🎥 LongE2V: A New Approach to Event-Based Video
The LongE2V framework, based on video diffusion models, has been developed to work with sparse event streams. The model effectively performs reconstruction, prediction, and frame interpolation using Autoregressive Unrolling and Adaptive Context Switching.
🌍 This technology solves the problem of instability and texture blurring when processing data from high-speed event cameras.
👤 This is a step toward ultra-clear video from "noisy" sensor data for robotics and autonomous driving.
Source 1: https://cdfan0627.github.io/LongE2V-page/ Source 2: https://github.com/cdfan0627/LongE2V