Browse million-row image tables without writing a GUI

What it does

Kangas is a Python library built around a DataGrid class that stores and displays large-scale multimedia tables. It ingests images, labels, bounding boxes, and metadata, then renders them in an interactive viewer for grouping, sorting, and filtering. The tool reads Pandas DataFrames, HuggingFace Datasets, Parquet, and CSV, and runs inside Jupyter notebooks or as a standalone desktop application.

The interesting bit

The project started inside Comet’s research team as a prototype for visualizing computer-vision corpora, then spun out into a standalone open-source tool. Its CV lineage means bounding boxes, labels, and scores are first-class data types that the UI recognizes and overlays automatically—no manual matplotlib choreography required.

Key highlights

• Built for scale: the UI groups, sorts, and filters across millions of data points.
• First-class computer-vision support: automatically parses bounding boxes, labels, and metadata attached to images.
• Runs inside Jupyter and Google Colab, or as a standalone app on Windows, macOS, and Linux.
• Ingests Pandas DataFrames, HuggingFace Datasets, Parquet, and CSV directly.
• Developed by Comet’s research team but remains free, open source, and independent of the Comet platform.

Caveats

• Explicitly in open beta; the README warns of bugs, rough edges, and rapid feature churn.
• A “very young project,” so long-term stability and API maturity are still unproven.

Verdict

Worth a spin if you need to explore large image or video datasets visually without building a frontend. Stick with Pandas if your data is small and purely tabular.