Datumaro is a flexible Python-based dataset management framework and command-line tool for building, analyzing, transforming, and converting computer vision datasets in many popular formats. It supports importing and exporting annotations and images across a wide variety of standards like COCO, PASCAL VOC, YOLO, ImageNet, Cityscapes, and many more, enabling easy integration with different training pipelines and tools. Datumaro makes it easy to merge datasets, split them into training/validation/test subsets, filter or transform annotations, and validate annotation quality — all while preserving metadata and supporting detailed statistics. It’s especially useful when you’re dealing with heterogeneous data sources or need to prepare complex datasets for machine learning workflows, freeing you from writing custom scripts for every format conversion.

Features

  • Import/export support for many dataset formats
  • Dataset merging and filtering tools
  • Custom annotation transformation utilities
  • Train/val/test splitting by configurable criteria
  • Dataset statistics and quality checking
  • Python API plus CLI for flexible workflows

Project Samples

Project Activity

See All Activity >

Categories

Data Management

License

MIT License

Follow Datumaro

Datumaro Web Site

Other Useful Business Software
8 Monitoring Tools in One APM. Install in 5 Minutes. Icon
8 Monitoring Tools in One APM. Install in 5 Minutes.

Errors, performance, logs, uptime, hosts, anomalies, dashboards, and check-ins. One interface.

AppSignal works out of the box for Ruby, Elixir, Node.js, Python, and more. 30-day free trial, no credit card required.
Start Free
Rate This Project
Login To Rate This Project

User Reviews

Be the first to post a review of Datumaro!

Additional Project Details

Programming Language

Python

Related Categories

Python Data Management System

Registered

2026-01-07