Kangas is an open-source tool from Comet for exploring, analyzing, and visualizing large-scale multimedia datasets. Published on GitHub under the Apache-2.0 license, it has gathered about 1,078 stars and was last actively updated in 2024. Kangas is built for machine learning practitioners who need to actually look at their data — images, video, audio, and text — rather than scrolling through folders of files.
At the heart of Kangas is the DataGrid, a dataframe-like structure designed for multimedia. You log datasets through a straightforward Python API, then explore them in a web interface that renders the media inline next to the metadata. Because it is designed for scale, Kangas can handle datasets far larger than what fits comfortably in a typical notebook-based workflow, and it integrates with the rest of the Comet experiment-tracking ecosystem while remaining usable entirely on its own.
Performance matters at dataset scale: Kangas can group, sort, and filter across millions of data points in seconds through its fast UI. It also has integrated computer vision support, visualizing and filtering bounding boxes, labels, and metadata with no extra setup. Deployment is flexible — run it as a standalone application on Windows, macOS, or Linux, remotely through Google Colab, or embedded in any Jupyter notebook environment, locally or on a server.
Kangas is completely free and open source under the Apache-2.0 license. There is no paid tier for the tool itself.
Machine learning engineers and data scientists who build training datasets from images, video, or audio and need a fast, visual way to audit quality, spot labeling problems, and understand what their models will learn from.