TIDES: Time-Derivative Event Simulation via Deformable Reconstruction
Researchers have developed TIDES, a novel continuous-time event simulator for event cameras. Unlike previous simulators that infer timestamps from frame sequences, TIDES uses a dynamic Gaussian splatting approach to derive per-pixel intensity dynamics directly from an explicit 3D scene representation. This method accurately predicts multiple event crossings per rendering step and models occlusion dynamics for adaptive time stepping, leading to higher fidelity event streams. The simulated events from TIDES have demonstrated better transferability to real-world downstream tasks compared to existing simulators. AI
IMPACT Improves simulation fidelity for event cameras, potentially accelerating research and development in areas like autonomous driving and robotics.