VGGT-Ω Model Adds Dynamic Scaling And Registers For Vision Tasks

VGGT-Ω @jianyuan_wang, @MinghaoChen23, Shangzhan Zhang, @n_karaev, Johannes Schönberger, @monsieurlabatut, @p_bojanowski, @davnov134, Andrea Vedaldi, Christian Rupprecht tl;dr: dynamic, single head, more data, more scalability, self-supervised->VGGT Registers for downstream applications, interesting https://arxiv.org/abs/2605.15195