November 2023 Summaries

This week's FiftyOne tips and tricks blog covers various questions and answers from the community. Topics include retrieving filepaths of samples not part of a view, clustering images based on similarity using embeddings, filtering with dynamic attributes, computing pairwise IoUs between predicted and ground truth objects, and converting CVAT coordinates into FiftyOne compatible ones. The blog also highlights the open-source nature of FiftyOne and encourages users to join its growing community.

This week's FiftyOne tips and tricks blog focuses on plugins. The latest feature from Voxel51, FiftyOne Plugins allows users to create UI elements and operators for easy and repeatable workflows using Python code. Highlights include creating stunning UIs with the Voxel51 Plugin Management and Development plugin, adding Python modules without worry, developing JavaScript plugins, and incorporating Node modules as dependencies. The blog encourages readers to explore existing plugins in the FiftyOne Plugins Repo and join the community on Slack for further discussions and collaboration.

The text discusses how to use FiftyOne, a platform for managing and analyzing computer vision data, to work with tracking datasets. It provides detailed steps on loading videos and detections into FiftyOne, computing statistics of trajectories, visualizing both ground truth and predicted trajectories, and creating an assessment using FiftyOne's evaluation methods. The text also explains how to evaluate false positive and false negative detections, as well as identity swaps in tracking. It concludes by emphasizing the power of FiftyOne for managing video datasets and gaining insights into model performance.

This week's FiftyOne tips and tricks blog covers various topics such as working with video object tracking, visualizing CVAT annotated skeletons, computing embeddings on video frames, geospatial queries in FiftyOne, and importing videos annotated in JSON with Labelbox. The community members share their questions and the experts provide detailed answers to help users navigate through these topics effectively.

This week's FiftyOne tips and tricks blog covers various questions and answers from the community. Topics include filtering samples that match a criteria, deduplicating data, deleting samples with specific tags, resolving not being able to see image thumbnails from a saved view, and performing evaluations on a subset of data. The blog also encourages readers to join the FiftyOne community and provides information about the open source machine learning toolset that helps improve computer vision models by curating high quality datasets, evaluating models, finding mistakes, visualizing embeddings, and getting to production faster.

The Voxel51 engineering team has released general availability updates for FiftyOneOne 0.22.3 and FiftyOne Teams 1.4.4, which bring numerous enhancements and fixes to streamline computer vision workflows. These updates include optimizations in sort_by_similarity(), bug fixes in the FiftyOne App, improvements in similarity backends for KNN queries, and more. Additionally, there are new features such as fiftyone delegated fail and fiftyone delegated delete commands. The community is encouraged to get involved through various channels like Slack, Meetups, workshops, GitHub contributions, and blogs.

This week's FiftyOne tips and tricks blog covers various questions and answers from the community. Topics include fast filepath replacements, changing the datatype of a field, grouping by multiple sample fields, sorting a dataset's string fields in numeric order, and efficiently merging samples. The blog also encourages users to join the FiftyOne community on Slack, GitHub, Stack Overflow, and Reddit for more tips and tricks related to this open-source machine learning toolset.

The AI, Machine Learning and Data Science Meetup on November 2, 2023 covered various topics such as Foundation Models for Electronic Health Records (EHRs), Exploring the Two Headed Classifier Use Case, and Using AI to Test Software. Michael Wornow discussed how foundation models can be used to improve healthcare systems by harnessing EHR data. Argo Saakyan presented a practical application of computer vision tasks using a two-headed classifier model. Justin Trugman shared insights on how AI can assist developers in testing better throughout the testing pyramid. The event also included Q&A sessions and highlighted upcoming Meetups and workshops.

Over ten weeks, various open source computer vision library plugins were developed using Python and JavaScript. These plugins include Image Quality Issues, Concept Interpolation, AI Art Gallery, Data Ingestion with Twilio, Visual Question Answering, In-App Youtube Player, Image Deduplication, Optical Character Recognition with PyTesseract, Keyword Search, Zero-Shot Prediction, Active Learning, Reverse Image Search, Concept Space Traversal, Plugin Management and Development, Audio Retrieval with ImageBind, and Semantic Document Search. These plugins provide a wide range of functionalities such as image quality assessment, concept interpolation, AI art gallery creation, data ingestion via Twilio, visual question answering, in-app YouTube player, image deduplication, optical character recognition using PyTesseract, keyword search, zero-shot prediction, active learning, reverse image search, concept space traversal, plugin management and development, audio retrieval with ImageBind, and semantic document search.

In this monthly blog series, we celebrate the fifth birthday of Voxel51 Inc., a company that enables developers, scientists, and organizations to build high-quality visual AI datasets and computer vision models. We highlight some key moments from the past year, including community spotlights, product releases, and updates on FiftyOne's GitHub repository. Additionally, we recap interesting questions and answers from the FiftyOne Community Slack channel in our Tips & Tricks blog series. Finally, we announce upcoming computer vision events and share new documentation, blogs, videos, tutorials, integrations, and cheat sheets to help users succeed with FiftyOne.