fiftyone: a visual AI development platform for dataset curation and computer vision model evaluation

What it solves

FiftyOne is designed to help developers build high-quality computer vision datasets and models. It addresses the difficulty of visualizing, labeling, and evaluating complex visual AI development workflows, allowing users to move beyond simple file explorers to a dedicated platform for data curation and model analysis.

How it works

FiftyOne provides a powerful visual interface (the FiftyOne App) that allows users to explore images, videos, and their associated labels. It integrates with popular deep learning libraries like PyTorch and Hugging Face, enabling users to visualize embeddings, identify failure modes in model performance, and create or edit 2D and 3D labels directly within the application.

Who it’s for

Computer vision engineers and data scientists working with visual AI, who need a tool for dataset curation, model evaluation, and label management.

Highlights

• Native Annotation: Create and edit 2D and 3D labels directly in the App or integrate with external tools.
• Visual Exploration: Explore images, videos, and associated labels in a unified interface.
• Embedding Exploration: Select and view samples based on their embedding points.
• Model Analysis: Evaluate performance and identify specific failure modes to fine-tune models.
• Data Curation: Quickly find and fix annotation errors and edge cases.
• Rich Integrations: Compatible with PyTorch, Hugging Face, and Ultralytics.