Skip to content
SPECIFICTION
Get listed

COCO Annotator

github.com

Free plan Open source WebWindowsLinuxAPISelf-hosted
Visit github.com
Spec score 9.0 #2 of 24 in AI Image Annotation Tools

What it is

COCO Annotator is a web-based tool for creating image annotations used in training data for localization and object detection. It supports bounding boxes, polygons, segmentation masks, keypoints, and points, along with disconnected shapes treated as one instance, multiple labels on an image segment, and custom metadata. Users can import existing COCO datasets and export annotations in COCO format. Assisted options include DEXTR, MaskRCNN, Magic Wand, semi-trained model annotation, and Google Images dataset generation. The software includes user authentication and a REST API, with a Swagger interface at localhost:5000/api. It is self-hosted on Linux or Windows, and Docker with docker-compose is required because Docker is the only supported installation method. Documentation covers development and production builds, centralized datasets, and external access. The server uses Flask, Eventlet, and Gunicorn, while long-running requests go to RabbitMQ workers. The project recommends HTTPS to encrypt browser-to-site communication and invites users to its Discord community. It is listed as free under an MIT-licensed plan.

Spec sheet

What it does
COCO Annotator is a web application for creating image-training annotations for object detection and localization.Read 15 Oct 2026 · github.com
Purpose
COCO Annotator is a web-based image annotation tool for creating training data for image localization and object detection.Read 1 Oct 2026 · github.com
Deployment
The documentation provides production and development Docker builds, and describes the production build as stable and suitable for a large user base.Read 1 Oct 2026 · github.com
Support
The project invites users to join its Discord community of machine-learning practitioners.Read 1 Oct 2026 · github.com
License
The repository is distributed under the MIT License, including commercial use rights with attribution.Read 15 Oct 2026 · github.com
Annotation API
The project exposes API endpoints for inspecting and analyzing annotation data.Read 15 Oct 2026 · github.com
Annotation features
It supports object segmentation, keypoints, disconnected objects as one instance, multiple labels per image segment, and custom metadata.Read 1 Oct 2026 · github.com
Annotation formats
It directly exports annotations to COCO format and imports datasets already annotated in COCO format.Read 1 Oct 2026 · github.com
Architecture
The web server uses Flask, Eventlet, and Gunicorn, while long-running requests are passed to workers through RabbitMQ.Read 1 Oct 2026 · github.com
Assisted tools
It includes DEXTR, MaskRCNN, Magic Wand, semi-trained model annotation, and Google Images dataset generation.Read 1 Oct 2026 · github.com
Authentication
The feature list includes a user authentication system.Read 1 Oct 2026 · github.com
Authentication and roles
The system includes authentication and Admin, Owner, and User permission roles.Read 15 Oct 2026 · github.com
Backups
Backups require copying the datasets directory and the Docker database volume.Read 15 Oct 2026 · github.com
COCO interoperability
Datasets can be imported from COCO and completed annotations exported directly as COCO data.Read 15 Oct 2026 · github.com
Custom metadata
Each object instance can carry user-defined metadata and multiple labels.Read 15 Oct 2026 · github.com
Data storage
Docker volumes store database-generated data and are described as compatible with both Linux and Windows containers.Read 1 Oct 2026 · github.com
Deployment endpoint
The production compose setup serves the application at localhost port 5000.Read 15 Oct 2026 · github.com
Developer mode
The development compose file separates frontend access at port 5000 from backend API access under /api.Read 15 Oct 2026 · github.com
Disconnected instances
Separate visible parts can be grouped and labeled as one object instance.Read 15 Oct 2026 · github.com
Google Images generation
A feature generates datasets from Google Images searches.Read 15 Oct 2026 · github.com
Independent adoption news
A Microsoft Azure community article dated December 7, 2019 describes using COCO Annotator with Azure storage, Azure ML pipelines, custom models, and Mask R-CNN proposals.Read 15 Oct 2026 · techcommunity.microsoft.com
Independent comparison
A Prolific buyer guide lists COCO Annotator among free open-source annotation tools and notes that such tools trade lower cost and customization for less support and fewer advanced collaboration features.Read 15 Oct 2026 · prolific.com
Installation
Docker and docker-compose are required because Docker is currently the only supported installation method.Read 1 Oct 2026 · github.com
Installation requirement
Docker and docker-compose are the only supported installation method documented by the project.Read 15 Oct 2026 · github.com
Keypoints
The annotation interface supports adding keypoint annotations.Read 15 Oct 2026 · github.com
Limit: annotation statistics bug
GitHub issue reports version 0.11.1 showing 17,453 annotations while reporting zero annotated images.Read 15 Oct 2026 · github.com
Limit: Docker deployment confusion
A Stack Overflow question reports local Docker users seeing RabbitMQ errors and confusion between the hosted demo address and localhost deployment.Read 15 Oct 2026 · stackoverflow.com
Limit: offline installation
A GitHub issue describes manual offline Docker installation problems, missing delete controls, and inability to configure keypoint classes.Read 15 Oct 2026 · github.com
Limit: small VM stability
A Stack Overflow user reported frequent hangs on a 1 GB Google Cloud VM; moving to 4 GB RAM resolved the problem.Read 15 Oct 2026 · stackoverflow.com
Maker identity
Justin Brooks's LinkedIn profile identifies him as the creator of COCO Annotator and places him in the Greater Toronto Area, Canada.Read 15 Oct 2026 · ca.linkedin.com
Model extensibility
External models can integrate by accepting an image POST and returning COCO-formatted annotation data.Read 15 Oct 2026 · github.com
Model-assisted labeling
Images can be sent to semi-trained models, including built-in Mask R-CNN and DEXTR integrations, to generate annotations.Read 15 Oct 2026 · github.com
Praise: COCO ecosystem
The ACM review says COCO import/export works well with frameworks and pretrained models such as Mask R-CNN and YOLO.Read 15 Oct 2026 · records.sigmm.org
Praise: local operation
An ACM SIGMM review praises that the tool runs locally, is free to download, uploads quickly, and offers efficient precise annotation tools.Read 15 Oct 2026 · records.sigmm.org
Praise: usability
The ACM review describes quick navigation, bounding-box, polygon, and eraser tools, and straightforward export.Read 15 Oct 2026 · records.sigmm.org
Release status
The releases page marks version 0.11.1 as the latest release and lists fixes covering environment booleans, keypoint edges, brush performance, thumbnails, file watching, Celery URLs, and exports.Read 15 Oct 2026 · github.com
REST API
The API uses resource-oriented REST URLs, HTTP response codes, and mostly JSON responses, with a Swagger interface at localhost:5000/api.Read 1 Oct 2026 · github.com
REST documentation
The REST API uses predictable resource URLs, JSON responses, HTTP status errors, and provides Swagger UI at /api.Read 15 Oct 2026 · github.com
Scaling
The dedicated-server guidance describes centralized datasets and external access for outsourcing, with a recommended basic instance of 2GB RAM and 2 CPU cores.Read 1 Oct 2026 · github.com
Security guidance
The deployment guide recommends HTTPS through Lets Encrypt and Nginx and says server root access is required.Read 15 Oct 2026 · github.com
Security posture
GitHub reports that the repository has no SECURITY.md policy and no published security advisories.Read 15 Oct 2026 · github.com
Segment annotation
Users can mark complete or partial image segments for training data.Read 15 Oct 2026 · github.com
System requirements
The VPS guide recommends 2 CPU cores and 2 GB RAM for a basic instance, with more memory for images above one megapixel.Read 15 Oct 2026 · github.com
Technology stack
The backend uses Flask, MongoDB, and MongoEngine while the frontend uses Vue, Axios, Paper.js, and Bootstrap.Read 15 Oct 2026 · github.com
Transport security
The deployment guide strongly recommends HTTPS because it encrypts communication between the browser and website.Read 1 Oct 2026 · github.com

Plans and prices

MIT-licensed software Free self-hosted · Docker required Read 1 Oct 2026 · github.com
Open-source self-hosted Free Docker deployment · local web interface · COCO import/export Read 15 Oct 2026 · records.sigmm.org

Every COCO Annotator plan and price →

Strong points and limits

Strong points

  • Exports annotations in COCO format.
  • Supports segmentation masks, keypoints, and bounding boxes.
  • Includes assisted annotation tools.
  • Provides a REST API and user authentication.
  • Free under the listed MIT-licensed plan.

Limits

  • Docker and docker-compose are required.
  • Docker is the only supported installation method.
  • The project recommends HTTPS for deployment.

Who it suits

COCO Annotator suits teams preparing labeled image datasets for localization or object detection, especially those able to deploy and manage a Docker-based self-hosted service. Its import and export support is designed around COCO-format datasets.

COCO Annotator in depth

Read the full write-up

COCO Annotator is an image-labeling application for building training datasets for localization and object detection. It is best for teams that want a broad set of annotation tools and COCO-format data handling in a self-hosted service. Its annotation breadth is appealing, but operating it requires Docker and docker-compose.

Overview

The workflow covers creating image annotations, importing existing COCO datasets and exporting COCO JSON. It supports bounding boxes, polygons, segmentation masks, keypoints and points, along with multiple labels per segment, disconnected objects grouped as one instance and custom metadata. That combination fits varied image-labeling tasks while keeping data centered on COCO; it is a poor fit for anyone who wants a conventional desktop installation.

Key features

Annotation and assistance

DEXTR, MaskRCNN, Magic Wand and semi-trained model annotation provide assisted labeling alongside the core annotation tools. Google Images dataset generation is another way to assemble image data. This range is useful when a project needs more than boxes, though teams should choose another tool if those annotation and COCO workflows do not match their needs.

API and deployment

The resource-oriented REST API uses HTTP response codes and mostly JSON responses; a Swagger interface is available at localhost:5000/api. User authentication is included. The server uses Flask, Eventlet and Gunicorn, and sends long-running requests to workers through RabbitMQ. This is a capable service-oriented setup, but the sole supported installation route—Docker with docker-compose—adds an operational requirement before teams can start.

Production and development Docker builds are documented, with the production build described as stable and suitable for a large user base. Dedicated-server guidance covers centralized datasets and external access for outsourcing, and recommends a basic instance with 2GB RAM and 2 CPU cores. Docker volumes store database-generated data and are described as compatible with Linux and Windows containers. The deployment guide strongly recommends HTTPS to encrypt browser-to-site communication. The project invites users to its Discord community of machine-learning practitioners. GitHub reports no detected SECURITY.md policy and no published security advisories, so organizations should account for that posture in their own security review.

Pricing

COCO Annotator is free under the MIT license: free for self-hosted use, with Docker required. There are no paid tiers or seat, quota, or renewal terms to weigh against the free option. The trade-off is operational rather than financial: users must provide and maintain the Docker environment.

Platforms

It runs as a web application and supports Linux and Windows, with API access and self-hosted deployment. These are service and deployment platforms rather than a native desktop app; Docker is the required installation method.

Who it's for

Choose COCO Annotator if your team needs multiple image-annotation types, assisted labeling and direct COCO import/export, and can operate a Docker-based service. It is less suitable for individuals seeking a standalone desktop tool or teams unwilling to manage self-hosted deployment.

Pros and cons

Pros: Multiple annotation types, custom metadata and grouping support accommodate more than basic box labeling.

Pros: COCO import and export keep datasets aligned with a common object-detection workflow.

Pros: Assisted tools, an API and documented production deployment support a wider service workflow.

Cons: Docker and docker-compose are mandatory, limiting the choices of teams without that operating environment.

Cons: The repository has no detected security policy or published advisories, a gap to consider in deployment planning.

Alternatives

AI Image Annotation Tools is a useful starting point for comparing more options.

Label Studio is worth considering for teams wanting a freemium alternative with a free trial and support for macOS alongside Linux, Windows, web, API and self-hosted platforms.

Amazon SageMaker Autopilot is a paid, usage-based option with a free trial and no minimum fees or upfront commitments; choose it if that pricing approach suits your workflow.

MakeSense.ai is a free browser-based photo-labeling tool, an alternative for readers looking for that format rather than COCO Annotator's Docker deployment.

Prodigy is a paid alternative with a free trial.

Roboflow offers a free tier with 10 credits per month, enough to train about 30 models or run 80,000 inferences; consider it if that allowance fits your usage.

Segments.ai has a free academic license for non-commercial research, with a license request required; it is a better fit for eligible student and researcher use.

Supervisely offers a free Community plan with 5 GB storage, 10,000 files and 2 members, plus unlimited projects and annotations; it is a fit when those hosted limits suit the team.

VGG Image Annotator is a free, BSD-licensed option without paid tiers or stated usage limits, for readers who prefer that alternative.

Verdict

COCO Annotator is a strong choice for teams building COCO-centered image datasets that need varied labels, assisted tools and API access without a license fee. Its main reason to choose it is the combination of annotation breadth and direct COCO handling; its main reason to look elsewhere is the firm Docker requirement and the responsibility of self-hosting.

Verdict

COCO Annotator offers varied annotation types and assisted labeling in a self-hosted, COCO-oriented workflow. Docker is a firm deployment requirement, so teams should be prepared to operate that environment.

Where Specifiction places it

Alternatives to COCO AnnotatorAll alternatives →