VGGT
A feed-forward 3D reconstruction model that regresses camera poses and geometric structure in a single forward pass, but does not provide uncertainty estimates.
Timeline (last 90 days)
The feed-forward 3D reconstruction model VGGT can regress camera poses and geometric structure in a single forward pass, but does not provide uncertainty estimates.
On the HardMatch benchmark, VGGT fails on 35% of the image pairs.
A linear probe predicting pose failure on VGGT's frozen features achieved an ROC-AUC of 93.5% on unseen categories.
The linear probe is attached to features of VGGT's fusion layers and only needs to learn a single weight vector to output a score for whether an image pair will fail.
All Facts (4)
The feed-forward 3D reconstruction model VGGT can regress camera poses and geometric structure in a single forward pass, but does not provide uncertainty estimates.
50%UnverifiedOn the HardMatch benchmark, VGGT fails on 35% of the image pairs.
50%UnverifiedA linear probe predicting pose failure on VGGT's frozen features achieved an ROC-AUC of 93.5% on unseen categories.
50%UnverifiedThe linear probe is attached to features of VGGT's fusion layers and only needs to learn a single weight vector to output a score for whether an image pair will fail.
50%