FiftyOne

FiftyOne is a local tool for computer vision teams that need to inspect their training data and understand where their models fail. It runs on macOS, Windows and Linux, with Docker support, and is open source under Apache 2.0. Its Python tools and visual App connect dataset curation with model evaluation.

The App displays images, videos and 3D data alongside labels, metadata and predictions. That lets you examine the samples behind an evaluation result rather than rely only on aggregate metrics. Supported workflows include object detection, medical imaging with DICOM and CT scans, and point clouds for 3D vision.

Embedding views help reveal patterns in a dataset and let you select points to inspect the corresponding samples and labels. You can find near-duplicates, pick difficult examples and identify data worth sending for annotation. FiftyOne also flags possible labeling errors, and you can create or edit 2D and 3D labels directly in the App.

For teams with an existing training workflow, integrations include PyTorch, Hugging Face and Ultralytics. You can connect annotation tools or extend FiftyOne for your own needs. The combination of sample-level inspection, label editing and evaluation suits work where dataset quality is a recurring part of model development.

The open-source software runs locally. FiftyOne Enterprise is a separate offering for collaborative cloud workloads, with connections to compute resources and workflow automation.