pytorch 如何把图像数据集进行划分成train,test和val(源文件数据)

1、手上目前拥有数据集是一大坨，没有train，test，val的划分

如图所示

2、目录结构：

|---data
  |---dslr
|---images
		|---back_pack
			|---a.jpg
			|---b.jpg
			...

3、转换后的格式如图

目录结构为：

|---datanews
  |---dslr
|---images
		|---test
		|---train
		|---valid
			|---back_pack
				|---a.jpg
				|---b.jpg
				...

4、代码如下：

4.1 先创建同样结构的层级结构

4.2 然后讲原始数据按照比例划分

4.3 移入到对应的文件目录里面

import os, random, shutil
def make_dir(source, target):
 '''
 创建和源文件相似的文件路径函数
 :param source: 源文件位置
 :param target: 目标文件位置
 '''
 dir_names = os.listdir(source)
 for names in dir_names:
  for i in ['train', 'valid', 'test']:
path = target + '/' + i + '/' + names
if not os.path.exists(path):
 os.makedirs(path)
def divideTrainValiTest(source, target):
 '''
  创建和源文件相似的文件路径
  :param source: 源文件位置
  :param target: 目标文件位置
 '''
 # 得到源文件下的种类
 pic_name = os.listdir(source)
 
 # 对于每一类里的数据进行操作
 for classes in pic_name:
  # 得到这一种类的图片的名字
  pic_classes_name = os.listdir(os.path.join(source, classes))
  random.shuffle(pic_classes_name)
  
  # 按照8：1：1比例划分
  train_list = pic_classes_name[0:int(0.8 * len(pic_classes_name))]
  valid_list = pic_classes_name[int(0.8 * len(pic_classes_name)):int(0.9 * len(pic_classes_name))]
  test_list = pic_classes_name[int(0.9 * len(pic_classes_name)):]
  
  # 对于每个图片，移入到对应的文件夹里面
  for train_pic in train_list:
shutil.copyfile(source + '/' + classes + '/' + train_pic, target + '/train/' + classes + '/' + train_pic)
  for validation_pic in valid_list:
shutil.copyfile(source + '/' + classes + '/' + validation_pic,
target + '/valid/' + classes + '/' + validation_pic)
  for test_pic in test_list:
shutil.copyfile(source + '/' + classes + '/' + test_pic, target + '/test/' + classes + '/' + test_pic)
if __name__ == '__main__':
 filepath = r'../data/dslr/images'
 dist = r'../datanews/dslr/images'
 make_dir(filepath, dist)
 divideTrainValiTest(filepath, dist)

补充：pytorch中数据集的划分方法及eError: take(): argument 'index' (position 1) must be Tensor, not numpy.ndarray错误原因

在使用pytorch框架时，难免需要对数据集进行训练集和验证集的划分，一般使用sklearn.model_selection中的train_test_split方法

该方法使用如下：

from sklearn.model_selection import train_test_split
import numpy as np
import torch
import torch.autograd import Variable
from torch.utils.data import DataLoader
 
traindata = np.load(train_path)# image_num * W * H
trainlabel = np.load(train_label_path)
train_data = traindata[:, np.newaxis, ...]
train_label_data = trainlabel[:, np.newaxis, ...]
 
x_tra, x_val, y_tra, y_val = train_test_split(train_data, train_label_data, test_size=0.1, random_state=0)  # 训练集和验证集使用9:1
 
x_tra = Variable(torch.from_numpy(x_tra))
x_tra = x_tra.float()
y_tra = Variable(torch.from_numpy(y_tra))
y_tra = y_tra.float()
 
x_val = Variable(torch.from_numpy(x_val))
x_val = x_val.float()
y_val = Variable(torch.from_numpy(y_val))
y_val = y_val.float()
 
# 训练集的DataLoader
traindataset = torch.utils.data.TensorDataset(x_tra, y_tra)
trainloader = DataLoader(dataset=traindataset, num_workers=opt.threads, batch_size=8, shuffle=True)  
 
# 验证集的DataLoader
validataset = torch.utils.data.TensorDataset(x_val, y_val)
valiloader = DataLoader(dataset=validataset, num_workers=opt.threads, batch_size=opt.batchSize, shuffle=True)

注意：如果按照如下方式使用，就会报eError: take(): argument 'index' (position 1) must be Tensor, not numpy.ndarray错误

from sklearn.model_selection import train_test_split
import numpy as np
import torch
import torch.autograd import Variable
from torch.utils.data import DataLoader
 
traindata = np.load(train_path)# image_num * W * H
trainlabel = np.load(train_label_path)
 
train_data = traindata[:, np.newaxis, ...]
train_label_data = trainlabel[:, np.newaxis, ...]
 
x_train = Variable(torch.from_numpy(train_data))
x_train = x_train.float()
y_train = Variable(torch.from_numpy(train_label_data))
y_train = y_train.float()
# 将原始的训练数据集分为训练集和验证集，后面就可以使用早停机制
x_tra, x_val, y_tra, y_val = train_test_split(x_train, y_train, test_size=0.1)  # 训练集和验证集使用9:1