问题描述
我正在尝试使用import cv2,os
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
#from keras.backend.tensorflow_backend import set_session
import keras,sys,time,warnings
from keras.models import *
from keras.layers import *
import pandas as pd
dir_data = "dataset1/"
dir_seg = dir_data + "/annotations_prepped_train/"
dir_img = dir_data + "/images_prepped_train/"
import cv2,os
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
## seaborn has white grid by default so I will get rid of this.
sns.set_style("whitegrid",{'axes.grid' : False})
import random
def give_color_to_seg_img(seg,n_classes):
'''
seg : (input_width,input_height,3)
'''
if len(seg.shape)==3:
seg = seg[:,:,0]
seg_img = np.zeros( (seg.shape[0],seg.shape[1],3) ).astype('float')
colors = sns.color_palette("hls",n_classes)
for c in range(n_classes):
segc = (seg == c)
seg_img[:,0] += (segc*( colors[c][0] ))
seg_img[:,1] += (segc*( colors[c][1] ))
seg_img[:,2] += (segc*( colors[c][2] ))
return(seg_img)
input_height,input_width = 224,224
output_height,output_width = 224,224
n_classes=20
def getimageArr( path,width,height ):
img = cv2.imread(path,1)
img = np.float32(cv2.resize(img,( width,height ))) / 127.5 - 1
return img
def getSegmentationArr( path,nClasses,height ):
seg_labels = np.zeros(( height,nClasses ))
img = cv2.imread(path,1)
img = cv2.resize(img,height ))
img = img[:,0]
for c in range(nClasses):
seg_labels[:,c ] = (img == c ).astype(int)
##seg_labels = np.reshape(seg_labels,( width*height,nClasses ))
return seg_labels
images = os.listdir(dir_img)
images.sort()
segmentations = os.listdir(dir_seg)
segmentations.sort()
X = []
Y = []
for im,seg in zip(images,segmentations) :
X.append( getimageArr(dir_img + im,input_width,input_height ) )
Y.append( getSegmentationArr( dir_seg + seg,n_classes,output_width,output_height ) )
X,Y = np.array(X),np.array(Y)
print(X.shape,Y.shape)
VGG_Weights_path = "C:/Users/umer Zukaib/Documents/python works/segmentation/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5"
def FCN8(nClasses,input_height=224,input_width=224):
## input_height and width must be devisible by 32 because maxpooling with filter size = (2,2) is operated 5 times,## which makes the input_height and width 2^5 = 32 times smaller
assert input_height%32 == 0
assert input_width%32 == 0
IMAGE_ORDERING = "channels_last"
img_input = Input(shape=(input_height,3)) ## Assume 224,224,3
## Block 1
x = Conv2D(64,(3,3),activation='relu',padding='same',name='block1_conv1',data_format=IMAGE_ORDERING )(img_input)
x = Conv2D(64,name='block1_conv2',data_format=IMAGE_ORDERING )(x)
x = MaxPooling2D((2,2),strides=(2,name='block1_pool',data_format=IMAGE_ORDERING )(x)
f1 = x
# Block 2
x = Conv2D(128,name='block2_conv1',data_format=IMAGE_ORDERING )(x)
x = Conv2D(128,name='block2_conv2',name='block2_pool',data_format=IMAGE_ORDERING )(x)
f2 = x
# Block 3
x = Conv2D(256,name='block3_conv1',data_format=IMAGE_ORDERING )(x)
x = Conv2D(256,name='block3_conv2',name='block3_conv3',name='block3_pool',data_format=IMAGE_ORDERING )(x)
pool3 = x
# Block 4
x = Conv2D(512,name='block4_conv1',data_format=IMAGE_ORDERING )(x)
x = Conv2D(512,name='block4_conv2',name='block4_conv3',data_format=IMAGE_ORDERING )(x)
pool4 = MaxPooling2D((2,name='block4_pool',data_format=IMAGE_ORDERING )(x)## (None,14,512)
# Block 5
x = Conv2D(512,name='block5_conv1',data_format=IMAGE_ORDERING )(pool4)
x = Conv2D(512,name='block5_conv2',name='block5_conv3',data_format=IMAGE_ORDERING )(x)
pool5 = MaxPooling2D((2,name='block5_pool',7,512)
#x = Flatten(name='flatten')(x)
#x = Dense(4096,name='fc1')(x)
# <--> o = ( Conv2D( 4096,( 7,7 ),data_format=IMAGE_ORDERING))(o)
# assuming that the input_height = input_width = 224 as in VGG data
#x = Dense(4096,name='fc2')(x)
# <--> o = ( Conv2D( 4096,( 1,1 ),data_format=IMAGE_ORDERING))(o)
# assuming that the input_height = input_width = 224 as in VGG data
#x = Dense(1000,activation='softmax',name='predictions')(x)
# <--> o = ( Conv2D( nClasses,kernel_initializer='he_normal',data_format=IMAGE_ORDERING))(o)
# assuming that the input_height = input_width = 224 as in VGG data
vgg = Model( img_input,pool5 )
vgg.load_weights(VGG_Weights_path) ## loading VGG weights for the encoder parts of FCN8
n = 4096
o = ( Conv2D( n,name="conv6",data_format=IMAGE_ORDERING))(pool5)
conv7 = ( Conv2D( n,name="conv7",data_format=IMAGE_ORDERING))(o)
## 4 times upsamping for pool4 layer
conv7_4 = Conv2DTranspose( nClasses,kernel_size=(4,4),strides=(4,use_bias=False,data_format=IMAGE_ORDERING )(conv7)
## (None,10)
## 2 times upsampling for pool411
pool411 = ( Conv2D( nClasses,name="pool4_11",data_format=IMAGE_ORDERING))(pool4)
pool411_2 = (Conv2DTranspose( nClasses,kernel_size=(2,data_format=IMAGE_ORDERING ))(pool411)
pool311 = ( Conv2D( nClasses,name="pool3_11",data_format=IMAGE_ORDERING))(pool3)
o = Add(name="add")([pool411_2,pool311,conv7_4 ])
o = Conv2DTranspose( nClasses,kernel_size=(8,8),strides=(8,data_format=IMAGE_ORDERING )(o)
#o = (Activation('softmax'))(o)
print('the value of O=',o)
new_s = o[None,10:,10,:]
print('the value of new_s=',new_s.values())
model = Model(img_input,o)
return model
model = FCN8(nClasses = n_classes,input_height = 224,input_width = 224)
model.summary()
from sklearn.utils import shuffle
train_rate = 0.85
index_train = np.random.choice(X.shape[0],int(X.shape[0]*train_rate),replace=False)
index_test = list(set(range(X.shape[0])) - set(index_train))
X,Y = shuffle(X,Y)
X_train,y_train = X[index_train],Y[index_train]
X_test,y_test = X[index_test],Y[index_test]
print(X_train.shape,y_train.shape)
print(X_test.shape,y_test.shape)
from keras import optimizers
sgd = optimizers.SGD(lr=1E-2,decay=5**(-4),momentum=0.9,nesterov=True)
#loss='mean_squared_error',optimizer='adam'
model.compile(loss='mean_squared_error',optimizer=sgd)
hist1 = model.fit(X_train,y_train,validation_data=(X_test,y_test),batch_size=32,epochs=1,verbose=2)
for key in ['loss','val_loss']:
plt.plot(hist1.history[key],label=key)
plt.legend()
plt.show()
y_pred = model.predict(X_test)
y_predi = np.argmax(y_pred,axis=3)
y_testi = np.argmax(y_test,axis=3)
print(y_testi.shape,y_predi.shape)
def IoU(Yi,y_predi):
## mean Intersection over Union
## Mean IoU = TP/(FN + TP + FP)
IoUs = []
Nclass = int(np.max(Yi)) + 1
for c in range(Nclass):
TP = np.sum( (Yi == c)&(y_predi==c) )
FP = np.sum( (Yi != c)&(y_predi==c) )
FN = np.sum( (Yi == c)&(y_predi != c))
IoU = TP/float(TP + FP + FN)
print("class {:02.0f}: #TP={:6.0f},#FP={:6.0f},#FN={:5.0f},IoU={:4.3f}".format(c,TP,FP,FN,IoU))
IoUs.append(IoU)
mIoU = np.mean(IoUs)
print("_________________")
print("Mean IoU: {:4.3f}".format(mIoU))
IoU(y_testi,y_predi)
将UIView转换为这种形状，目前我只能做左下角的操作，寻求添加其他角的帮助。
仅用于左下角的代码。
UIBezierPath我做了几次尝试添加其他角的尝试，但是我的 let mask = CAShapeLayer()
mask.frame = self.innerLayout.layer.bounds
let path = UIBezierPath()
let radius: CGFloat = 50
let rect = mask.bounds
path.move(to: rect.origin)
path.addLine(to: CGPoint(x: rect.maxX,y: rect.minY))
path.addLine(to: CGPoint(x: rect.maxX,y: rect.maxY))
path.addLine(to: CGPoint(x: rect.minX + radius,y: rect.maxY))
path.addArc(withCenter: CGPoint(x: rect.minX,y: rect.maxY),radius: radius,startAngle: 0,endAngle: CGFloat(M_PI_2 * 3),clockwise: false)
mask.path = path.cgPath
self.innerLayout.layer.mask = mask
的形状很有趣。我只是通过复制和粘贴（以及更改原点）来添加它，我相信我们将使用这部分代码4次以添加4个角
UIView解决方法
这是要在Playground中运行的完整示例。这样的几何形状总是棘手的。弄错坐标或使弧线走错方向非常容易。
import pandas as pd
df = pd.DataFrame({'foo':[1,2,3],'bar':[4,5,6]})
def my_function(x):
#print(x)
# some computation
# returns multiple values (tuple)
# simplified here
return 1,1,1
df = df[df.foo > 10]
df['r1'],df['r2'],df['r3'] = zip(*df.apply(my_function,axis=1))
df