DvsGesture

该数据集名为DvsGesture，由IBM使用动态视觉传感器（DVS）录制，包含29名受试者在不同光照条件下执行的11种不同动作的记录。在仅60个周期后，该数据集达到了92.7%的准确率，并且可以与现有最先进的深度网络进行比较。数据集规模方面，包含1176个训练样本和288个测试样本；训练集记录时长为7602秒（约2小时），而测试集的记录时长为1960秒。该数据集的任务是动作识别。This dataset is named DvsGesture. It was recorded by IBM using dynamic vision sensors (DVS), and contains recordings of 11 distinct actions performed by 29 subjects under varying lighting conditions. After merely 60 training epochs, it attains an accuracy of 92.7%, which is on par with current state-of-the-art deep neural networks. Regarding dataset scale, it consists of 1176 training samples and 288 test samples. The total recording duration of the training set is 7602 seconds (approximately 2 hours), while that of the test set is 1960 seconds. The task of this dataset is action recognition.

搜集汇总

背景与挑战

背景概述

该数据集是Auryn脉冲神经网络模拟器的一个分支，实现了事件驱动随机反向传播（eRBP）方法，用于尖峰时序依赖的深度学习。它包含构建和训练脉冲神经网络的源代码、示例脚本以及相关文档，适用于神经形态计算和深度学习研究。