FiftyOne
voxel51.com
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What FiftyOne is
FiftyOne Annotation for images helps machine-learning teams find samples for labeling, automate annotation, and connect labeling with model evaluation. Supported annotation types include bounding boxes, polygons, segmentation masks, classifications, polylines, and SAM2 click-to-segment instance segmentation. A natural-language labeling agent can create pre-labels for human review. Workflows support multiple stages, review steps, rejection loops, and quality gates. Listed integrations include CVAT, Label Studio, V7, and Labelbox, plus Qdrant, Redis, and Pinecone for vector search. Deployment options cover cloud or multi-cloud, hybrid, on-premise, air-gapped, and managed services. FiftyOne states that datasets remain in the customer environment rather than being stored or transmitted by the platform. Data anonymization supports protection of personally identifiable and protected health information. Voxel51 says it maintains ISO 27001 certification and undergoes regular third-party penetration and vulnerability tests. The open-source product is available through API, Linux, macOS, self-hosted, web, and Windows. Team, Growth, and Custom plans list enterprise support and deployment options, with prices available through sales.
Who it suits
It is aimed at AI builders and machine-learning teams handling multimodal data at scale. Organizations needing self-hosted, air-gapped, or collaborative annotation workflows may benefit.
Good to know, and what to watch
- Supports six listed image annotation types
- Natural-language agent creates pre-labels
- Includes review loops and quality gates
- Datasets remain in the customer environment
- Offers cloud, hybrid, and air-gapped deployment
- Plan prices require contacting sales
- Advanced support varies by plan
- Annotation depends on listed integrations
Plans
Every plan and priceCustom
Growth
Team
Open source
The verdict
FiftyOne combines open-source annotation workflows with model-assisted labeling and controlled deployment choices. Review the plan-specific support and deployment terms with your team’s infrastructure.
What the maker publishes
- Agentic labeling
- Users can prompt a labeling agent in natural language to generate pre-labels and combine them with human review. source
- Annotation types
- Image annotation supports bounding boxes, polygons, segmentation masks, classifications, polylines, and SAM2 click-to-segment instance segmentation. source
- Company founding
- Voxel51 says it was founded in 2018 at the University of Michigan by Professor Jason Corso and PhD student Brian Moore. source
- Data handling
- The security page states that FiftyOne does not store or transmit datasets and that data remains in the customer's environment. source
- Deployment
- FiftyOne deployment options include cloud or multi-cloud, hybrid, on-premise, air-gapped, and managed services. source
- Founded
- 2018 source
- Image workflows
- The image annotation platform supports multi-stage workflows with review stages, rejection loops, and quality gates. source
- Integrations
- The official integrations documentation lists annotation integrations including CVAT, Label Studio, V7, and Labelbox, and vector search integrations including Qdrant, Redis, and Pinecone. source
- Intended users
- Voxel51 describes FiftyOne as a platform for AI builders and machine-learning teams working with multimodal data at scale. source
- Purpose
- FiftyOne Annotation for images helps identify samples for labeling, automate annotation with AI, and connect labeling with model evaluation. source
- Security
- Voxel51 states that it maintains ISO 27001 certification and undergoes regular third-party penetration and vulnerability tests. source
- Sensitive data
- FiftyOne supports data anonymization to protect personally identifiable and protected health information. source
- Support
- The Team plan lists standard enterprise support, Growth lists premium enterprise support, and Custom lists a dedicated customer success team. source
Checked against the maker's pages on 5 Oct 2026.