Nikhila Ravi

Nikhila Ravi has helped turn image segmentation from a specialist computer-vision task into an open, promptable platform. As co-leader of the Segment Anything programme, she coordinates research, engineering and deployment across models that identify, isolate and track visual concepts in images and video.

Nikhila Ravi is a director at Meta Superintelligence Labs and co-leads the Segment Anything team. A Cambridge-trained engineer, she has built a career across computer-vision research, large-scale engineering and product translation. Her central work is the Segment Anything model family, developed by broad teams she co-led rather than by one inventor. SAM 3, released in November 2025 and accepted to ICLR 2026, unified detection, segmentation and tracking around text or image-exemplar concept prompts. The project built a data engine with four million unique concept labels across images and videos, including hard negatives, and released the SA-Co benchmark. On that benchmark, the authors reported roughly twice the promptable-concept-segmentation performance of existing systems in both image and video settings. The team also documented important limits, including weaker zero-shot handling of specialised out-of-domain concepts and complex relational phrases. In March 2026, SAM 3.1 introduced object multiplexing, tracking as many as 16 objects in one forward pass. The organisation reported that this doubled medium-object video throughput from 16 to 32 frames per second on one H100 without an accuracy loss. Ravi’s contribution is organisational as well as technical: sustaining an open research platform through successive model generations and practical deployments.