UGG: Unified Generative Geometry Model
One model learns appearance, depth, and camera motion through conditional completion. We explain how joint training helps, and how image-based verification selects among geometric predictions at inference time.
29 min read · 2026