Researchers have developed a novel statistical noise model for hybrid event-frame sensors, which combine Event Vision Sensors (EVS) and Active Pixel Sensors (APS) on a single chip. This model uniquely describes the noise characteristics of both APS and EVS pixels, accounting for factors like photon shot noise, dark current, fixed-pattern noise, and quantization noise. The team also created a calibration pipeline to estimate noise parameters from real-world data and introduced H-ESIM, a simulator capable of generating realistic RAW frames and events based on these calibrated noise statistics. Experiments demonstrated the model's effectiveness across various imaging tasks, showing strong performance when transferring simulations to real-world data. AI
IMPACT This research could lead to more accurate simulations for computer vision tasks, improving the development of AI systems that rely on visual data.
RANK_REASON The cluster contains an academic paper detailing a new technical model and simulator. [lever_c_demoted from research: ic=1 ai=0.7]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
