DINOv3
Summary
DINOv3 is a scaled self-supervised vision foundation model suite focused on frozen transfer and dense feature quality.
Role In The Wiki
DINOv3 is the vision SSL reference point for comparing predictive embedding methods, semantic latents, and pixel-space unified models.
FlowWM adds a distinct use: DINOv3 features become the state space of a stochastic future-trajectory flow. This supports their downstream semantic utility while leaving dense-fidelity, calibrated uncertainty, and action-conditioned planning boundaries explicit.
Evidence
Relation To Foundation TSFM Agenda
Use the source-level agenda mapping in dinov3-2025 rather than duplicating verdict rows here.
At the entity level, DINOv3 is the vision SSL reference point for comparing predictive embedding methods, semantic latents, and pixel-space unified models. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.