__         _     __                     __    
   / /_  _____(_)___/ /___ ____        ____/ /____
  / __ \/ ___/ / __  / __ `/ _ \______/ __  / ___/
 / /_/ / /  / / /_/ / /_/ /  __/_____/ /_/ (__  ) 
/_.___/_/  /_/\__,_/\__, /\___/      \__,_/____/  
                   /____/                         

bridge-ds is a lightweight Python framework designed to provide a unified interface to deep learning datasets from different modalities: Perform global operations, aggregations and queries with a Pandas-like experience, and handle individual samples and raw data using a class-based, tab-completion-ey interface.

• Contents
• Key Features
• Installation
• Documentation

Browse

Browse through your datasets with ease using an intuitive interface.

Work with tables

View your data as tables.

Plot your data

Visualize your data quickly and effectively with the exposed Pandas Plotting API.

Assign, sort and filter

Perform common data operations like assigning new columns, sorting, and filtering with Pandas-like syntax.

Augment

Apply and visualize data augmentations directly within your workflow.

You can install the latest version of Bridge's from PyPI. It comes in a few flavors:

Core: The core package includes the basic functionality of Bridge.

$ uv add bridge-ds

Vision: The vision package includes the core package and additional (opinionated) functionality for working with image datasets.

$ uv add bridge-ds[vision]

• NOTE: to run the demo notebooks locally, you'll need the vision package.

To learn more about bridge-ds, please visit the official documentation.

$ git clone https://github.com/guybuk/bridge-ds.git
$ cd bridge-ds
$ uv sync --all-extras

# Testing
# Default: fast unit tests across core + vision (notebooks excluded)
$ pytest

# Run notebook integration tests (slow, ~5 min, requires `vision` extras)
$ pytest -m slow

# Run everything (CI uses this)
$ pytest -m "slow or not slow"

# Building the docs
$ sudo apt install pandoc
$ cd docs
$ make html

bridge-ds is under active development, currently in a pre-alpha stage.

The following is a rough roadmap of the planned features:

• Video Support
  • DataIO for video
  • DisplayEngine (video player)
  • DatasetProviders (for popular video datasets)
  • Transforms (clipping, sampling, augmentation)
• Text
  • DatasetProviders
  • DisplayEngine (adapt existing engine to work with classic text tasks: translation, Q&A, etc.)
• Core
  • DualDatasets (for tasks with two main elements e.g. image-image, image-text,text-text)
  • Stress testing (currently have no capacity to test huge datasets)