Search papers, labs, and topics across Lattice.
Affiliation:
4
0
5
Standard perspective backbones can solve feed-forward 3D reconstruction across arbitrary mixtures of uncalibrated perspective, fisheye, and 360掳 panoramas in a single pass, beating camera-conditioned baselines by over 26 mAA without any architectural redesign.
Achieve higher-fidelity, one-shot head avatars by autoregressively splitting 3D Gaussians, guided by a GNN, to capture fine-grained facial details.
Achieve real-time cattle mounting pose estimation in complex environments with FSMC-Pose, a framework that outperforms existing methods while drastically reducing computational costs.
Finally, digital humans can have realistic, socially aware conversations: DyaDiT generates dyadic gestures that users strongly prefer over existing methods.
