Skip to content

Celebrating Three Years of FiftyOne!

Last week the team at Voxel51 got together to meet, greet, and work towards achieving our mission of bringing transparency and clarity to the world’s data. 

While we were away … open source FiftyOne achieved three amazing milestones. In this post, we share and celebrate those milestones!

FiftyOne 0.21.5 and 0.21.6 are here! 

Last week we announced FiftyOne 0.21.5 and 0.21.6 What’s new? Here are some highlights:

Models :rocket::rocket:

Other goodies :lollipop::chocolate_bar:

Bug fixes :bug::ladybug:

  • Improved robustness when updating datasets in multiple processes concurrently
  • Improved handling of group datasets whose groups may contain missing samples for certain slices
  • Resolved bugs with similarity queries using the sklearn backend
  • Fixed text and checkbox attribute usage when using our CVAT 2.5 integration

Community contributions :raised_hands: :thank_you:

Special thanks to these awesome community members for contributing to this release!

:first_place_medal: Rusteam added Segment Anything to the model zoo #3019

:first_place_medal: timmermansjoy added support for using MPS devices when running Torch models on macOS #2843

:first_place_medal: smidm fixed a bug when exporting keypoints with NaN coordinates in COCO format #3316

:first_place_medal: Sa-Schmi fixed a bug with custom Visualizers in the App #3357

:first_place_medal: NeoKish squashed a number of documentation bugs #3283, #3289, #3290

:first_place_medal: glenn-jocher updated the YOLOv5 exporter to support Ultralytics’ latest dataset format #3393

:first_place_medal: mys007 added bazel support for the App #3338

:first_place_medal: helioshe4 updated the model zoo to officially support torchvision>=0.15.0 #3348

Check out the release notes for a full rundown of the new features! :hugging_face::chart_with_upwards_trend:

Open source FiftyOne crossed 4000 stars on GitHub!

If you’re like me, you star repos as a way to show support for open source projects you love, or to bookmark a repo you visit frequently or want to dive into later. For us at Voxel51, stars are one way we feel the love that the capabilities we’re building in the open source project are meaningful to members of the community.

Last week we crossed 4000 stars! Thank you to everyone who’s been a part of the amazing journey with us so far, and we look forward to continuing to grow and many more milestones ahead!

FiftyOne turns three!

Three years ago, we launched open source FiftyOne, the world’s first (and today’s most prolific!) open source tool for building high-quality datasets and computer vision models. Not yet familiar? FiftyOne is an open source machine learning toolset that enables data science teams to improve the performance of their computer vision models by helping them curate high quality datasets, evaluate models, find mistakes, visualize embeddings, and get to production faster.

It’s been an awesome ride so far and we wanted to take a quick walk back down memory lane and celebrate key accomplishments in the FiftyOne community.

A quick trip down memory lane: key dates

October 18, 2018: Started Voxel51 Inc. to enable developers, scientists, and organizations to build high-quality datasets and computer vision models

August 8, 2019: Announced Seed Funding

June 1, 2020: Released FiftyOne 0.1 to a few dozen private-beta users

August 11, 2020: Open sourced FiftyOne, making it the open source tool for building high-quality datasets and computer vision models

Late July, 2021: Began working with dozens of startups and Fortune 500 enterprises as early adopters of FiftyOne Teams

September 21, 2022: Announced Series A funding and the public availability of FiftyOne Teams

… And today (in honor of August 11, 2023 last week): We celebrate 3 years of open source FiftyOne!

Join the FiftyOne community!

Join the thousands of engineers and data scientists already using FiftyOne to solve some of the most challenging problems in computer vision today!