脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - pytorch Dataset,DataLoader产生自定义的训练数据案例

pytorch Dataset,DataLoader产生自定义的训练数据案例

2021-09-13 00:25pan_jinquan Python

这篇文章主要介绍了pytorch Dataset, DataLoader产生自定义的训练数据案例,具有很好的参考价值,希望对大家有所帮助。一起跟随小编过来看看吧

1. torch.utils.data.Dataset

datasets这是一个pytorch定义的dataset的源码集合。下面是一个自定义Datasets的基本框架,初始化放在__init__()中,其中__getitem__()和__len__()两个方法是必须重写的。

__getitem__()返回训练数据,如图片和label,而__len__()返回数据长度。

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
class CustomDataset(data.Dataset):#需要继承data.Dataset
 def __init__(self):
  # TODO
  # 1. Initialize file path or list of file names.
  pass
 def __getitem__(self, index):
  # TODO
  # 1. Read one data from file (e.g. using numpy.fromfile, PIL.Image.open).
  # 2. Preprocess the data (e.g. torchvision.Transform).
  # 3. Return a data pair (e.g. image and label).
  #这里需要注意的是,第一步:read one data,是一个data
  pass
 def __len__(self):
  # You should change 0 to the total size of your dataset.
  return 0

2. torch.utils.data.DataLoader

DataLoader(object)可用参数:

dataset(Dataset) 传入的数据集

batch_size(int, optional)每个batch有多少个样本

shuffle(bool, optional)在每个epoch开始的时候,对数据进行重新排序

sampler(Sampler, optional) 自定义从数据集中取样本的策略,如果指定这个参数,那么shuffle必须为False

batch_sampler(Sampler, optional) 与sampler类似,但是一次只返回一个batch的indices(索引),需要注意的是,一旦指定了这个参数,那么batch_size,shuffle,sampler,drop_last就不能再制定了(互斥——Mutually exclusive)

num_workers (int, optional) 这个参数决定了有几个进程来处理data loading。0意味着所有的数据都会被load进主进程。(默认为0)

collate_fn (callable, optional) 将一个list的sample组成一个mini-batch的函数

pin_memory (bool, optional) 如果设置为True,那么data loader将会在返回它们之前,将tensors拷贝到CUDA中的固定内存(CUDA pinned memory)中.

drop_last (bool, optional) 如果设置为True:这个是对最后的未完成的batch来说的,比如你的batch_size设置为64,而一个epoch只有100个样本,那么训练的时候后面的36个就被扔掉了。 如果为False(默认),那么会继续正常执行,只是最后的batch_size会小一点。

timeout(numeric, optional) 如果是正数,表明等待从worker进程中收集一个batch等待的时间,若超出设定的时间还没有收集到,那就不收集这个内容了。这个numeric应总是大于等于0。默认为0

worker_init_fn (callable, optional) 每个worker初始化函数 If not None, this will be called on eachworker subprocess with the worker id (an int in [0, num_workers - 1]) as input, after seeding and before data loading. (default: None)

3. 使用Dataset, DataLoader产生自定义训练数据

假设TXT文件保存了数据的图片和label,格式如下:第一列是图片的名字,第二列是label

?
1
2
3
4
5
6
7
8
9
10
0.jpg 0
1.jpg 1
2.jpg 2
3.jpg 3
4.jpg 4
5.jpg 5
6.jpg 6
7.jpg 7
8.jpg 8
9.jpg 9

也可以是多标签的数据,如:

?
1
2
3
4
5
6
7
8
9
10
0.jpg 0 10
1.jpg 1 11
2.jpg 2 12
3.jpg 3 13
4.jpg 4 14
5.jpg 5 15
6.jpg 6 16
7.jpg 7 17
8.jpg 8 18
9.jpg 9 19

图库十张原始图片放在./dataset/images目录下,然后我们就可以自定义一个Dataset解析这些数据并读取图片,再使用DataLoader类产生batch的训练数据

3.1 自定义Dataset

首先先自定义一个TorchDataset类,用于读取图片数据,产生标签:

注意初始化函数:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
import torch
from torch.autograd import Variable
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
import numpy as np
from utils import image_processing
import os
 
class TorchDataset(Dataset):
 def __init__(self, filename, image_dir, resize_height=256, resize_width=256, repeat=1):
  '''
  :param filename: 数据文件TXT:格式:imge_name.jpg label1_id labe2_id
  :param image_dir: 图片路径:image_dir+imge_name.jpg构成图片的完整路径
  :param resize_height 为None时,不进行缩放
  :param resize_width 为None时,不进行缩放,
        PS:当参数resize_height或resize_width其中一个为None时,可实现等比例缩放
  :param repeat: 所有样本数据重复次数,默认循环一次,当repeat为None时,表示无限循环<sys.maxsize
  '''
  self.image_label_list = self.read_file(filename)
  self.image_dir = image_dir
  self.len = len(self.image_label_list)
  self.repeat = repeat
  self.resize_height = resize_height
  self.resize_width = resize_width
 
  # 相关预处理的初始化
  '''class torchvision.transforms.ToTensor'''
  # 把shape=(H,W,C)的像素值范围为[0, 255]的PIL.Image或者numpy.ndarray数据
  # 转换成shape=(C,H,W)的像素数据,并且被归一化到[0.0, 1.0]的torch.FloatTensor类型。
  self.toTensor = transforms.ToTensor()
 
  '''class torchvision.transforms.Normalize(mean, std)
  此转换类作用于torch. * Tensor,给定均值(R, G, B) 和标准差(R, G, B),
  用公式channel = (channel - mean) / std进行规范化。
  '''
  # self.normalize=transforms.Normalize()
 
 def __getitem__(self, i):
  index = i % self.len
  # print("i={},index={}".format(i, index))
  image_name, label = self.image_label_list[index]
  image_path = os.path.join(self.image_dir, image_name)
  img = self.load_data(image_path, self.resize_height, self.resize_width, normalization=False)
  img = self.data_preproccess(img)
  label=np.array(label)
  return img, label
 
 def __len__(self):
  if self.repeat == None:
   data_len = 10000000
  else:
   data_len = len(self.image_label_list) * self.repeat
  return data_len
 
 def read_file(self, filename):
  image_label_list = []
  with open(filename, 'r') as f:
   lines = f.readlines()
   for line in lines:
    # rstrip:用来去除结尾字符、空白符(包括\n、\r、\t、' ',即:换行、回车、制表符、空格)
    content = line.rstrip().split(' ')
    name = content[0]
    labels = []
    for value in content[1:]:
     labels.append(int(value))
    image_label_list.append((name, labels))
  return image_label_list
 
 def load_data(self, path, resize_height, resize_width, normalization):
  '''
  加载数据
  :param path:
  :param resize_height:
  :param resize_width:
  :param normalization: 是否归一化
  :return:
  '''
  image = image_processing.read_image(path, resize_height, resize_width, normalization)
  return image
 
 def data_preproccess(self, data):
  '''
  数据预处理
  :param data:
  :return:
  '''
  data = self.toTensor(data)
  return data

3.2 DataLoader产生批训练数据

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
if __name__=='__main__':
 train_filename="../dataset/train.txt"
 # test_filename="../dataset/test.txt"
 image_dir='../dataset/images'
 
 epoch_num=2 #总样本循环次数
 batch_size=7 #训练时的一组数据的大小
 train_data_nums=10
 max_iterate=int((train_data_nums+batch_size-1)/batch_size*epoch_num) #总迭代次数
 
 train_data = TorchDataset(filename=train_filename, image_dir=image_dir,repeat=1)
 # test_data = TorchDataset(filename=test_filename, image_dir=image_dir,repeat=1)
 train_loader = DataLoader(dataset=train_data, batch_size=batch_size, shuffle=False)
 # test_loader = DataLoader(dataset=test_data, batch_size=batch_size,shuffle=False)
 
 # [1]使用epoch方法迭代,TorchDataset的参数repeat=1
 for epoch in range(epoch_num):
  for batch_image, batch_label in train_loader:
   image=batch_image[0,:]
   image=image.numpy()#image=np.array(image)
   image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
   image_processing.cv_show_image("image",image)
   print("batch_image.shape:{},batch_label:{}".format(batch_image.shape,batch_label))
   # batch_x, batch_y = Variable(batch_x), Variable(batch_y)

上面的迭代代码是通过两个for实现,其中参数epoch_num表示总样本循环次数,比如epoch_num=2,那就是所有样本循环迭代2次。

但这会出现一个问题,当样本总数train_data_nums与batch_size不能整取时,最后一个batch会少于规定batch_size的大小,比如这里样本总数train_data_nums=10,batch_size=7,第一次迭代会产生7个样本,第二次迭代会因为样本不足,只能产生3个样本。

我们希望,每次迭代都会产生相同大小的batch数据,因此可以如下迭代:注意本人在构造TorchDataset类时,就已经考虑循环迭代的方法,因此,你现在只需修改repeat为None时,就表示无限循环了,调用方法如下:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
'''
下面两种方式,TorchDataset设置repeat=None可以实现无限循环,退出循环由max_iterate设定
'''
train_data = TorchDataset(filename=train_filename, image_dir=image_dir,repeat=None)
train_loader = DataLoader(dataset=train_data, batch_size=batch_size, shuffle=False)
# [2]第2种迭代方法
for step, (batch_image, batch_label) in enumerate(train_loader):
 image=batch_image[0,:]
 image=image.numpy()#image=np.array(image)
 image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
 image_processing.cv_show_image("image",image)
 print("step:{},batch_image.shape:{},batch_label:{}".format(step,batch_image.shape,batch_label))
 # batch_x, batch_y = Variable(batch_x), Variable(batch_y)
 if step>=max_iterate:
  break
# [3]第3种迭代方法
# for step in range(max_iterate):
#  batch_image, batch_label=train_loader.__iter__().__next__()
#  image=batch_image[0,:]
#  image=image.numpy()#image=np.array(image)
#  image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
#  image_processing.cv_show_image("image",image)
#  print("batch_image.shape:{},batch_label:{}".format(batch_image.shape,batch_label))
#  # batch_x, batch_y = Variable(batch_x), Variable(batch_y)

3.3 附件:image_processing.py

上面代码,用到image_processing,这是本人封装好的图像处理包,包含读取图片,画图等基本方法:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
# -*-coding: utf-8 -*-
"""
 @Project: IntelligentManufacture
 @File : image_processing.py
 @Author : panjq
 @E-mail : pan_jinquan@163.com
 @Date : 2019-02-14 15:34:50
"""
 
import os
import glob
import cv2
import numpy as np
import matplotlib.pyplot as plt
 
def show_image(title, image):
 '''
 调用matplotlib显示RGB图片
 :param title: 图像标题
 :param image: 图像的数据
 :return:
 '''
 # plt.figure("show_image")
 # print(image.dtype)
 plt.imshow(image)
 plt.axis('on') # 关掉坐标轴为 off
 plt.title(title) # 图像题目
 plt.show()
 
def cv_show_image(title, image):
 '''
 调用OpenCV显示RGB图片
 :param title: 图像标题
 :param image: 输入RGB图像
 :return:
 '''
 channels=image.shape[-1]
 if channels==3:
  image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR) # 将BGR转为RGB
 cv2.imshow(title,image)
 cv2.waitKey(0)
 
def read_image(filename, resize_height=None, resize_width=None, normalization=False):
 '''
 读取图片数据,默认返回的是uint8,[0,255]
 :param filename:
 :param resize_height:
 :param resize_width:
 :param normalization:是否归一化到[0.,1.0]
 :return: 返回的RGB图片数据
 '''
 
 bgr_image = cv2.imread(filename)
 # bgr_image = cv2.imread(filename,cv2.IMREAD_IGNORE_ORIENTATION|cv2.IMREAD_COLOR)
 if bgr_image is None:
  print("Warning:不存在:{}", filename)
  return None
 if len(bgr_image.shape) == 2: # 若是灰度图则转为三通道
  print("Warning:gray image", filename)
  bgr_image = cv2.cvtColor(bgr_image, cv2.COLOR_GRAY2BGR)
 
 rgb_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2RGB) # 将BGR转为RGB
 # show_image(filename,rgb_image)
 # rgb_image=Image.open(filename)
 rgb_image = resize_image(rgb_image,resize_height,resize_width)
 rgb_image = np.asanyarray(rgb_image)
 if normalization:
  # 不能写成:rgb_image=rgb_image/255
  rgb_image = rgb_image / 255.0
 # show_image("src resize image",image)
 return rgb_image
 
def fast_read_image_roi(filename, orig_rect, ImreadModes=cv2.IMREAD_COLOR, normalization=False):
 '''
 快速读取图片的方法
 :param filename: 图片路径
 :param orig_rect:原始图片的感兴趣区域rect
 :param ImreadModes: IMREAD_UNCHANGED
      IMREAD_GRAYSCALE
      IMREAD_COLOR
      IMREAD_ANYDEPTH
      IMREAD_ANYCOLOR
      IMREAD_LOAD_GDAL
      IMREAD_REDUCED_GRAYSCALE_2
      IMREAD_REDUCED_COLOR_2
      IMREAD_REDUCED_GRAYSCALE_4
      IMREAD_REDUCED_COLOR_4
      IMREAD_REDUCED_GRAYSCALE_8
      IMREAD_REDUCED_COLOR_8
      IMREAD_IGNORE_ORIENTATION
 :param normalization: 是否归一化
 :return: 返回感兴趣区域ROI
 '''
 # 当采用IMREAD_REDUCED模式时,对应rect也需要缩放
 scale=1
 if ImreadModes == cv2.IMREAD_REDUCED_COLOR_2 or ImreadModes == cv2.IMREAD_REDUCED_COLOR_2:
  scale=1/2
 elif ImreadModes == cv2.IMREAD_REDUCED_GRAYSCALE_4 or ImreadModes == cv2.IMREAD_REDUCED_COLOR_4:
  scale=1/4
 elif ImreadModes == cv2.IMREAD_REDUCED_GRAYSCALE_8 or ImreadModes == cv2.IMREAD_REDUCED_COLOR_8:
  scale=1/8
 rect = np.array(orig_rect)*scale
 rect = rect.astype(int).tolist()
 bgr_image = cv2.imread(filename,flags=ImreadModes)
 
 if bgr_image is None:
  print("Warning:不存在:{}", filename)
  return None
 if len(bgr_image.shape) == 3: #
  rgb_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2RGB) # 将BGR转为RGB
 else:
  rgb_image=bgr_image #若是灰度图
 rgb_image = np.asanyarray(rgb_image)
 if normalization:
  # 不能写成:rgb_image=rgb_image/255
  rgb_image = rgb_image / 255.0
 roi_image=get_rect_image(rgb_image , rect)
 # show_image_rect("src resize image",rgb_image,rect)
 # cv_show_image("reROI",roi_image)
 return roi_image
 
def resize_image(image,resize_height, resize_width):
 '''
 :param image:
 :param resize_height:
 :param resize_width:
 :return:
 '''
 image_shape=np.shape(image)
 height=image_shape[0]
 width=image_shape[1]
 if (resize_height is None) and (resize_width is None):#错误写法:resize_height and resize_width is None
  return image
 if resize_height is None:
  resize_height=int(height*resize_width/width)
 elif resize_width is None:
  resize_width=int(width*resize_height/height)
 image = cv2.resize(image, dsize=(resize_width, resize_height))
 return image
def scale_image(image,scale):
 '''
 :param image:
 :param scale: (scale_w,scale_h)
 :return:
 '''
 image = cv2.resize(image,dsize=None, fx=scale[0],fy=scale[1])
 return image
 
def get_rect_image(image,rect):
 '''
 :param image:
 :param rect: [x,y,w,h]
 :return:
 '''
 x, y, w, h=rect
 cut_img = image[y:(y+ h),x:(x+w)]
 return cut_img
def scale_rect(orig_rect,orig_shape,dest_shape):
 '''
 对图像进行缩放时,对应的rectangle也要进行缩放
 :param orig_rect: 原始图像的rect=[x,y,w,h]
 :param orig_shape: 原始图像的维度shape=[h,w]
 :param dest_shape: 缩放后图像的维度shape=[h,w]
 :return: 经过缩放后的rectangle
 '''
 new_x=int(orig_rect[0]*dest_shape[1]/orig_shape[1])
 new_y=int(orig_rect[1]*dest_shape[0]/orig_shape[0])
 new_w=int(orig_rect[2]*dest_shape[1]/orig_shape[1])
 new_h=int(orig_rect[3]*dest_shape[0]/orig_shape[0])
 dest_rect=[new_x,new_y,new_w,new_h]
 return dest_rect
 
def show_image_rect(win_name,image,rect):
 '''
 :param win_name:
 :param image:
 :param rect:
 :return:
 '''
 x, y, w, h=rect
 point1=(x,y)
 point2=(x+w,y+h)
 cv2.rectangle(image, point1, point2, (0, 0, 255), thickness=2)
 cv_show_image(win_name, image)
 
def rgb_to_gray(image):
 image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
 return image
 
def save_image(image_path, rgb_image,toUINT8=True):
 if toUINT8:
  rgb_image = np.asanyarray(rgb_image * 255, dtype=np.uint8)
 if len(rgb_image.shape) == 2: # 若是灰度图则转为三通道
  bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_GRAY2BGR)
 else:
  bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)
 cv2.imwrite(image_path, bgr_image)
 
def combime_save_image(orig_image, dest_image, out_dir,name,prefix):
 '''
 命名标准:out_dir/name_prefix.jpg
 :param orig_image:
 :param dest_image:
 :param image_path:
 :param out_dir:
 :param prefix:
 :return:
 '''
 dest_path = os.path.join(out_dir, name + "_"+prefix+".jpg")
 save_image(dest_path, dest_image)
 
 dest_image = np.hstack((orig_image, dest_image))
 save_image(os.path.join(out_dir, "{}_src_{}.jpg".format(name,prefix)), dest_image)

3.4 完整的代码

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
# -*-coding: utf-8 -*-
"""
 @Project: pytorch-learning-tutorials
 @File : dataset.py
 @Author : panjq
 @E-mail : pan_jinquan@163.com
 @Date : 2019-03-07 18:45:06
"""
import torch
from torch.autograd import Variable
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
import numpy as np
from utils import image_processing
import os
 
class TorchDataset(Dataset):
 def __init__(self, filename, image_dir, resize_height=256, resize_width=256, repeat=1):
  '''
  :param filename: 数据文件TXT:格式:imge_name.jpg label1_id labe2_id
  :param image_dir: 图片路径:image_dir+imge_name.jpg构成图片的完整路径
  :param resize_height 为None时,不进行缩放
  :param resize_width 为None时,不进行缩放,
        PS:当参数resize_height或resize_width其中一个为None时,可实现等比例缩放
  :param repeat: 所有样本数据重复次数,默认循环一次,当repeat为None时,表示无限循环<sys.maxsize
  '''
  self.image_label_list = self.read_file(filename)
  self.image_dir = image_dir
  self.len = len(self.image_label_list)
  self.repeat = repeat
  self.resize_height = resize_height
  self.resize_width = resize_width
 
  # 相关预处理的初始化
  '''class torchvision.transforms.ToTensor'''
  # 把shape=(H,W,C)的像素值范围为[0, 255]的PIL.Image或者numpy.ndarray数据
  # 转换成shape=(C,H,W)的像素数据,并且被归一化到[0.0, 1.0]的torch.FloatTensor类型。
  self.toTensor = transforms.ToTensor()
 
  '''class torchvision.transforms.Normalize(mean, std)
  此转换类作用于torch. * Tensor,给定均值(R, G, B) 和标准差(R, G, B),
  用公式channel = (channel - mean) / std进行规范化。
  '''
  # self.normalize=transforms.Normalize()
 
 def __getitem__(self, i):
  index = i % self.len
  # print("i={},index={}".format(i, index))
  image_name, label = self.image_label_list[index]
  image_path = os.path.join(self.image_dir, image_name)
  img = self.load_data(image_path, self.resize_height, self.resize_width, normalization=False)
  img = self.data_preproccess(img)
  label=np.array(label)
  return img, label
 
 def __len__(self):
  if self.repeat == None:
   data_len = 10000000
  else:
   data_len = len(self.image_label_list) * self.repeat
  return data_len
 
 def read_file(self, filename):
  image_label_list = []
  with open(filename, 'r') as f:
   lines = f.readlines()
   for line in lines:
    # rstrip:用来去除结尾字符、空白符(包括\n、\r、\t、' ',即:换行、回车、制表符、空格)
    content = line.rstrip().split(' ')
    name = content[0]
    labels = []
    for value in content[1:]:
     labels.append(int(value))
    image_label_list.append((name, labels))
  return image_label_list
 
 def load_data(self, path, resize_height, resize_width, normalization):
  '''
  加载数据
  :param path:
  :param resize_height:
  :param resize_width:
  :param normalization: 是否归一化
  :return:
  '''
  image = image_processing.read_image(path, resize_height, resize_width, normalization)
  return image
 
 def data_preproccess(self, data):
  '''
  数据预处理
  :param data:
  :return:
  '''
  data = self.toTensor(data)
  return data
 
if __name__=='__main__':
 train_filename="../dataset/train.txt"
 # test_filename="../dataset/test.txt"
 image_dir='../dataset/images'
 
 epoch_num=2 #总样本循环次数
 batch_size=7 #训练时的一组数据的大小
 train_data_nums=10
 max_iterate=int((train_data_nums+batch_size-1)/batch_size*epoch_num) #总迭代次数
 
 train_data = TorchDataset(filename=train_filename, image_dir=image_dir,repeat=1)
 # test_data = TorchDataset(filename=test_filename, image_dir=image_dir,repeat=1)
 train_loader = DataLoader(dataset=train_data, batch_size=batch_size, shuffle=False)
 # test_loader = DataLoader(dataset=test_data, batch_size=batch_size,shuffle=False)
 
 # [1]使用epoch方法迭代,TorchDataset的参数repeat=1
 for epoch in range(epoch_num):
  for batch_image, batch_label in train_loader:
   image=batch_image[0,:]
   image=image.numpy()#image=np.array(image)
   image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
   image_processing.cv_show_image("image",image)
   print("batch_image.shape:{},batch_label:{}".format(batch_image.shape,batch_label))
   # batch_x, batch_y = Variable(batch_x), Variable(batch_y)
 
 '''
 下面两种方式,TorchDataset设置repeat=None可以实现无限循环,退出循环由max_iterate设定
 '''
 train_data = TorchDataset(filename=train_filename, image_dir=image_dir,repeat=None)
 train_loader = DataLoader(dataset=train_data, batch_size=batch_size, shuffle=False)
 # [2]第2种迭代方法
 for step, (batch_image, batch_label) in enumerate(train_loader):
  image=batch_image[0,:]
  image=image.numpy()#image=np.array(image)
  image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
  image_processing.cv_show_image("image",image)
  print("step:{},batch_image.shape:{},batch_label:{}".format(step,batch_image.shape,batch_label))
  # batch_x, batch_y = Variable(batch_x), Variable(batch_y)
  if step>=max_iterate:
   break
 # [3]第3种迭代方法
 # for step in range(max_iterate):
 #  batch_image, batch_label=train_loader.__iter__().__next__()
 #  image=batch_image[0,:]
 #  image=image.numpy()#image=np.array(image)
 #  image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
 #  image_processing.cv_show_image("image",image)
 #  print("batch_image.shape:{},batch_label:{}".format(batch_image.shape,batch_label))
 #  # batch_x, batch_y = Variable(batch_x), Variable(batch_y)

以上为个人经验,希望能给大家一个参考,也希望大家多多支持服务器之家。如有错误或未考虑完全的地方,望不吝赐教。

原文链接:https://blog.csdn.net/guyuealian/article/details/88343924

延伸 · 阅读

精彩推荐