Ding-Kexin / IF_CALC

The repository contains the implementations for Coupled Adversarial Learning for Fusion Classification of Hyperspectral and LiDAR Data
15 stars 6 forks source link

训练测试数据集划分 #2

Open icey-zhang opened 1 year ago

icey-zhang commented 1 year ago

请问muffl trento按什么比例划分的测试和训练集啊?论文里面好像没说欸

Ding-Kexin commented 1 year ago

这两个数据集我也是,每类随机选取20个样本用于训练,其余的标记样本用于测试

在 2023-02-07 10:14:52,"zhang jiaqing" @.***> 写道:

请问muffl trento按什么比例划分的测试和训练集啊?论文里面好像没说欸

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

icey-zhang commented 1 year ago

请教一下,我用原始划分数据集的方法和随机取20样本,反而是20样本的结果更好,为什么训练样本点变少了,反而效果更好了呢?

Ding-Kexin commented 1 year ago

您所说的原始划分数据集的方法是指官方提供的标准训练、测试数据集划分吗? 如果是的话,我的理解是,以trento为例,虽然标准的训练样本数量更多,但是训练样本点的分布很紧凑,对于取patch的方法来说,会使得网络处理的信息有冗余甚至有矛盾产生。