@GithubProjects: Open-source image annotation: draw polygons, boxes, circles, and points https://osp.fyi/labelme
Summary
A look at Labelme, an open-source Python/Qt image annotation tool for drawing polygons, boxes, circles, and points, featuring AI-assisted annotation via SAM, EfficientSAM, YOLO-World, and SAM3, plus VOC/CCO-format exports and video annotation support.
View Cached Full Text
Cached at: 11/03/26, 12:32 PM
Open-source image annotation: draw polygons, boxes, circles, and points
https://t.co/o17JJfPZt7 https://t.co/l8eeKnjjwe
Image annotation in Python with polygons, rectangles, circles, and points
Source: https://www.opensourceprojects.dev/post/labelme
Labelme: A Python Tool for Annotating Images Without the Headache
If you’ve ever tried to build a computer vision dataset, you know the pain: you’ve got hundreds of images, and you need to draw polygons around every dog, every car, every whatever-it-is you’re trying to teach a model to recognize. Doing that by hand in a general-purpose image editor is slow and error-prone. Labelme is a graphical annotation tool written in Python that’s built specifically for this job—and it’s been quietly powering vision datasets for years.
What It Does
Labelme is a desktop annotation tool inspired by the original LabelMe project from MIT. It’s written in Python and uses Qt for its graphical interface, so you get a real native-feeling app rather than a browser tab. You load up images (or video), draw shapes over the things you care about, and label them. The output is structured annotation data you can feed into a training pipeline.
The tool supports the primitives you’d expect: polygons, rectangles, circles, lines, and points. On top of that, it handles image-level flag annotation (useful for classification and dataset cleaning), video annotation, and GUI customization—things like predefined labels, auto-saving, and label validation. It also ships with multi-language support, covering English, Chinese, Japanese, Korean, German, French, and more, with 20 languages available in total.
Why It’s Cool
- **It exports to formats you actually use.**Rather than locking you into some proprietary format, Labelme exports VOC-format datasets for both semantic and instance segmentation, plus COCO-format datasets for instance segmentation. If you’ve worked with these formats before, you know how much time that saves.
- **The AI assistance is a genuine time-saver.**The README lists AI-assisted point-to-polygon and mask annotation using SAM and EfficientSAM models, plus text-to-annotation via YOLO-world and SAM3. In practice, that means you can click a point or type a label and let the model do the tedious boundary-drawing for you. For anyone who’s spent an afternoon tracing object outlines pixel by pixel, this is the kind of feature that changes your workflow.
- **It’s built for real annotation work, not demos.**The GUI customization options (predefined labels, auto-saving, label validation) suggest the maintainers understand that annotation is a long, repetitive task where small conveniences add up. Auto-save alone will save you from at least one minor tragedy.
- **Video annotation is included.**A lot of annotation tools treat video as an afterthought. Labelme doesn’t.
- **There’s a standalone option.**If you want to annotate without touching Python or Qt, there’s a standalone app available at labelme.io. That’s a thoughtful escape hatch for teams where not everyone wants to set up a Python environment.
How to Try It
Since Labelme is a Python package, installation is straightforward. You’ll want to grab it from PyPI:
Then launch it:
From there, you can open an image, start drawing shapes, and assign labels. The repository includes a tutorial underexamples/tutorialthat walks through the basics, along with dedicated example directories for semantic segmentation, instance segmentation, bbox detection, classification, and video annotation. Those examples are worth skimming—they show the full path from annotation to exported dataset.
If you’d rather skip the Python setup entirely, head tolabelme.iofor the standalone app. And if you run into questions, there’s a Discord community linked from the repo.
You can find the source, examples, and installation details atgithub.com/wkentaro/labelme.
Final Thoughts
Labelme isn’t trying to be everything to everyone—it’s a focused annotation tool that does its job well and exports to formats the vision community already relies on. The addition of SAM and YOLO-based assistance brings it in line with where the field is heading, without abandoning the manual controls that still matter for precise work. If you’re building a segmentation or detection dataset and you’re tired of cobbling together scripts and image editors, this is a sensible place to start. It’s mature, actively maintained, and the standalone option means you don’t have to be a Python developer to use it.
Follow @githubprojects for more developer tools and open source projects.
Similar Articles
Show HN: Free tool to mark points and polygon regions
Tack is a free, browser-based tool that allows users to mark points, draw polygons, and export coordinate data as JSON or YAML without uploading images to a server.
@op7418: https://x.com/op7418/status/2074728162018152817
The user released an open-source AI illustration skill called guizang-material-illustration, based on GPT-Image 2.0 and Codex Agent, which can generate 3D material explanation diagrams with Chinese labels, suitable for articles, weekly reports, PPTs, etc.
@tobi: Better annotating screenshot tool coming to Omarchy Quattro. https://github.com/tobi/omasnap
Omasnap is a new native Wayland screenshot and annotation tool for Omarchy and Hyprland, offering vector layers, OCR capture, and clean window capture via Wayland protocols.
@opensourcelab9: 【Shocking】 An OSS that lets Claude Code handle video editing has racked up 15K stars on GitHub Its name is video-use He…
An open-source tool called video-use lets users edit videos by chatting with Claude Code, automating tasks like cutting filler words, adding subtitles, and color grading, and has gained 15K stars on GitHub.
@helloiamleonie: OpenMed Liquid AI
OpenMed demonstrated using Liquid AI's LFM2.5-VL-3B model to analyze a generated skin image, mapping regions and measuring dimensions locally on a Mac Studio for visual review.