← Back to feed

Déjà View: Looping Transformers for Multi-View 3D Reconstruction

AI · Studies · 2026-11-13 20:35

NVIDIA introduces Déjà View, a looping Transformer model for multi-view 3D reconstruction that outperforms billion-parameter baselines on five benchmarks with only 117 million parameters, significantly reducing computational costs.

Source: research.nvidia.com


Auto-generated summary for reference only. RSS