文章目录
0.环境要求1.加载包和数据集1.1加载包1.2加载数据
2.加载和增强图像Generator3.对训练集做增强用于下面的训练模型3.构建ResUNet模型4.Loss & Compile5.Training6.Testing
0.环境要求
Crack 500数据集下载： https://download.csdn.net/download/QH2107/87423329
创建一个环境，python版本为3.6.13
建一个requirements.txt文件
#新建requirements.txt
absl-py==0.15.0
aiohttp==3.7.4.post0
albumentations==1.3.0
argon2-cffi==20.1.0
astor==0.8.1
async-generator==1.10
async-timeout==3.0.1
attrs==21.4.0
backcall==0.2.0
bleach==4.1.0
blinker==1.4
brotlipy==0.7.0
cachetools==4.2.2
certifi==2021.5.30
cffi==1.14.6
chardet==4.0.0
charset-normalizer==2.0.4
click==8.0.3
colorama==0.4.4
coverage==5.5
cryptography==3.4.7
cycler==0.11.0
Cython==0.29.24
dataclasses==0.8
decorator==5.1.1
defusedxml==0.7.1
entrypoints==0.3
gast==0.2.2
google-auth==2.6.0
google-auth-oauthlib==0.4.4
google-pasta==0.2.0
graphviz==0.19.1
grpcio==1.36.1
h5py==2.10.0
idna==3.3
idna-ssl==1.1.0
imageio==2.15.0
importlib-metadata==4.8.1
ipykernel==5.3.4
ipython==7.16.1
ipython-genutils==0.2.0
ipywidgets==7.6.5
jedi==0.17.0
Jinja2==3.0.3
joblib==1.1.1
jsonschema==3.0.2
jupyter==1.0.0
jupyter-client==7.1.2
jupyter-console==6.4.3
jupyter-contrib-core==0.4.0
jupyter-contrib-nbextensions==0.7.0
jupyter-core==4.8.1
jupyter-highlight-selected-word==0.2.0
jupyter-latex-envs==1.4.6
jupyter-nbextensions-configurator==0.6.1
jupyterlab-pygments==0.1.2
jupyterlab-widgets==1.0.0
Keras==2.3.1
Keras-Applications==1.0.8
Keras-Preprocessing==1.1.2
kiwisolver==1.3.1
lxml==3.8.0
Markdown==3.3.4
MarkupSafe==2.0.1
matplotlib==3.3.4
mistune==0.8.4
mkl-fft==1.3.0
mkl-random==1.1.1
mkl-service==2.3.0
multidict==5.1.0
nb-conda==2.2.1
nb-conda-kernels==2.3.1
nbclient==0.5.3
nbconvert==6.0.7
nbformat==5.1.3
nest-asyncio==1.5.1
networkx==2.5.1
notebook==6.4.3
numpy==1.19.2
oauthlib==3.2.0
olefile==0.46
opencv-python==4.5.5.62
opt-einsum==3.3.0
packaging==21.3
pandas==1.1.5
pandocfilters==1.5.0
parso==0.8.3
pickleshare==0.7.5
Pillow==8.4.0
pip==21.3.1
prometheus-client==0.13.1
prompt-toolkit==3.0.20
protobuf==3.17.2
pyasn1==0.4.8
pyasn1-modules==0.2.8
pycparser==2.21
pydot==1.4.2
pydot-ng==2.0.0
pydotplus==2.0.2
Pygments==2.11.2
PyJWT==2.1.0
pyOpenSSL==21.0.0
pyparsing==3.0.4
pyreadline==2.1
pyrsistent==0.17.3
PySocks==1.7.1
python-dateutil==2.8.2
pytz==2022.7.1
PyWavelets==1.1.1
pywin32==228
pywinpty==0.5.7
PyYAML==6.0
pyzmq==22.2.1
qtconsole==5.2.2
QtPy==2.0.1
qudida==0.0.4
requests==2.27.1
requests-oauthlib==1.3.0
rsa==4.7.2
scikit-image==0.17.2
scikit-learn==0.24.2
scipy==1.5.2
seaborn==0.11.2
Send2Trash==1.8.0
setuptools==58.0.4
six==1.16.0
sklean==0.0.3
tensorboard==2.4.0
tensorboard-plugin-wit==1.6.0
tensorflow==2.1.0
tensorflow-estimator==2.1.0
termcolor==1.1.0
terminado==0.9.4
testpath==0.5.0
tf-unet==0.1.2
threadpoolctl==3.1.0
tifffile==2020.9.3
tornado==6.1
traitlets==4.3.3
typing_extensions==4.1.1
urllib3==1.26.8
wcwidth==0.2.5
webencodings==0.5.1
Werkzeug==0.16.1
wheel==0.37.1
widgetsnbextension==3.5.1
win-inet-pton==1.1.0
wincertstore==0.2
wrapt==1.12.1
yarl==1.6.3
zipp==3.6.0
通过下面命令安全所需要的包
conda install --yes --file requirements.txt
或
pip install -r requirements.txt
1.加载包和数据集
! nvidia-smi
Mon Feb 6 12:48:07 2023
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 512.78 Driver Version: 512.78 CUDA Version: 11.6 |
|-------------------------------+----------------------+----------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... WDDM | 00000000:01:00.0 Off | N/A |
| N/A 43C P0 26W / N/A | 0MiB / 6144MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
1.1加载包
import os
import cv2
import shutil
import math
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
sns.set()
import tensorflow as tf
from tensorflow import keras
import tensorflow.keras.backend as K
from tensorflow.keras.utils import Sequence
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, Activation, MaxPooling2D, Conv2DTranspose, Add, concatenate, average, Dropout
from tensorflow.keras.losses import binary_crossentropy
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
from sklearn.metrics import classification_report, roc_auc_score, accuracy_score
from albumentations import Compose, OneOf, Flip, Rotate, RandomContrast, RandomGamma, RandomBrightness, ElasticTransform, GridDistortion, OpticalDistortion, RGBShift, CLAHE
from tensorflow.keras.losses import binary_crossentropy
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
from skimage.transform import resize
from sklearn.metrics import classification_report
1.2加载数据
#数据集的文件夹路径
train_image_dir = r'E:\dataset\CRACK500\train\image'
train_mask_dir = r'E:\dataset\CRACK500\train\mask'
valid_image_dir = r'E:\dataset\CRACK500\validation\image'
valid_mask_dir =r'E:\dataset\CRACK500\validation\mask'
test_image_dir = r'E:\dataset\CRACK500\test\image'
test_mask_dir = r'E:\dataset\CRACK500\test\mask'
#数据集的文件路径（image对应mask）
# 测试集
test_image_paths = sorted([os.path.join(test_image_dir, fname) for fname in os.listdir(test_image_dir) if fname.endswith(".png") and not fname.startswith(".")])
test_mask_paths = sorted([os.path.join(test_mask_dir, fname) for fname in os.listdir(test_mask_dir) if fname.endswith(".png") and not fname.startswith(".")])
print("Number of testing images : ", len(test_image_paths))
print("Number of testing masks : ", len(test_mask_paths))
# 训练集
train_image_files = sorted([os.path.join(train_image_dir, fname) for fname in os.listdir(train_image_dir) if fname.endswith(".png") and not fname.startswith(".")])
train_mask_files = sorted([os.path.join(train_mask_dir, fname) for fname in os.listdir(train_mask_dir) if fname.endswith(".png") and not fname.startswith(".")])
print("Number of training images : ", len(train_image_files))
print("Number of training masks : ", len(train_mask_files))
#验证集
valid_image_files = sorted([os.path.join(valid_image_dir, fname) for fname in os.listdir(valid_image_dir) if fname.endswith(".png") and not fname.startswith(".")])
valid_mask_files = sorted([os.path.join(valid_mask_dir, fname) for fname in os.listdir(valid_mask_dir) if fname.endswith(".png") and not fname.startswith(".")])
print("Number of validing images : ", len(valid_image_files))
print("Number of validing masks : ", len(valid_mask_files))
#结果
Number of testing images : 1124
Number of testing masks : 1124
Number of training images : 1896
Number of training masks : 1896
Number of validing images : 348
Number of validing masks : 348
batch_size = 4 #批大小，显存不够可以再小一点
img_dim=(320, 640) #图像大小
2.加载和增强图像Generator
class Generator(Sequence):
def __init__(self, x_set, y_set, batch_size=5, img_dim=(128, 128), augment=False):
self.x = x_set
self.y = y_set
self.batch_size = batch_size
self.img_dim = img_dim
self.augment = augment
def __len__(self):
return math.ceil(len(self.x) / self.batch_size)
augmentations = Compose(
[
Flip(p=0.7),
Rotate(p=0.7),
OneOf([
RandomContrast(),
RandomGamma(),
RandomBrightness()
], p=0.3),
OneOf([
ElasticTransform(alpha=120, sigma=120 * 0.05, alpha_affine=120 * 0.03),
GridDistortion(),
OpticalDistortion(distort_limit=2, shift_limit=0.5)
], p=0.3),
])
def __getitem__(self, idx):
batch_x = self.x[idx * self.batch_size:(idx + 1) * self.batch_size]
batch_y = self.y[idx * self.batch_size:(idx + 1) * self.batch_size]
batch_x = np.array([cv2.resize(cv2.cvtColor(cv2.imread(file_name, -1), cv2.COLOR_BGR2RGB), (self.img_dim[1], self.img_dim[0])) for file_name in batch_x])
batch_y = np.array([(cv2.resize(cv2.imread(file_name, -1), (self.img_dim[1], self.img_dim[0]))>0).astype(np.uint8) for file_name in batch_y])
if self.augment is True:
aug = [self.augmentations(image=i, mask=j) for i, j in zip(batch_x, batch_y)]
batch_x = np.array([i['image'] for i in aug])
batch_y = np.array([j['mask'] for j in aug])
batch_y = np.expand_dims(batch_y, -1)
return batch_x/255, batch_y/1
# 对测试集处理
test1_generator=Generator(test_image_paths,test_mask_paths,batch_size,img_dim,False)
# 生成的样本 (未增强)
for i, j in test1_generator:
break
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Original Images (test)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(i[:4], axes[:4]):
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 4, figsize=(13,3))
fig.suptitle('Original Masks (test)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(j[:4], axes[:4]):
ax.imshow(np.squeeze(img, -1), cmap='gray')
ax.axis('off')
plt.tight_layout()
plt.show()
# 对训练集和验证集做同样的处理
train_generator = Generator(train_image_files, train_mask_files,batch_size,img_dim,False)
validation_generator = Generator(valid_image_files, valid_mask_files,batch_size,img_dim,False)
for i, j in train_generator:
break
print(i.shape)
print(j.shape)
(4, 320, 640, 3)
(4, 320, 640, 1)
for i, j in validation_generator:
break
print(i.shape)
print(j.shape)
(4, 320, 640, 3)
(4, 320, 640, 1)
# 生成的样本(未增强)
for i, j in train_generator:
break
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Original Images (train)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(i[:4], axes[:4]):
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Original Masks (train)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(j[:4], axes[:4]):
ax.imshow(np.squeeze(img, -1), cmap='gray')
ax.axis('off')
plt.tight_layout()
plt.show()
# 生成的样本 (未增强)
for i, j in validation_generator:
break
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Original Images (validation)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(i[:4], axes[:4]):
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Original Masks (validation)', fontsize=15)
axes = axes.flatten()
for img, ax in zip(j[:4], axes[:4]):
ax.imshow(np.squeeze(img, -1), cmap='gray')
ax.axis('off')
plt.tight_layout()
plt.show()
3.对训练集做增强用于下面的训练模型
tg = Generator(train_image_files, train_mask_files, batch_size, img_dim, augment = True) #训练集
vg = Generator(valid_image_files, valid_mask_files, batch_size, img_dim, augment = False)#验证集
for i, j in tg:
break
print(i.shape)
print(j.shape)
#结果
(4, 320, 640, 3)
(4, 320, 640, 1)
for i, j in vg:
break
print(i.shape)
print(j.shape)
#结果
(4, 320, 640, 3)
(4, 320, 640, 1)
# Augmented train
for i, j in tg:
break
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Augmented Images', fontsize=15)
axes = axes.flatten()
for img, ax in zip(i[:4], axes[:4]):
ax.imshow(img)
ax.axis('off')
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 4, figsize=(13,2.5))
fig.suptitle('Augmented Masks', fontsize=15)
axes = axes.flatten()
for img, ax in zip(j[:4], axes[:4]):
ax.imshow(np.squeeze(img, -1), cmap='gray')
ax.axis('off')
plt.tight_layout()
plt.show()
3.构建ResUNet模型
import numpy as np
from tensorflow.keras.backend import int_shape
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, UpSampling2D, Add, BatchNormalization, Input, Activation, Concatenate
from keras.regularizers import l2
# BatchNormalization and Activation
def BN_Act(x, act = True):
x = BatchNormalization()(x)
if act == True:
x = Activation("relu")(x)
return x
#conv2d block
def conv2d_block(x, filters, kernel_size = (3, 3), padding = "same", strides = 1):
conv = BN_Act(x)
conv = Conv2D(filters, kernel_size, padding = padding, strides = strides)(conv)
return conv
#Fixed layer.
def stem(x, filters, kernel_size=(3, 3), padding="same", strides=1):
conv = Conv2D(filters, kernel_size, padding = padding, strides = strides)(x)
conv = conv2d_block(conv, filters, kernel_size = kernel_size, padding = padding, strides = strides)
#skip
shortcut = Conv2D(filters, kernel_size = (1, 1), padding = padding, strides = strides)(x)
shortcut = BN_Act(shortcut, act = False) # No activation in skip connection
output = Add()([conv, shortcut])
return output
# Residual Block
def residual_block(x, filters, kernel_size = (3, 3), padding = "same", strides = 1):
res = conv2d_block(x, filters, kernel_size = kernel_size, padding = padding, strides = strides)
res = conv2d_block(res, filters, kernel_size = kernel_size, padding = padding, strides = 1)
shortcut = Conv2D(filters, kernel_size = (1, 1), padding = padding, strides = strides)(x)
shortcut = BN_Act(shortcut, act = False) # No activation in skip connection
output = Add()([shortcut, res])
return output
# Upsampling Concatenation block
def upsample_concat_block(x, xskip):
u = UpSampling2D((2, 2))(x)
c = Concatenate()([u, xskip])
return c
# MODEL
def ResUNet():
f = [16, 32, 64, 128, 256]
inputs = Input((img_dim[0], img_dim[1], 3))
## Encoder/downsampling/contracting path
e0 = inputs
e1 = stem(e0, f[0])
e2 = residual_block(e1, f[1], strides = 2)
e3 = residual_block(e2, f[2], strides = 2)
e4 = residual_block(e3, f[3], strides = 2)
e5 = residual_block(e4, f[4], strides = 2)
## Bridge/Bottleneck
b0 = conv2d_block(e5, f[4], strides = 1)
b1 = conv2d_block(b0, f[4], strides = 1)
## Decoder/upsampling/expansive path
u1 = upsample_concat_block(b1, e4)
d1 = residual_block(u1, f[4])
u2 = upsample_concat_block(d1, e3)
d2 = residual_block(u2, f[3])
u3 = upsample_concat_block(d2, e2)
d3 = residual_block(u3, f[2])
u4 = upsample_concat_block(d3, e1)
d4 = residual_block(u4, f[1])
outputs = Conv2D(1, (1, 1), padding = "same", activation = "sigmoid")(d4)
model = Model(inputs, outputs)
return model
K.clear_session()
model = ResUNet()
model.summary()
#结果
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 320, 640, 3) 0
__________________________________________________________________________________________________
conv2d (Conv2D) (None, 320, 640, 16) 448 input_1[0][0]
__________________________________________________________________________________________________
batch_normalization (BatchNorma (None, 320, 640, 16) 64 conv2d[0][0]
__________________________________________________________________________________________________
activation (Activation) (None, 320, 640, 16) 0 batch_normalization[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 320, 640, 16) 64 input_1[0][0]
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 320, 640, 16) 2320 activation[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 320, 640, 16) 64 conv2d_2[0][0]
__________________________________________________________________________________________________
add (Add) (None, 320, 640, 16) 0 conv2d_1[0][0]
batch_normalization_1[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 320, 640, 16) 64 add[0][0]
__________________________________________________________________________________________________
activation_1 (Activation) (None, 320, 640, 16) 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 160, 320, 32) 4640 activation_1[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 160, 320, 32) 128 conv2d_3[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 160, 320, 32) 544 add[0][0]
__________________________________________________________________________________________________
activation_2 (Activation) (None, 160, 320, 32) 0 batch_normalization_3[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 160, 320, 32) 128 conv2d_5[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 160, 320, 32) 9248 activation_2[0][0]
__________________________________________________________________________________________________
add_1 (Add) (None, 160, 320, 32) 0 batch_normalization_4[0][0]
conv2d_4[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 160, 320, 32) 128 add_1[0][0]
__________________________________________________________________________________________________
activation_3 (Activation) (None, 160, 320, 32) 0 batch_normalization_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 80, 160, 64) 18496 activation_3[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, 80, 160, 64) 256 conv2d_6[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 80, 160, 64) 2112 add_1[0][0]
__________________________________________________________________________________________________
activation_4 (Activation) (None, 80, 160, 64) 0 batch_normalization_6[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, 80, 160, 64) 256 conv2d_8[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 80, 160, 64) 36928 activation_4[0][0]
__________________________________________________________________________________________________
add_2 (Add) (None, 80, 160, 64) 0 batch_normalization_7[0][0]
conv2d_7[0][0]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, 80, 160, 64) 256 add_2[0][0]
__________________________________________________________________________________________________
activation_5 (Activation) (None, 80, 160, 64) 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 40, 80, 128) 73856 activation_5[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, 40, 80, 128) 512 conv2d_9[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 40, 80, 128) 8320 add_2[0][0]
__________________________________________________________________________________________________
activation_6 (Activation) (None, 40, 80, 128) 0 batch_normalization_9[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, 40, 80, 128) 512 conv2d_11[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 40, 80, 128) 147584 activation_6[0][0]
__________________________________________________________________________________________________
add_3 (Add) (None, 40, 80, 128) 0 batch_normalization_10[0][0]
conv2d_10[0][0]
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None, 40, 80, 128) 512 add_3[0][0]
__________________________________________________________________________________________________
activation_7 (Activation) (None, 40, 80, 128) 0 batch_normalization_11[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, 20, 40, 256) 295168 activation_7[0][0]
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None, 20, 40, 256) 1024 conv2d_12[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D) (None, 20, 40, 256) 33024 add_3[0][0]
__________________________________________________________________________________________________
activation_8 (Activation) (None, 20, 40, 256) 0 batch_normalization_12[0][0]
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None, 20, 40, 256) 1024 conv2d_14[0][0]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, 20, 40, 256) 590080 activation_8[0][0]
__________________________________________________________________________________________________
add_4 (Add) (None, 20, 40, 256) 0 batch_normalization_13[0][0]
conv2d_13[0][0]
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None, 20, 40, 256) 1024 add_4[0][0]
__________________________________________________________________________________________________
activation_9 (Activation) (None, 20, 40, 256) 0 batch_normalization_14[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D) (None, 20, 40, 256) 590080 activation_9[0][0]
__________________________________________________________________________________________________
batch_normalization_15 (BatchNo (None, 20, 40, 256) 1024 conv2d_15[0][0]
__________________________________________________________________________________________________
activation_10 (Activation) (None, 20, 40, 256) 0 batch_normalization_15[0][0]
__________________________________________________________________________________________________
conv2d_16 (Conv2D) (None, 20, 40, 256) 590080 activation_10[0][0]
__________________________________________________________________________________________________
up_sampling2d (UpSampling2D) (None, 40, 80, 256) 0 conv2d_16[0][0]
__________________________________________________________________________________________________
concatenate (Concatenate) (None, 40, 80, 384) 0 up_sampling2d[0][0]
add_3[0][0]
__________________________________________________________________________________________________
batch_normalization_16 (BatchNo (None, 40, 80, 384) 1536 concatenate[0][0]
__________________________________________________________________________________________________
activation_11 (Activation) (None, 40, 80, 384) 0 batch_normalization_16[0][0]
__________________________________________________________________________________________________
conv2d_17 (Conv2D) (None, 40, 80, 256) 884992 activation_11[0][0]
__________________________________________________________________________________________________
batch_normalization_17 (BatchNo (None, 40, 80, 256) 1024 conv2d_17[0][0]
__________________________________________________________________________________________________
conv2d_19 (Conv2D) (None, 40, 80, 256) 98560 concatenate[0][0]
__________________________________________________________________________________________________
activation_12 (Activation) (None, 40, 80, 256) 0 batch_normalization_17[0][0]
__________________________________________________________________________________________________
batch_normalization_18 (BatchNo (None, 40, 80, 256) 1024 conv2d_19[0][0]
__________________________________________________________________________________________________
conv2d_18 (Conv2D) (None, 40, 80, 256) 590080 activation_12[0][0]
__________________________________________________________________________________________________
add_5 (Add) (None, 40, 80, 256) 0 batch_normalization_18[0][0]
conv2d_18[0][0]
__________________________________________________________________________________________________
up_sampling2d_1 (UpSampling2D) (None, 80, 160, 256) 0 add_5[0][0]
__________________________________________________________________________________________________
concatenate_1 (Concatenate) (None, 80, 160, 320) 0 up_sampling2d_1[0][0]
add_2[0][0]
__________________________________________________________________________________________________
batch_normalization_19 (BatchNo (None, 80, 160, 320) 1280 concatenate_1[0][0]
__________________________________________________________________________________________________
activation_13 (Activation) (None, 80, 160, 320) 0 batch_normalization_19[0][0]
__________________________________________________________________________________________________
conv2d_20 (Conv2D) (None, 80, 160, 128) 368768 activation_13[0][0]
__________________________________________________________________________________________________
batch_normalization_20 (BatchNo (None, 80, 160, 128) 512 conv2d_20[0][0]
__________________________________________________________________________________________________
conv2d_22 (Conv2D) (None, 80, 160, 128) 41088 concatenate_1[0][0]
__________________________________________________________________________________________________
activation_14 (Activation) (None, 80, 160, 128) 0 batch_normalization_20[0][0]
__________________________________________________________________________________________________
batch_normalization_21 (BatchNo (None, 80, 160, 128) 512 conv2d_22[0][0]
__________________________________________________________________________________________________
conv2d_21 (Conv2D) (None, 80, 160, 128) 147584 activation_14[0][0]
__________________________________________________________________________________________________
add_6 (Add) (None, 80, 160, 128) 0 batch_normalization_21[0][0]
conv2d_21[0][0]
__________________________________________________________________________________________________
up_sampling2d_2 (UpSampling2D) (None, 160, 320, 128 0 add_6[0][0]
__________________________________________________________________________________________________
concatenate_2 (Concatenate) (None, 160, 320, 160 0 up_sampling2d_2[0][0]
add_1[0][0]
__________________________________________________________________________________________________
batch_normalization_22 (BatchNo (None, 160, 320, 160 640 concatenate_2[0][0]
__________________________________________________________________________________________________
activation_15 (Activation) (None, 160, 320, 160 0 batch_normalization_22[0][0]
__________________________________________________________________________________________________
conv2d_23 (Conv2D) (None, 160, 320, 64) 92224 activation_15[0][0]
__________________________________________________________________________________________________
batch_normalization_23 (BatchNo (None, 160, 320, 64) 256 conv2d_23[0][0]
__________________________________________________________________________________________________
conv2d_25 (Conv2D) (None, 160, 320, 64) 10304 concatenate_2[0][0]
__________________________________________________________________________________________________
activation_16 (Activation) (None, 160, 320, 64) 0 batch_normalization_23[0][0]
__________________________________________________________________________________________________
batch_normalization_24 (BatchNo (None, 160, 320, 64) 256 conv2d_25[0][0]
__________________________________________________________________________________________________
conv2d_24 (Conv2D) (None, 160, 320, 64) 36928 activation_16[0][0]
__________________________________________________________________________________________________
add_7 (Add) (None, 160, 320, 64) 0 batch_normalization_24[0][0]
conv2d_24[0][0]
__________________________________________________________________________________________________
up_sampling2d_3 (UpSampling2D) (None, 320, 640, 64) 0 add_7[0][0]
__________________________________________________________________________________________________
concatenate_3 (Concatenate) (None, 320, 640, 80) 0 up_sampling2d_3[0][0]
add[0][0]
__________________________________________________________________________________________________
batch_normalization_25 (BatchNo (None, 320, 640, 80) 320 concatenate_3[0][0]
__________________________________________________________________________________________________
activation_17 (Activation) (None, 320, 640, 80) 0 batch_normalization_25[0][0]
__________________________________________________________________________________________________
conv2d_26 (Conv2D) (None, 320, 640, 32) 23072 activation_17[0][0]
__________________________________________________________________________________________________
batch_normalization_26 (BatchNo (None, 320, 640, 32) 128 conv2d_26[0][0]
__________________________________________________________________________________________________
conv2d_28 (Conv2D) (None, 320, 640, 32) 2592 concatenate_3[0][0]
__________________________________________________________________________________________________
activation_18 (Activation) (None, 320, 640, 32) 0 batch_normalization_26[0][0]
__________________________________________________________________________________________________
batch_normalization_27 (BatchNo (None, 320, 640, 32) 128 conv2d_28[0][0]
__________________________________________________________________________________________________
conv2d_27 (Conv2D) (None, 320, 640, 32) 9248 activation_18[0][0]
__________________________________________________________________________________________________
add_8 (Add) (None, 320, 640, 32) 0 batch_normalization_27[0][0]
conv2d_27[0][0]
__________________________________________________________________________________________________
conv2d_29 (Conv2D) (None, 320, 640, 1) 33 add_8[0][0]
==================================================================================================
Total params: 4,723,057
Trainable params: 4,715,761
Non-trainable params: 7,296
__________________________________________________________________________________________________
4.Loss & Compile
smooth = 1.
def dice_coef(y_true, y_pred):
y_true_f = K.flatten(y_true)
y_pred_f = K.flatten(y_pred)
intersection = tf.reduce_sum(y_true_f * y_pred_f)
return (2. * intersection + smooth) / (tf.reduce_sum(y_true_f) + tf.reduce_sum(y_pred_f) + smooth)
def dice_coef_loss(y_true, y_pred):
return 1.0 - dice_coef(y_true, y_pred)
def IOU(y_true, y_pred):
y_true = K.flatten(y_true)
y_pred = K.flatten(y_pred)
thresh = 0.5
y_true = K.cast(K.greater_equal(y_true, thresh), 'float32')
y_pred = K.cast(K.greater_equal(y_pred, thresh), 'float32')
union = K.sum(K.maximum(y_true, y_pred)) + K.epsilon()
intersection = K.sum(K.minimum(y_true, y_pred)) + K.epsilon()
iou = intersection/union
return iou
def lr_schedule(epoch):
lr =0.0035
if epoch >150:
lr *=2**-1
elif epoch >80:
lr *=2**(-1)
elif epoch >50:
lr *=2**(-1)
elif epoch >30:
lr *=2**(-1)
print('Learning rate: ', lr)
return lr
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.callbacks import LearningRateScheduler
from keras.optimizers import SGD
import time
start_time = time.time()
# Prepare callbacks for model saving and for learning rate adjustment.
lr_scheduler = LearningRateScheduler(lr_schedule)
lr_reducer = ReduceLROnPlateau(factor=np.sqrt(0.1),
cooldown=0,
patience=5,
min_lr=0.5e-6)
callbacks = [lr_reducer, lr_scheduler]
import tensorflow as tf
optimiser=tf.keras.optimizers.Adam(
learning_rate=lr_schedule(0),
beta_1=0.9,
beta_2=0.999,
epsilon=1e-07,
amsgrad=True,
name="Adam"
)
model.compile(optimizer =optimiser , loss = dice_coef_loss, metrics = ['accuracy', IOU, dice_coef])
#结果
Learning rate: 0.0035
5.Training
train_steps = len(train_image_files)//batch_size
valid_steps = len(valid_image_files)//batch_size
history = model.fit(
tg, #基于上面的训练集
steps_per_epoch=train_steps,
initial_epoch = 0,
epochs=3, #这里只设为3个
validation_data = vg, #基于上面的验证集
validation_steps = valid_steps,callbacks=callbacks)
WARNING:tensorflow:sample_weight modes were coerced from
...
to
['...']
WARNING:tensorflow:sample_weight modes were coerced from
...
to
['...']
Train for 474 steps, validate for 87 steps
Learning rate: 0.0035
Epoch 1/3
474/474 [==============================] - 10471s 22s/step - loss: 0.4093 - accuracy: 0.9513 - IOU: 0.4435 - dice_coef: 0.5907 - val_loss: 0.4153 - val_accuracy: 0.9350 - val_IOU: 0.4442 - val_dice_coef: 0.5847
Learning rate: 0.0035
Epoch 2/3
474/474 [==============================] - 5823s 12s/step - loss: 0.3736 - accuracy: 0.9561 - IOU: 0.4787 - dice_coef: 0.6264 - val_loss: 0.4357 - val_accuracy: 0.8781 - val_IOU: 0.4241 - val_dice_coef: 0.5643
Learning rate: 0.0035
Epoch 3/3
474/474 [==============================] - 6416s 14s/step - loss: 0.3337 - accuracy: 0.9618 - IOU: 0.5186 - dice_coef: 0.6663 - val_loss: 0.4244 - val_accuracy: 0.9004 - val_IOU: 0.4297 - val_dice_coef: 0.5756
train_loss = history.history['loss']
valid_loss = history.history['val_loss']
train_acc = history.history['accuracy']
valid_acc = history.history['val_accuracy']
fig, axes = plt.subplots(1, 2, figsize=(13,4))
axes = axes.flatten()
axes[0].plot(train_acc, label='training')
axes[0].plot(valid_acc, label='validation')
axes[0].set_title('Accuracy Curve')
axes[0].set_xlabel('epochs')
axes[0].set_ylabel('accuracy')
axes[0].legend()
axes[1].plot(train_loss, label='training')
axes[1].plot(valid_loss, label='validation')
axes[1].set_title('Loss Curve')
axes[1].set_xlabel('epochs')
axes[1].set_ylabel('loss')
axes[1].legend()
plt.show()
train_dice = history.history['dice_coef']
valid_dice = history.history['val_dice_coef']
train_IOU = history.history['IOU']
valid_IOU = history.history['val_IOU']
fig, axes = plt.subplots(1, 2, figsize=(20,7))
axes = axes.flatten()
axes[0].plot(train_IOU, label='training')
axes[0].plot(valid_IOU, label='validation')
axes[0].set_title('IOU Curve [Adam lr : 0.0001]')
axes[0].set_xlabel('epochs')
axes[0].set_ylabel('IOU')
axes[0].legend()
axes[1].plot(train_dice, label='training')
axes[1].plot(valid_dice, label='validation')
axes[1].set_title('Dice coefficient Curve [Adam lr : 0.0001]')
axes[1].set_xlabel('epochs')
axes[1].set_ylabel('dice_coef')
axes[1].legend()
plt.show()
6.Testing
test_generator = Generator(valid_image_files, valid_mask_files, 396, img_dim)
for x_test, y_test in test_generator:
break
y_pred = model.predict(x_test)
yy_true = (y_test>0.5).flatten()
yy_pred = (y_pred>0.5).flatten()
report = classification_report(yy_true, yy_pred, output_dict=True)
Accuracy = accuracy_score(yy_true, yy_pred)
Precision = report['True']['precision']
Recall = report['True']['recall']
F1_score = report['True']['f1-score']
Sensitivity = Recall
Specificity = report['False']['recall']
AUC = roc_auc_score(y_test.flatten(), y_pred.flatten())
IOU = (Precision*Recall)/(Precision+Recall-Precision*Recall)
print("Accuracy: {0:.4f}\n".format(Accuracy))
print("Precision: {0:.4f}\n".format(Precision))
print("Recall: {0:.4f}\n".format(Recall))
print("F1-Score: {0:.4f}\n".format(F1_score))
print("Sensitivity: {0:.4f}\n".format(Sensitivity))
print("Specificity: {0:.4f}\n".format(Specificity))
print("AUC: {0:.4f}\n".format(AUC))
print("IOU: {0:.4f}\n".format(IOU))
print('-'*50,'\n')
print(classification_report(yy_true, yy_pred))
#结果
Accuracy: 0.9004
Precision: 0.3555
Recall: 0.8970
F1-Score: 0.5092
Sensitivity: 0.8970
Specificity: 0.9006
AUC: 0.9202
IOU: 0.3415
--------------------------------------------------
precision recall f1-score support
False 0.99 0.90 0.94 67167304
True 0.36 0.90 0.51 4103096
accuracy 0.90 71270400
macro avg 0.67 0.90 0.73 71270400
weighted avg 0.96 0.90 0.92 71270400
for i, j in test1_generator:
break
print(i.shape)
print(j.shape)
#结果
(4, 320, 640, 3)
(4, 320, 640, 1)
ttg = Generator(test_image_paths,test_mask_paths, batch_size, img_dim, augment = False)
for i, j in ttg:
break
print(i.shape)
print(j.shape)
#结果
(4, 320, 640, 3)
(4, 320, 640, 1)
test_generator1 = Generator(test_image_paths,test_mask_paths,1124, img_dim)
for x_test, y_test in test_generator:
break
y_pred = model.predict(x_test)
yy_true = (y_test>0.5).flatten()
yy_pred = (y_pred>0.5).flatten()
report = classification_report(yy_true, yy_pred, output_dict=True)
Accuracy = accuracy_score(yy_true, yy_pred)
Precision = report['True']['precision']
Recall = report['True']['recall']
F1_score = report['True']['f1-score']
Sensitivity = Recall
Specificity = report['False']['recall']
AUC = roc_auc_score(y_test.flatten(), y_pred.flatten())
IOU = (Precision*Recall)/(Precision+Recall-Precision*Recall)
print("Accuracy: {0:.4f}\n".format(Accuracy))
print("Precision: {0:.4f}\n".format(Precision))
print("Recall: {0:.4f}\n".format(Recall))
print("F1-Score: {0:.4f}\n".format(F1_score))
print("Sensitivity: {0:.4f}\n".format(Sensitivity))
print("Specificity: {0:.4f}\n".format(Specificity))
print("AUC: {0:.4f}\n".format(AUC))
print("IOU: {0:.4f}\n".format(IOU))
print('-'*50,'\n')
print(classification_report(yy_true, yy_pred))
#结果
Accuracy: 0.9004
Precision: 0.3555
Recall: 0.8970
F1-Score: 0.5092
Sensitivity: 0.8970
Specificity: 0.9006
AUC: 0.9202
IOU: 0.3415
--------------------------------------------------
precision recall f1-score support
False 0.99 0.90 0.94 67167304
True 0.36 0.90 0.51 4103096
accuracy 0.90 71270400
macro avg 0.67 0.90 0.73 71270400
weighted avg 0.96 0.90 0.92 71270400
参考资料：https://github.com/Subham2901/Concrete_Crack_Segmentation 欢迎关注公众号【智能建造小硕】（分享计算机编程、人工智能、智能建造、日常学习、科研和写作经验等，欢迎大家关注交流。）
参考阅读
您阅读本篇文章共花了：
发表评论