from nolearn.lasagne import NeuralNet
from nolearn.lasagne import visualize
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import sklearn
from lasagne.layers import InputLayer, Conv2DLayer, DenseLayer, MaxPool2DLayer, InverseLayer
### DATASET
dataSet = myTools.loadImages('../../images', 1024, 1024, 4)
dataSet = myTools.oneDimension(dataSet)
dataSet = dataSet.astype(numpy.uint8)
dataSet = myTools.cropCenter(dataSet, 80)
dataSet = myTools.augmentData(dataSet,
numOfTiles=4,
overlap=False,
imageWidth=819,
imageHeight=819)
dataSet = dataSet.astype(numpy.float32)
#plt.show(plt.imshow(dataSet[0][0], cmap=cm.binary))
### MASKS
masks = myTools.loadImages('../../masks', 819, 819, 1)
for x in numpy.nditer(masks, op_flags=['readwrite']):
if x > 0:
#open the image
image = Image.open(imagePath)
#image as numpy array
image = numpy.asarray(image)
#keep only one dimension
image = image[:, :, 0]
#most needed type casting
image = image.astype(numpy.uint8)
#crop the center
image = myTools.cropCenter1(image, 100)
#this step is mysteriously needed
image = myTools.augmentMasks(image.reshape(1, 1, image.shape[0],
image.shape[1]),
numOfTiles=1,
overlap=False,
imageWidth=image.shape[0],
imageHeight=image.shape[1])
#another vital type casting
image = image.astype(numpy.float32)
#setting parameters for the network
data_size = (None, 1, image[0][0].shape[0], image[0][0].shape[1])
#load the pretrained network
myNet = myTools.createPretrainedNN(data_size)
#make predictions for the image
res = myNet(image)
#crop the center of the mask
res = myTools.cropCenter(res, 80)
plt.show(plt.imshow(res[0][0], cmap=cm.binary))
scipy.misc.imsave(outName, res[0][0])
image=numpy.asarray(image) #keep only one dimension image=image[:,:,0] #most needed type casting image=image.astype(numpy.uint8) #crop the center image=myTools.cropCenter1(image, 83.1) #split image to 4 splits=myTools.augmentData(image.reshape(1,1,image.shape[0],image.shape[1]), numOfTiles=4, overlap=False, imageWidth=image.shape[0], imageHeight=image.shape[1]) #another vital type casting splits=splits.astype(numpy.float32) #keep only the 4 original tiles splits=splits[0:4,:,:,:] #setting parameters for the network data_size=(None,1,splits[0][0].shape[0],splits[0][0].shape[1]) #load the pretrained network myNet=myTools.createPretrainedNN(data_size) #make predictions for the 4 tiles print(splits.dtype) res=myNet(splits) #crop the center of the predictions res=myTools.cropCenter(res, 93) #concatenate on the x axis top=np.concatenate((res[0][0],res[2][0]),axis=1) bot=np.concatenate((res[1][0],res[3][0]),axis=1) #concatenate the two halves to get the full image res=np.concatenate((top,bot),axis=0) plt.show(plt.imshow(res, cmap=cm.binary)) scipy.misc.imsave(outName, res)
#this step is mysteriously needed image=myTools.augmentData(image.reshape(1,1,image.shape[0],image.shape[1]), numOfTiles=1, overlap=False, imageWidth=image.shape[0], imageHeight=image.shape[1]) #another vital type casting image=image.astype(numpy.float32) #setting parameters for the network data_size=(None,1,image[0][0].shape[0],image[0][0].shape[1]) imageMean=numpy.mean(image, keepdims=True) image=image-imageMean #load the pretrained network myNet1=myTools.createPretrainedNN2(data_size, modelFile='myModel102Plus90.npz', filters=32) #make predictions for the image res1=myNet1(image) #crop the center of the mask res1=myTools.cropCenter(res1, 80) #load the pretrained network myNet2=myTools.createPretrainedNN2(data_size, modelFile='myModel103Plus90.npz', filters=32) #make predictions for the image res2=myNet2(image) #crop the center of the mask res2=myTools.cropCenter(res2, 80) #load the pretrained network myNet3=myTools.createPretrainedNN2(data_size, modelFile='myModel104Plus90.npz', filters=32) #make predictions for the image res3=myNet3(image) #crop the center of the mask res3=myTools.cropCenter(res3, 80) '''
#another vital type casting
image = image.astype(numpy.float32)
#setting parameters for the network
data_size = (None, 1, image[0][0].shape[0], image[0][0].shape[1])
imageMean = numpy.mean(image, keepdims=True)
image = image - imageMean
#load the pretrained network
myNet1 = myTools.createPretrainedNN2(data_size,
modelFile='myModel102Plus90.npz',
filters=32)
#make predictions for the image
res1 = myNet1(image)
#crop the center of the mask
res1 = myTools.cropCenter(res1, 80)
#load the pretrained network
myNet2 = myTools.createPretrainedNN2(data_size,
modelFile='myModel103Plus90.npz',
filters=32)
#make predictions for the image
res2 = myNet2(image)
#crop the center of the mask
res2 = myTools.cropCenter(res2, 80)
#load the pretrained network
myNet3 = myTools.createPretrainedNN2(data_size,
modelFile='myModel104Plus90.npz',
filters=32)
#make predictions for the image
image=image.astype(numpy.uint8) #crop the center #image=myTools.cropCenter1(image, 100) #this step is mysteriously needed image=myTools.augmentMasks(image.reshape(1,1,image.shape[0],image.shape[1]), numOfTiles=1, overlap=False, imageWidth=image.shape[0], imageHeight=image.shape[1]) #another vital type casting image=image.astype(numpy.float32) #setting parameters for the network data_size=(None,1,image[0][0].shape[0],image[0][0].shape[1]) #load the pretrained network myNet1=myTools.createPretrainedNN2(data_size, modelFile='model020New.npz', filters=64) #make predictions for the image res1=myNet1(image) #crop the center of the mask res1=myTools.cropCenter(res1, 80) #load the pretrained network myNet2=myTools.createPretrainedNN2(data_size, modelFile='model20NewsallLR688images.npz', filters=64) #make predictions for the image res2=myNet2(image) #crop the center of the mask res2=myTools.cropCenter(res2, 80) #load the pretrained network myNet3=myTools.createPretrainedNN2(data_size, modelFile='model20NewSmallLR.npz', filters=64) #make predictions for the image res3=myNet3(image) #crop the center of the mask res3=myTools.cropCenter(res3, 80)
from nolearn.lasagne import visualize
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import sklearn
from lasagne.layers import InputLayer, Conv2DLayer, DenseLayer, MaxPool2DLayer, InverseLayer
### DATASET
dataSet=myTools.loadImages('../../images', 1024, 1024, 4)
dataSet=myTools.oneDimension(dataSet)
dataSet=dataSet.astype(numpy.uint8)
dataSet=myTools.cropCenter(dataSet, 80)
dataSet=myTools.augmentData(dataSet, numOfTiles=4, overlap=False, imageWidth=819, imageHeight=819)
dataSet=dataSet.astype(numpy.float32)
#plt.show(plt.imshow(dataSet[0][0], cmap=cm.binary))
### MASKS
masks=myTools.loadImages('../../masks', 819, 819, 1)
for x in numpy.nditer(masks, op_flags=['readwrite']):
if x>0:
x[...]=1
masks=masks.astype(numpy.float32)
overlap=False,
imageWidth=image.shape[0],
imageHeight=image.shape[1])
#another vital type casting
image = image.astype(numpy.float32)
#setting parameters for the network
data_size = (None, 1, image[0][0].shape[0], image[0][0].shape[1])
#load the pretrained network
myNet1 = myTools.createPretrainedNN2(data_size,
modelFile='model020New.npz',
filters=64)
#make predictions for the image
res1 = myNet1(image)
#crop the center of the mask
res1 = myTools.cropCenter(res1, 80)
#load the pretrained network
myNet2 = myTools.createPretrainedNN2(data_size,
modelFile='model20NewsallLR688images.npz',
filters=64)
#make predictions for the image
res2 = myNet2(image)
#crop the center of the mask
res2 = myTools.cropCenter(res2, 80)
#load the pretrained network
myNet3 = myTools.createPretrainedNN2(data_size,
modelFile='model20NewSmallLR.npz',
filters=64)
#make predictions for the image
image=numpy.asarray(image) #keep only one dimension image=image[:,:,0] #most needed type casting image=image.astype(numpy.uint8) #crop the center #image=myTools.cropCenter1(image, 100) #this step is mysteriously needed image=myTools.augmentData(image.reshape(1,1,image.shape[0],image.shape[1]), numOfTiles=1, overlap=False, imageWidth=image.shape[0], imageHeight=image.shape[1]) #another vital type casting image=image.astype(numpy.float32) imageMean=numpy.mean(image, keepdims=True) image=image-imageMean #setting parameters for the network data_size=(None,1,image[0][0].shape[0],image[0][0].shape[1]) #load the pretrained network myNet=myTools.createPretrainedNN2(data_size, modelFile=modelFilename, filters=numberOfFilters) #make predictions for the image res=myNet(image) #crop the center of the mask res=myTools.cropCenter(res, 83.1) plt.show(plt.imshow(res[0][0], cmap=cm.binary)) scipy.misc.imsave(outName, res[0][0])
argEpochs=int(sys.argv[3])
print('Learning rate: %f' % (argLR))
print('Weight decay: %f' % (argWD))
print('Number of epochs: %d' % (argEpochs))
random.seed(123)
### DATASET
dataSet=myTools.loadImages('../../images', 1024, 1024, 4)
dataSet=myTools.oneDimension(dataSet)
dataSet=dataSet.astype(numpy.uint8)
dataSet=myTools.cropCenter(dataSet, 83.1)#81.2 83.1
dataSet=myTools.augmentData(dataSet, numOfTiles=1, overlap=False, imageWidth=850, imageHeight=850)#830 850
dataSet=dataSet.astype(numpy.float32)
imageMeans=numpy.mean(dataSet, axis=(1,2,3), keepdims=True)
dataSet=dataSet-imageMeans
### MASKS
masks=myTools.loadImages('../../masksExpertBig', 850, 850, 1)
