We are a global, multidisciplinary team advancing AI, computer vision, and graphics publishing at top conferences and collaborating with leading research institutions worldwide.

A transformer-based approach that separates a scene-specific map from a scene-agnostic regressor pre-trained across tens of thousands of environments — maintaining accuracy under novel viewpoints, weather, and lighting conditions without retraining.

Completes full offline scene reconstruction in 13.8 seconds on edge hardware, with no cloud dependency, and supports persistent scene models that update as ground truth changes — directly relevant to ISR workflows in dynamic or denied environments.

Enhances reconstruction and localization by applying probabilistic and learned 3D priors, improving scene coherence, camera pose accuracy, and downstream tasks like relocalization and view synthesis.