Labelme

AIGPL-3.0

Image annotation tool for polygons, boxes and points.

Latest v7.11.0 · by Kentaro WadaWritten in PythonWebsitewkentaro/labelmeRSS

Release activity

Release activity — 14 releases across 12 days since Aug 21, 2026. Each cell is one day; darker means more releases that day. Nothing is recorded before Aug 21, 2026. Older weeks are hidden at this screen width.
AugSepOctNov
SundayNo releases on Aug 23, 2026No releases on Aug 30, 2026No releases on Sep 6, 2026No releases on Sep 13, 2026No releases on Sep 20, 2026No releases on Sep 27, 2026No releases on Oct 4, 2026No releases on Oct 11, 2026No releases on Oct 18, 20261 release on Oct 25, 2026No releases on Nov 1, 2026No releases on Nov 8, 2026No releases on Nov 15, 2026
MondayNo releases on Aug 24, 2026No releases on Aug 31, 2026No releases on Sep 7, 20261 release on Sep 14, 2026No releases on Sep 21, 2026No releases on Sep 28, 2026No releases on Oct 5, 2026No releases on Oct 12, 2026No releases on Oct 19, 2026No releases on Oct 26, 2026No releases on Nov 2, 2026No releases on Nov 9, 2026
TuesdayNo releases on Aug 25, 2026No releases on Sep 1, 2026No releases on Sep 8, 2026No releases on Sep 15, 2026No releases on Sep 22, 2026No releases on Sep 29, 2026No releases on Oct 6, 2026No releases on Oct 13, 20261 release on Oct 20, 2026No releases on Oct 27, 2026No releases on Nov 3, 2026No releases on Nov 10, 2026
WednesdayNo releases on Aug 26, 20261 release on Sep 2, 2026No releases on Sep 9, 2026No releases on Sep 16, 2026No releases on Sep 23, 2026No releases on Sep 30, 20261 release on Oct 7, 2026No releases on Oct 14, 2026No releases on Oct 21, 2026No releases on Oct 28, 2026No releases on Nov 4, 2026No releases on Nov 11, 2026
Thursday1 release on Aug 27, 20262 releases on Sep 3, 2026No releases on Sep 10, 2026No releases on Sep 17, 2026No releases on Sep 24, 2026No releases on Oct 1, 2026No releases on Oct 8, 2026No releases on Oct 15, 2026No releases on Oct 22, 2026No releases on Oct 29, 2026No releases on Nov 5, 2026No releases on Nov 12, 2026
Friday1 release on Aug 21, 2026No releases on Aug 28, 2026No releases on Sep 4, 2026No releases on Sep 11, 20262 releases on Sep 18, 2026No releases on Sep 25, 2026No releases on Oct 2, 2026No releases on Oct 9, 2026No releases on Oct 16, 2026No releases on Oct 23, 2026No releases on Oct 30, 2026No releases on Nov 6, 20261 release on Nov 13, 2026
SaturdayNo releases on Aug 22, 2026No releases on Aug 29, 2026No releases on Sep 5, 2026No releases on Sep 12, 2026No releases on Sep 19, 2026No releases on Sep 26, 20261 release on Oct 3, 2026No releases on Oct 10, 2026No releases on Oct 17, 2026No releases on Oct 24, 2026No releases on Oct 31, 2026No releases on Nov 7, 20261 release on Nov 14, 2026

14 releases since Aug 21, 2026, busiest day 2

Changelog

v7.11.0

Latest
Changed 4
  • SAM 3 asks to be downloaded again, as 1.8 GB on a GPU or 3.7 GB on the CPU, and asks again for its CPU files when AI falls back to the CPU
  • Files from earlier versions of SAM 3 stay on disk
  • AI no longer retries a request on the CPU when the GPU fails partway through; the next request loads the model again
  • The status bar no longer shows "Preparing image…"
Fixed 4
  • The model menu shows the download size for the AI device, so SAM 3 shows about 1.8 GB on a GPU instead of its CPU size
  • Deleting an AI model no longer freezes the window while AI runs, and other loaded models stay loaded
  • Quitting during AI no longer leaves the AI server running for up to 5 seconds
  • An interrupted model download no longer leaves partial files behind

From Labelme

Changed
  • SAM 3 asks to be downloaded again, as 1.8 GB on a GPU or 3.7 GB on the CPU, and asks again for its CPU files when AI falls back to the CPU. Its files from earlier versions stay on disk. (#3099)
  • AI no longer retries a request on the CPU when the GPU fails partway through; the next request loads the model again. The status bar no longer shows "Preparing image…". (#3100)
Fixed
  • The model menu shows the download size for the AI device: SAM 3 shows about 1.8 GB on a GPU instead of its CPU size. (#3104)
  • Deleting an AI model no longer freezes the window while AI runs, and other loaded models stay loaded. (#3105)
  • Quitting during AI no longer leaves the AI server running for up to 5 seconds, and an interrupted model download no longer leaves partial files behind. (#3106)
View originalPermalink
How v7.11.0 went

v7.10.0

Added 20
  • Allow drawing Polygon and LineStrip paths by dragging, simplifying strokes into fewer vertices while preserving click-to-add input
  • Add 3-point circle mode (circle_3p) to draw circles by clicking three points on the circumference
  • Open JPEG 2000 Images (.jp2, .jpx, .j2k, .j2c) through Pillow with OpenJPEG support, including folder scanning and drag-and-drop, while preserving decoded pixels and transparency in memory
  • labelme.log records the platform, the installed AI runtime versions, and how long each AI model takes to load, encode an image and generate annotations
  • A Compute device Setting runs AI on the CPU, and the status bar shows while AI is running and whether it runs on the CPU
  • AI-Find All replaces the AI Text Prompt, finding every object that matches a text prompt or a box around one example refined with more examples and counter-examples drawn while holding Shift, then added with Enter or discarded with Esc
Changed 7
  • Right-clicking a Shape opens a menu of Shape actions, and right-clicking empty canvas opens a menu of drawing and canvas actions
  • AI models run on the GPU through WebGPU (Metal, Vulkan or Direct3D 12) by default instead of the CPU, and CUDA when onnxruntime-gpu is installed; SAM3 switches to a new model and downloads again
  • SAM3 is the default model for AI-Find All instead of YOLO-World, and users on YOLO-World must pick it again in Settings to keep it
  • AI inference runs in a separate process, so a slow or crashing model no longer freezes or closes the window
  • AI-Segment and AI-Find All retry on the CPU when the GPU fails, instead of failing
  • A slow AI-Segment prompt shows a spinning ring where it was placed, and the status bar names the running task and reports how many new objects AI-Find All found
  • The AI Assist and AI Text Prompt toolbar groups merge into one AI group, and Run reads Finding… while it runs
View originalRead the notes
How v7.10.0 went

v7.9.1

Fixed 2
  • A point dragged inside a box is no longer lost on image switch, because the move is now saved and added to undo history.
  • Hovering where Shapes overlap now highlights the Shape a click selects, so a point drawn inside a box is highlighted and picked over the box.
View originalRead the notes
How v7.9.1 went

v7.9.0

Added 7
  • Allow double-clicking a shape on the canvas to open the label-edit dialog
  • Lock shapes against editing with Lock Shapes / Unlock Shapes (Ctrl+Shift+L), with the lock saved per shape as "locked": true in the annotation file
  • Add Fit to Selection (Ctrl+Shift+F) to the View menu and toolbar, a zoom mode that fits the view to the selected shapes with padding whenever the selection changes
  • Remember the Annotation Folder for each opened folder across restarts, including one given with --output, which is now created at startup if missing
  • Add a footer to the Image List showing the Annotation Folder and how many images are annotated, with Change Annotation Folder and an option to return to saving next to images, plus a warning when a remembered folder is missing
  • Choose a model that is not yet downloaded from a toolbar model picker to download it and select it when the files arrive, with SAM3 showing its license notice first
  • Add Restore Defaults and Close buttons to the Brightness/Contrast dialog
Changed 14
  • Show Fit to Window and Fit to Selection as icon buttons beside the toolbar zoom field, with a caption naming the active zoom mode
  • Image List shows image paths relative to the opened folder with the full path on hover, and File Search matches the shown path
  • Menus, dialogs, Settings, and CLI help now say "annotation file" instead of "label file", and File menu "Delete File" is now "Delete Annotation File"
  • The File List dock is renamed Image List
  • Output Dir is renamed Annotation Folder and Open Dir is renamed Open Folder, with the Annotation Folder status message translated
  • AI Models opens in its own window from File > AI Models or Manage models in a toolbar model picker instead of as a Settings page
Fixed 7
  • An image too large to display now shows the size error immediately instead of after decoding the whole image, which previously took tens of seconds and gigabytes of memory for large GeoTIFFs
  • Multi-band TIFFs stored with separate band planes now open at their real size and colors instead of as a thin, garbled strip
  • Scrolling a zoomed-in image with many shapes is faster because shapes outside the view are skipped
  • Labelme starts faster by loading SciPy only when AI results are turned into shapes
  • Folders with many images open faster
  • Switching to the next or previous image is faster because neighboring images are decoded in the background
  • Image List checkmarks now follow Save As and deletes of any image's Annotation File, not only the open image's
Removed 1
  • Remove Fit to Width from the View menu, with Ctrl+Shift+F now triggering Fit to Selection and configs setting shortcuts.fit_width still loading
View originalRead the notes
How v7.9.0 went

v7.8.1

Fixed 1
  • Avoid image hashing when loading images while preserving AI embedding reuse when revisiting unchanged images.
View originalRead the notes
How v7.8.1 went

v7.8.0

Changed 3
  • Save annotations in the background during editing, with revision-aware save status and safe waiting before navigation, closing, Save As, or deletion
  • Manual Save and Save As replace pending auto-save snapshots, and canceling Save As preserves them
  • Refreshed the app icon with the new Labelme logo, using a rounded macOS-style icon in the Dock
Fixed 8
  • Reduce undo delays in images with many distinct labels
  • Keep auto-save failures visible in the status bar with a Retry action
  • Speed up annotation redraws by drawing vertex outlines individually while preserving overlapping marker fills and highlights
  • Fix Annotation save destinations, stored image references, and File List checkmarks when embedded metadata differs from the selected Image
  • Fix Previous and Next navigation when filtering hides the active Image, and disable navigation actions without a visible neighbor
  • Preserve the current Annotation and unsaved edits after Annotation File deletion fails, and show an error dialog instead of closing the application
  • Give toolbar overflow controls distinct accessible names for screen readers
  • Keep Image references valid when saving and reopening Annotation Files through directory aliases, including macOS /tmp and /private/tmp
View originalRead the notes
How v7.8.0 went

v7.7.0

Changed 2
  • Highlight SAM 2 Balanced and SAM 3 as recommended AI model downloads
  • Lower the default Polygon detail from 80 to 60 so AI-generated polygons have fewer vertices to adjust by hand
View originalRead the notes
How v7.7.0 went

v7.6.0

Added 7
  • Navigate between neighboring shapes from the Edit menu or with configurable Ctrl+Arrow shortcuts (Command+Arrow on macOS)
  • Manage AI models in Settings with background downloads, a cancellable queue, model license links, and deletion
  • Annotation model pickers now show downloaded models
  • Search Settings by name, config key, or related terms in English and the current UI language, then jump directly to a control with the mouse or keyboard
  • Offer to reset an invalid configuration to defaults, saving a backup first and showing its path
  • Configure predefined Shape Flags in Settings with Label patterns, one flag name per line, live updates, and inline validation
  • Existing Shape Flags are preserved when rules or Labels change
Changed 3
  • Delete selected shapes immediately without a confirmation dialog; use Undo to restore them
  • Remove the default Ctrl+P (⌘P on macOS) shortcut for Keep Previous Annotation; the option remains available in the menu and its shortcut can still be customized
  • Use theme-aware neutral icons for Delete Shapes and Delete Selected Vertex, consistent with other editing actions
Fixed 9
  • Explain missing images referenced by annotations with the expected location, and move the underlying error into expandable Details
  • Fixed the missing icon on the Merge Shapes action
  • Restore icons for Copy to Clipboard and Paste from Clipboard in the Edit and canvas context menus
  • Remember the AI Text Prompt model across restarts and keep its Settings control synchronized with the annotation panel
  • Preserve supported settings when upgrading from older versions, and prevent retired overrides from returning after settings are saved
  • Closing the application or cancelling or deleting a model now stops stalled downloads promptly without allowing them to recreate removed files
View originalRead the notes
How v7.6.0 went

v7.5.0

Added 2
  • Accessible empty-state guidance with image, directory, and drag-and-drop entry points when Labelme launches without input
  • Merge selected Mask Shapes sharing one Label into a single Mask Shape via Edit > Merge Shapes or the canvas context menu, preserving every mask pixel in exports; one Undo restores the inputs
Changed 10
  • AI Assist point previews reuse predictions across repaints and keep the previous preview visible while updating
  • Refreshed Simplified Chinese, Traditional Chinese, and Turkish interface translations
  • Use native step arrows and text alignment for the zoom percentage control
  • Clarify the rectangle drawing hint to explain holding Shift for a square
  • Reword the image embedding and vertex removal actions, with the new labels translated in every language
  • Clarify file-dialog captions and the prompt for unsaved annotation changes in all supported languages
Fixed 17
  • AI Text Prompt reports errors during Shape generation instead of closing the app
  • Keep point shapes inside the image when dragging them together with other shapes while out-of-bounds points are disallowed
  • Recalculate Fit to Window and Fit to Width after resizing docks or navigating between images, while preserving saved pan and scroll positions during navigation
  • Improved accessible names for AI and zoom controls and aligned accessible brightness and contrast values with the displayed percentages
  • Preserve typed shape labels when tabbing into suggestions or opening a label with a case-insensitive match, while applying explicit mouse and keyboard choices before acceptance
  • Allow users to override Qt's image allocation limit when opening large images that remain within the raster engine's supported dimensions
Removed 1
  • Remove the right-drag copy/move gesture and its Copy here/Move here menu from the canvas; to copy a shape, use Copy and Paste or Duplicate, then left-drag it into place
View originalRead the notes
How v7.5.0 went

v7.4.1

Fixed 4
  • Synchronized the status bar with shape creation and editing modes immediately
  • Fixed the label dialog appearing unnamed to assistive technology; it and its Label, Group ID, and Description editors now expose localized names
  • Fixed the shape-deletion confirmation claiming the deletion could not be undone when Undo restores the shape, its Shape List row, and the saved shape count
  • Clear the label dialog's stale list highlight when the current label is not in the list
View originalRead the notes
How v7.4.1 went

v7.4.0

Changed 3
  • Improved the Label dialog so blank labels cannot be submitted and shared fields can be edited across Shapes with different Labels when validate_label is exact
  • Reworded the LineStrip tooltip to "Click to place linestrip points; Ctrl+click places the last one"; label, shortcut, icon, and behavior are unchanged
  • Polygon, linestrip, and AI-Points clicks share one canvas handler; Ctrl+Shift+click now also places the last linestrip point, matching AI-Points
Fixed 2
  • Prevent a crash when resizing the window after closing an image in Fit to Window or Fit to Width mode
  • Fixed Keep Previous Scale restoring stale per-Image viewports during navigation
View originalRead the notes
How v7.4.0 went

v7.3.0

Changed 7
  • Renamed the View-menu actions Original Size, Fit Window, and Fit Width to Actual Size, Fit to Window, and Fit to Width, and reworded the zoom tooltips to describe the effect on the displayed image; shortcuts and behavior are unchanged
  • Renamed the Copy Shapes and Paste Shapes actions to Copy to Clipboard and Paste from Clipboard
  • Regrouped the Edit menu and canvas context menu so Undo, clipboard, and shape-editing actions sit in separate sections
  • Reworded the Save As tooltip to Save the labels under a new file name so it no longer reads like the Save tooltip; label, shortcut, icon, and behavior are unchanged
  • The Brightness/Contrast dialog applies adjustments with lookup tables while preserving prior rendering and transparency, and shows each slider value as a percentage
  • Shape labels shown on the canvas now include their group ID, matching the Shape List
  • The dock listing the current Image's Shapes is now titled Shape List
View originalRead the notes
How v7.3.0 went

v7.2.0

Added 1
  • Added Shape Color controls to Settings for automatic, uniform, and per-Label fallback coloring.
Changed 1
  • Improved AI Assist polygon output with cleaner boundaries, grouped disconnected lands, and adjustable detail.
Fixed 4
  • Fixed Undo appearing enabled immediately after opening an image whose Shapes were carried forward via Keep Previous Annotation / Ctrl+Shift, which could silently discard the carried-forward Shapes if clicked before any edit on the new image.
  • Fixed Undo remaining disabled after committing the first Shape on a raw Image; Undo now removes that Shape from the canvas, Annotation List, and both manually saved and auto-saved Annotation Files.
  • Kept the image point beneath the cursor stationary during Ctrl/Cmd-wheel zoom without canvas flicker.
  • Fixed `labelme --help` showing a Python executable and temporary launcher path instead of the stable `labelme` command on Windows with Python 3.14.
View originalRead the notes
How v7.2.0 went

v7.1.0

Added 1
  • Expanded the Settings dialog with a scrollable page of workflow-oriented groups, an icon-labeled section index for quick navigation, and explanatory descriptions, so that auto-save, saving image data, keep-previous annotation/zoom/brightness-contrast, fill-drawing, label sorting/completion/text-field behavior, and the default AI model can all be changed in-app with immediate apply
Changed 4
  • Changed the Settings dialog to keep a stable, resizable window size, with scrolling when the available height cannot fit its content
  • Grouped Shape coloring under one shape_color setting with explicit auto, uniform, and by_label modes, replacing the default_shape_color, shift_auto_shape_color, and label_colors keys; existing Config Files and inline --config values are migrated in memory, preserving their behavior and override precedence
  • Replaced the zoom control's What's This help with a plain hover tooltip naming the Ctrl+Wheel canvas zoom shortcut
  • Changed the release workflow to publish PEP 440 prerelease tags to PyPI with the current Unreleased notes and mark their GitHub releases as prereleases, while stable tags continue to use their exact version sections
Fixed 8
  • Fixed clicking or hovering near a linestrip's unrendered closing line being treated as a hit on the shape; edge hover, add-point-to-edge, and body selection now ignore that phantom segment
  • Fixed the --output guard rejecting only lowercase .json paths; the guard now uses the case-insensitive is_label_file_path helper so upper- or mixed-case file paths are rejected as expected
  • Fixed read_label_file accepting boolean values where a shape's group_id or point coordinates were expected; booleans are now rejected with a clear read error
  • Fixed the AI-model download progress dialog formatting sizes incorrectly by adding a GB tier and rounding-safe unit boundaries
  • Fixed shapes_to_label mishandling a mask shape whose bounding box extends past the image edge; the destination rectangle and matching mask sub-slice are now clipped to the image canvas
  • Fixed the application window and taskbar icon silently falling back to a blank icon by loading the real icons/icon-256.png, and the new_icon tests now assert the icon is non-null
  • Fixed a crash when opening palette-mode images that need re-encoding; non-transparent palette images are converted to RGB before JPEG save, indexed-transparency P images are saved as PNG, and PA images are converted to RGBA before PNG save
  • Fixed the outline of a Mask shape being drawn one pixel down-and-right of its filled pixels by subtracting the contour padding offset so the outline aligns with the fill
Removed 1
  • Removed the logger_level config key, which had no effect; a logger_level entry remaining in an existing ~/.labelmerc is dropped on load
View originalRead the notes
How v7.1.0 went
View all

Discussion

If you publish Labelme, you can claim this product by proving you administer its repository.