Event6D: Event-based Novel Object 6D Pose Tracking

Event6D introduces an event-depth tracking framework for real-time 6D object pose tracking in fast and dynamic scenes. The method reconstructs both intensity and depth from sparse event streams at arbitrary timestamps conditioned on the latest depth observation, enabling robust tracking of novel objects without object-specific training. We also introduce a comprehensive benchmark suite consisting of a large-scale synthetic training dataset and complementary real and simulated event datasets for evaluation.