Rapid Data Experimentation

FiftyOne is the first open-source dataset experimentation tool transforming the ways CV/ML scientists evaluate their data. Search, filter and explore visual data to curate datasets that enhance predictive model performance.

Intuitive & Extensible API

FiftyOne datasets are flexible and scalable, allowing you to naturally integrate into existing workflows with a few lines of code. FiftyOne has a schema-free data model and user-defined fields without dataset size limits to put the power of data in your hands.

With FiftyOne...

Evaluate Pre-Training


Evaluate the quality of your dataset before you even train your model.

Optimize for Efficiency


Iterate on your data to achieve better performance more quickly.

Use Anytime, Anywhere


We meet needs for cloud, remote, and on-prem for wherever your data lives.

Download FiftyOne to start experimenting with your data in just minutes.

Curate the right data for your models.

With a scientific foundation and established, repeatable workflows, stop wrangling and take control of your data.

Create Datasets

Work with public datasets or create your own from scratch

Find Mistakes

Detect and visualize potential label mistakes in your data

De-Duplication

Find and remove duplicate images from your dataset

Inference-in-the-Loop

Integrate FiftyOne with your existing machine learning workflows

Uniqueness

Find diverse samples and near-duplicates in your data

Take the Dark Arts out of
Data Curation

Without professional tools or established best practices, machine learning and data scientists are left to develop custom solutions and workflows that lack evidence-based methods and repeatability. Stop asking "Why didn't this work?" or "Why did this work? And how do I do it again?" with no clear answers. FiftyOne takes the guesswork and witchcraft out of curating quality data.

Lightweight Tooling, Impactful Features

Documentation & Code Links

Install Guide Docs
Tutorials Docs
Recipes Docs
Release Notes Docs
API Docs
Filter, augment, and transform data
Connect to your data locally, remote, on the cloud... anywhere
Quantitatively measure data attributes, including hardness and uniqueness
Dynamically search, filter, and sort your data with infinite scrolling
Convert data into formats ready for common libraries
Connect with annotation provider's API
Interactive handoff between code and GUI
Visualize 2D detections, instance and semantic segmentation, and more