DVSMOTION20

该数据集旨在增强基于进度事件的光流算法。使用具有 346 x 260 空间分辨率的 IniVation DAViS346 相机收集数据。数据集分为相机运动数据（静止场景和运动相机）和物体运动数据（静止相机和运动物体）。摄像机运动数据包含四个真实的室内序列（即棋盘、教室、会议室和会议室翻译），并从 IMU 推断出地面真实运动。该类相机的运动受到云台的限制，每次采集前都会对IMU进行校准。对象运动数据包括包含多个对象运动的两个真实序列（称为手和汽车）。由于无法从 IMU 推断出物体运动，因此该类别没有地面真实运动。

This dataset is designed to enhance event-based optical flow algorithms. Data was collected using an IniVation DAVIS346 camera with a spatial resolution of 346 × 260. The dataset is divided into two categories: camera motion data (static scenes with moving camera) and object motion data (static camera with moving objects). The camera motion data category contains four real-world indoor sequences, namely Checkerboard, Classroom, Meeting Room and Translation Room, and the ground-truth motion is inferred from IMU readings. The camera motion in this category is constrained by a gimbal, and the IMU is calibrated prior to each data acquisition session. The object motion data category includes two real-world sequences with multiple moving objects, referred to as Hand and Car. Since object motion cannot be inferred from the IMU, this category has no available ground-truth motion.

搜集汇总

背景与挑战

背景概述

DVSMOTION20是一个专为光流算法设计的数据集，包含相机运动和物体运动两类数据，分别提供和不提供地面真实运动信息，适用于视频CV和光流估计研究。