“VGGT-Ω” 3D reconstruction models is scaling like LLMs. Instead of relying on slow optimization pipelines, VGGT-Ω predicts cameras and depth in one feed-forward pass, even for dynamic video. This research makes that scaling practical with scene registers, lighter prediction
VGGT-Ω: Scaling 3D Reconstruction Models with Feed-Forward Neural Networks
By
–
