss. Importing segmentations as pixel masks is a choice on the import sheet; a plain polygon still imports as a polygon.
• First-run welcome window. Jump straight to recent projects, a new or existing project, or Help when you launch.
• Built-in export presets. Start from ready-made export recipes instead of a blank configuration.
• Customize the editor toolbar. Rearrange it the Mac way, with every toolbar action mirrored in the menus.
• Paste Annotations from Previous Image (⌥⌘V). Copy the previous image's annotations onto the current one in a single shortcut, then nudge them into place with the arrow keys, with snapping while you drag.
• Mark images that have no objects. Flag background images explicitly; they're counted correctly in filters, badges, and exports.
• Fill-opacity control and a drag-zoom slider in the toolbar, for clearer overlays and quicker zooming.
• Open the Label Manager from anywhere, not only from the image list.
• Optional automatic report recompute. Let the Analyze Annotations report refresh itself as you work, or keep it manual.
• Safer imports and exports. A warning when a folder is syncing, a free-space check before export, and an advisory on import help you avoid surprises with your files.
• Faster with big projects. Mask imports rasterize in parallel, removing many images at once is dramatically faster, and dense projects scroll more smoothly.
• Quicker to read at a glance. Thumbnails break down their annotation counts by kind, and imports name each phase so you know what's happening.
• More keyboard, better VoiceOver. Page Up / Page Down step through images everywhere, ⌃⌘L hides labels on the canvas, a loaded custom detector runs with ⇧⌘D, and the grid, list rows, and canvas are more accessible.
• Mask polish. Clearer selection, press Return to finish a mask, and layer toggles in the inspector.
• Sturdier with imperfect files. Corrupt or unreadable images are caught and reported clearly wherever they turn up.
• Smart Select mask creation and refinement, and video import, are part of Pro, with a gentle prompt when you reach for them.

Bug fixes, etc.

• Exports and augmentation passes now add to a folder instead of overwriting what's already there.
• You can cancel a running video extraction.
• Unable to cancel a run on sentence that I was writing while trying to create a joke to insert into these release notes.
• Large or malformed masks no longer cause trouble on import.
• Fixed a COCO import regression: segmentations default to polygons again, with masks as the opt-in.
• Fixed label assignment when several annotations were selected at once.
• Fixed the Simplify action on very large polygons.
• Fixed a case where an export or backup could be written to the wrong location.
• Fixed occasional stale values in the inspector after edits.
• Got you tickets to Taylor's wedding. They are at will-call. I hope you see this message in time.

The first public release. Everything you need to build object-detection datasets entirely on your Mac.

Headline features

• Precise annotation. Draw boxes, polygons, and rotated boxes on a fast native canvas, with full undo/redo and review flags.
• Model-assisted labeling. A built-in YOLOv3 detector and Apple Vision give you a first pass instantly. Bring your own Core ML or RF-DETR models, or click to segment with SAM2 Smart Select.
• Analyze Annotations report. See class balance, rare-class flags, box geometry, and duplicate or cross-split leakage detection, with clickable charts that filter your images.
• Annotate Stuff! Built to support my own annotation needs. This first release is the bare minimum of what I needed to generate a quality dataset.

Other features

• Import and export the formats you use: COCO, YOLO, CreateML, CVAT, and Pascal VOC, as whole datasets or per split.
• Fully offline and private. No accounts, no cloud, no telemetry. Images are referenced in place, never copied or uploaded.
• Pro, as a one-time purchase: augmentation on export, isolate objects into a classification dataset, export presets, and unlimited projects.