Keras 数据增强ImageDataGenerator多输入多输出实例

缺乏、安全感 2022-12-26 00:58 94阅读 0赞

更多编程教程请到：[菜鸟教程][Link 1] https://www.piaodoo.com/

友情链接：

[高州阳光论坛][Link 2]https://www.hnthzk.com/

[人人影视][Link 3]http://www.op-kg.com/

我就废话不多说了，大家还是直接看代码吧~

import os 
    os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID" 
    os.environ["CUDA_VISIBLE_DEVICES"]=""
    import sys
    import gc
    import time
    import cv2
    import random
    import numpy as np
    import pandas as pd
    import seaborn as sns
    import matplotlib.pyplot as plt
    from tqdm import tqdm

from random\_eraser import get\_random\_eraser  
from keras.preprocessing.image import ImageDataGenerator, array\_to\_img, img\_to\_array, load\_img

datagen = ImageDataGenerator(  
rotation\_range=20, \#旋转  
width\_shift\_range=0.1, \#水平位置平移

# height\_shift\_range=0.2, \#上下位置平移 #

shear\_range=0.5, \#错切变换，让所有点的x坐标(或者y坐标)保持不变，而对应的y坐标(或者x坐标)则按比例发生平移  
zoom\_range=\[0.9,0.9\], \# 单方向缩放，当一个数值时两个方向等比例缩放，参数为list时长宽不同程度缩放。参数大于0小于1时，执行的是放大操作，当参数大于1时，执行的是缩小操作。  
channel\_shift\_range = 40, \#偏移通道数值，改变图片颜色，越大颜色越深  
horizontal\_flip=True, \#水平翻转，垂直翻转vertical\_flip  
fill\_mode=‘nearest’, \#操作导致图像缺失时填充方式。“constant”、“nearest”（默认）、“reflect”和“wrap”  
preprocessing\_function = get\_random\_eraser(p=0.7,v\_l=0,v\_h=255,s\_l=0.01,s\_h=0.03,r\_1=1,r\_2=1.5,pixel\_level=True)  
)

# train\_generator = datagen.flow\_from\_directory( #

# ‘base/Images/’, #

# save\_to\_dir = ‘base/fake/’, #

# batch\_size=1 #

# ) #

# for i in range(5): #

# train\_generator.next() #

# ！ #

# df\_train = pd.read\_csv(‘base/Annotations/label.csv’, header=None) #

# df\_train.columns = \[‘image\_id’, ‘class’, ‘label’\] #

# classes = \[‘collar\_design\_labels’, ‘neckline\_design\_labels’, ‘skirt\_length\_labels’, #

# ‘sleeve\_length\_labels’, ‘neck\_design\_labels’, ‘coat\_length\_labels’, ‘lapel\_design\_labels’, #

# ‘pant\_length\_labels’\] #

# ！ #

# classes = \[‘collar\_design\_labels’\] #

# ！ #

# for i in range(len(classes)): #

# gc.enable() #

# \# 单个分类 #

# cur\_class = classes\[i\] #

# df\_load = df\_train\[(df\_train\[‘class’\] == cur\_class)\].copy() #

# df\_load.reset\_index(inplace=True) #

# del df\_load\[‘index’\] #

# \# print(cur\_class) #

# \# 加载数据和label #

# n = len(df\_load) #

# \# n\_class = len(df\_load\[‘label’\]\[0\]) #

# \# width = 256 #

# \# X = np.zeros((n,width, width, 3), dtype=np.uint8) #

# \# y = np.zeros((n, n\_class), dtype=np.uint8) #

# print(f’starting load trainset \{cur\_class\} \{n\}’) #

# sys.stdout.flush() #

# for i in tqdm(range(n)): #

# \# tmp\_label = df\_load\[‘label’\]\[i\] #

# img = load\_img(‘base/\{0\}’.format(df\_load\[‘image\_id’\]\[i\])) #

# x = img\_to\_array(img) #

# x = x.reshape((1,) + x.shape) #

# m=0 #

# for batch in datagen.flow(x,batch\_size=1): #

# \# plt.imshow(array\_to\_img(batch\[0\])) #

# \# print(batch) #

# array\_to\_img(batch\[0\]).save(f’base/fake/\{format(df\_load\[“image\_id”\]\[i\])\}-\{m\}.jpg’) #

# m+=1 #

# if m>3: #

# break #

# gc.collect() #

# ！ #

img = load\_img(‘base/Images/collar\_design\_labels/2f639f11de22076ead5fe1258eae024d.jpg’)  
plt.figure()  
plt.imshow(img)  
x = img\_to\_array(img)

x = x.reshape((1,) + x.shape)

i = 0  
for batch in datagen.flow(x,batch\_size=5):  
plt.figure()  
plt.imshow(array\_to\_img(batch\[0\]))

# print(len(batch)) #

i += 1  
if i >0:  
break

#多输入，设置随机种子
    # Define the image transformations here
    gen = ImageDataGenerator(horizontal_flip = True,
           vertical_flip = True,
           width_shift_range = 0.1,
           height_shift_range = 0.1,
           zoom_range = 0.1,
           rotation_range = 40)

# Here is the function that merges our two generators #

# We use the exact same generator with the same random seed for both the y and angle arrays #

def gen\_flow\_for\_two\_inputs(X1, X2, y):  
genX1 = gen.flow(X1,y, batch\_size=batch\_size,seed=666)  
genX2 = gen.flow(X1,X2, batch\_size=batch\_size,seed=666)  
while True:  
X1i = genX1.next()  
X2i = genX2.next()  
\#Assert arrays are equal - this was for peace of mind, but slows down training  
\#np.testing.assert\_array\_equal(X1i\[0\],X2i\[0\])  
yield \[X1i\[0\], X2i\[1\]\], X1i\[1\]

#手动构造，直接输出多label
    generator = ImageDataGenerator(rotation_range=5.,
            width_shift_range=0.1, 
            height_shift_range=0.1, 
            horizontal_flip=True, 
            vertical_flip=True)

def generate\_data\_generator(generator, X, Y1, Y2):  
genX = generator.flow(X, seed=7)  
genY1 = generator.flow(Y1, seed=7)  
while True:  
Xi = genX.next()  
Yi1 = genY1.next()  
Yi2 = function(Y2)  
yield Xi, \[Yi1, Yi2\]  
model.fit\_generator(generate\_data\_generator(generator, X, Y1, Y2),  
epochs=epochs)

def batch_generator(generator,X,Y):
     Xgen = generator.flow(X)
     while True:
      yield Xgen.next(),Y
    h = model.fit_generator(batch_generator(datagen, X_all, y_all), 
           steps_per_epoch=len(X_all)//32+1,
           epochs=80,workers=3,
           callbacks=[EarlyStopping(patience=3), checkpointer,ReduceLROnPlateau(monitor='val_loss',factor=0.5,patience=1)], 
           validation_data=(X_val,y_val))

**补充知识：****读取图片成numpy数组,裁剪并保存 和 数据增强(ImageDataGenerator)**

我就废话不多说了，大家还是直接看代码吧~

from PIL import Image
    import numpy as np
    from PIL import Image
    from keras.preprocessing import image
    import matplotlib.pyplot as plt
    import os
    import cv2
    # from scipy.misc import toimage
    import matplotlib
    # 生成图片地址和对应标签
    file_dir = '../train/'
    image_list = []
    label_list = []
    cate = [file_dir + x for x in os.listdir(file_dir) if os.path.isdir(file_dir + x)]
    for name in cate:
     temp = name.split('/')
     path = '../train_new/' + temp[-1]
     isExists = os.path.exists(path)
     if not isExists:
      os.makedirs(path) # 目录不存在则创建
     class_path = name + "/"

for file in os.listdir(class\_path):  
print(file)  
img\_obj = Image.open(class\_path + file) \# 读取图片  
img\_array = np.array(img\_obj)  
resized = cv2.resize(img\_array, (256, 256)) \# 裁剪  
resized = resized.astype(‘float32’)  
resized /= 255.

# plt.imshow(resized) #

# plt.show() #

save\_path = path + ‘/’ + file  
matplotlib.image.imsave(save\_path, resized) \# 保存

keras之数据增强

from PIL import Image
    import numpy as np
    from PIL import Image
    from keras.preprocessing import image
    import os
    import cv2
    # 生成图片地址和对应标签
    file_dir = '../train/'

label\_list = \[\]  
cate = \[file\_dir + x for x in os.listdir(file\_dir) if os.path.isdir(file\_dir + x)\]  
for name in cate:  
image\_list = \[\]  
class\_path = name + “/”  
for file in os.listdir(class\_path):  
image\_list.append(class\_path + file)  
batch\_size = 64  
if len(image\_list) < 10000:  
num = int(10000 / len(image\_list))  
else:  
num = 0

# 设置生成器参数 #

datagen = image.ImageDataGenerator(fill\_mode=‘wrap’, \# 填充模式  
rotation\_range=40, \# 指定旋转角度范围  
width\_shift\_range=0.2, \# 水平位置平移  
height\_shift\_range=0.2, \# 上下位置平移  
horizontal\_flip=True, \# 随机对图片执行水平翻转操作  
vertical\_flip=True, \# 对图片执行上下翻转操作  
shear\_range=0.2,  
rescale=1./255, \# 缩放  
data\_format=‘channels\_last’)  
if num > 0:  
temp = name.split(’/’)  
path = ‘…/train\_datage/’ + temp\[-1\]  
isExists = os.path.exists(path)  
if not isExists:  
os.makedirs(path)

for image\_path in image\_list:  
i = 1  
img\_obj = Image.open(image\_path) \# 读取图片  
img\_array = np.array(img\_obj)  
x = img\_array.reshape((1,) + img\_array.shape) \#要求为4维  
name\_image = image\_path.split(’/’)  
print(name\_image)  
for batch in datagen.flow(x,  
batch\_size=1,  
save\_to\_dir=path,  
save\_prefix=name\_image\[-1\]\[:-4\] + ‘\_’,  
save\_format=‘jpg’):  
i += 1  
if i > num:  
break

以上这篇Keras 数据增强ImageDataGenerator多输入多输出实例就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持菜鸟教程www.piaodoo.com。

[Link 1]: https://www.piaodoo.com/
[Link 2]: https://www.hnthzk.com/
[Link 3]: http://www.op-kg.com/