Datasaur
NLP data labeling platform with AI-assisted automation, quality workflows, and private LLM options
Labelme
Intuitive Python tool for polygonal image and video annotation
Labelme provides a graphical interface for creating polygon, rectangle, circle, line, point, and image‑level flag annotations, supporting VOC/COCO export, video labeling, and customizable label sets.
Overview
Overview
Labelme is a Python‑based graphical annotation application built with Qt. It targets computer‑vision researchers, dataset curators, and developers who need precise polygonal, rectangular, circular, line, point, or image‑level flag annotations. The interface runs on Windows, macOS, and Linux, allowing users to label single images or entire directories with minimal setup.
Capabilities & Deployment
The tool supports export to VOC and COCO formats, facilitating seamless integration with popular segmentation and detection pipelines. Video annotation is available directly in the GUI, and label sets can be predefined, auto‑saved, and validated to reduce errors. Installation is flexible: a simple pip install labelme, a downloadable standalone executable that bundles all dependencies, or native packages from major Linux distributions. Annotations are saved as JSON files, optionally omitting embedded image data for lightweight storage. Extensive examples and a command‑line interface further streamline batch processing and custom workflows.
Highlights
Supports polygon, rectangle, circle, line, point, and image‑level flag annotations
Exports to VOC and COCO formats for segmentation and detection tasks
Built‑in video annotation within the same graphical interface
Customizable label sets, auto‑saving, and validation options
Pros
- Cross‑platform Qt GUI works on Windows, macOS, and Linux
- Rich set of primitives covers most 2D annotation needs
- Multiple installation paths: pip, standalone app, or distro packages
- Active community with extensive examples and documentation
Considerations
- GPL‑3.0 license may limit use in proprietary software
- Limited to 2D image and video annotation, no 3D support
- Performance can degrade with very large images due to Qt rendering
- No built‑in collaborative web server for team labeling
Managed products teams compare with
When teams consider Labelme, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Computer‑vision researchers building training datasets
- Dataset curators needing VOC or COCO export capabilities
- Teams requiring quick video frame annotation
- Developers preferring a Python‑based, locally run tool
Not ideal when
- Projects that need collaborative, web‑based labeling platforms
- Real‑time annotation pipelines with strict latency requirements
- 3D point‑cloud or volumetric data annotation tasks
- Organizations requiring a permissive MIT/BSD license
How teams use it
Semantic segmentation dataset creation
Generate VOC/COCO masks for training segmentation models
Instance segmentation labeling
Annotate object polygons to feed instance detection pipelines
Video frame labeling
Produce per‑frame annotations for action‑recognition datasets
Image classification flagging
Assign image‑level tags for dataset cleaning and filtering
Tech snapshot
Python99%
Makefile1%
Frequently asked questions
How can I install Labelme?
Install via pip (`pip install labelme`), download the standalone executable, or use your Linux distribution’s package manager.
What file format does Labelme export?
Annotations are saved as JSON and can be exported to VOC or COCO formats for segmentation and detection tasks.
Can I annotate videos with Labelme?
Yes, the GUI includes video annotation support, allowing frame‑by‑frame labeling.
How do I customize label sets?
Predefine labels in a text file or configure them via the GUI; options include auto‑saving and validation.
Is there a way to omit image data from JSON files?
Use the `--nodata` flag to store only relative image paths, keeping JSON files lightweight.
Project at a glance
Active- Stars
- 15,493
- Watchers
- 15,493
- Forks
- 3,637
LicenseGPL-3.0
Repo age9 years old
Last commityesterday
Primary languagePython
Last synced 4 hours ago