DDD20 (DAVIS Driving Dataset 2020)

DDD20 是第一个公共端到端汽车驾驶训练数据集的扩展版本（数据量是原始 DDD17 的 4 倍多），使用神经形态仿生硅视网膜事件相机，DAVIS 事件+帧相机在研究所的传感器组开发。苏黎世 UZH-ETH 神经信息学系。
DDD20包括转向、油门、刹车等汽车数据，可用于评估汽车驾驶辅助应用的帧和事件数据融合情况。

DDD20 is an extended version of the first public end-to-end automotive driving training dataset, with a dataset size exceeding four times that of the original DDD17. The data was collected using the neuromorphic biomimetic silicon retinal event camera DAVIS (event + frame camera), developed by the Sensor Group of the Department of Neuroinformatics, University of Zurich and ETH Zurich (UZH-ETH Zurich).
DDD20 includes automotive driving data such as steering commands, throttle inputs, and brake signals, which can be used to evaluate the fusion performance of frame and event data for automotive driving assistance applications.

搜集汇总

背景与挑战

背景概述

DDD20是一个公共端到端汽车驾驶训练数据集，使用神经形态事件相机（DAVIS事件+帧相机）采集，数据量是前代DDD17的4倍多。该数据集包含转向、油门和刹车等驾驶数据，旨在支持汽车驾驶辅助应用中帧和事件数据的融合研究。