Summary
SphereTrust ranks candidate masks using the geometry of frozen self-supervised features rather than an additional quality model. Across eight candidate pools, it exceeded the strongest evaluated external baseline on six pools by 1.7 to 9.3 Dice points.
Why it matters
More reliable pseudo-label selection can reduce annotation costs and improve the quality of training data for vision systems.
Related coverage
- Leopard Imaging and Lumotive introduce software-defined long-range 3D perception
- Envision launches an AI-optimised 8 MW onshore wind turbine
- AgentX demonstrates long-horizon autonomous model research