PKU-DAVIS-SOD

该数据集是一个大规模的多模态神经形态目标检测数据集，采用DAVIS346摄像头构建，包含了在各种条件下带有标记物体的混合驾驶序列。该数据集包含276,000个时间戳，并分为训练集（671,300个样本）、验证集（194,700个样本）和测试集（214,100个样本）。它捕捉了关于物体类别、规模、速度和光照强度等多种条件的多样性。数据集包含220个混合驾驶序列，共有1,080,100个边界框，其任务是流式目标检测。

This dataset is a large-scale multimodal neuromorphic object detection dataset constructed with DAVIS346 cameras, encompassing mixed driving sequences with labeled objects under diverse conditions. It contains 276,000 timestamps, and is partitioned into three subsets: training set (671,300 samples), validation set (194,700 samples) and test set (214,100 samples). The dataset covers diverse conditions including object categories, scales, velocities and illumination intensities. It comprises 220 mixed driving sequences with a total of 1,080,100 bounding boxes, and the targeted task is streaming object detection.

搜集汇总

背景与挑战

背景概述

PKU-DAVIS-SOD是一个多模态流式目标检测数据集，包含事件和帧数据，适用于不同场景如正常、低光和环境模糊下的目标检测任务。数据集需通过特定脚本将原始数据转换为同步帧和事件格式以供使用。