def batchProcessor(self, pathsX, Y):
X = []
for filename in pathsX:
img = ImageOps.readImage(os.path.join(os.path.join(self.__path, self.__imageDir), filename))
img = ImageOps.resizeImage(img, self.imageSize, self.imageSize)
img = img / 255
X.append(img)
return np.array(X), np.array(Y)
def imageOps(self):
if self.isLoaded:
dialog = ImageOps(self.image)
if dialog.isChanged():
self.image = dialog.getAfterQImage()
self.pixmap = QPixmap.fromImage(self.image)
self.internalPixmap.setPixmap(self.pixmap)
self.scene.update()
self.graphicsView.update()
QApplication.processEvents()
def batchProcessor(self, paths):
X = []
Y = []
for path in paths:
temp = []
path = os.path.join(path, 'cat_')
for i in range(5):
temp.append(ImageOps.readImage(path + str(i) + '.jpg') / 255)
Y.append(ImageOps.readImage(path + 'result.jpg') / 255)
X.append(np.array(temp))
return np.array(X), np.array(Y)
def batchProcessor(self, filenames):
X = []
Y = []
for filename in filenames:
img = ImageOps.readImage(os.path.join(self.imagesDir, filename + '.jpg'))
img = ImageOps.resizeImage(img, self.imageSize, self.imageSize)
if len(img.shape) == 2:
img = np.pad(img.reshape((self.imageSize, self.imageSize, 1)), [(0, 0), (0, 0), (0, 2)],
mode='constant', constant_values=0)
X.append(img)
Y.append(h5py.File(os.path.join(self.annotationsDir, filename + '.mat'), 'r'))
return np.array(X), np.array(Y)
def batchProcessor(self, pathX, Y):
resultX = []
resultY = []
for i in range(len(pathX)):
try:
x = ImageOps.readImage(pathX[i])
x = ImageOps.resizeImageWithAspectRatio(x, self.imageSize)
x = np.reshape(x, (-1))
if x is None or len(x) != self.imageSize * self.imageSize * 3:
raise Exception
resultX.append(np.array(x)/255.0)
resultY.append(Y[i])
except:
print("file removed:", pathX[i])
#os.remove(pathX[i])
return np.concatenate(resultX, axis=0).reshape((-1, self.imageSize, self.imageSize, 3)), np.array(
resultY).reshape((-1))
import sys
from Model import Vgg16
from Trainer import Trainer
import numpy as np
from ImageOps import ImageOps
imageOps = ImageOps()
batchsize = 32
imageSize = 64
model = Vgg16()
model.noOfClasses = 2
model.hiddenUnits = 2048
model.imageShape = [imageSize, imageSize, 3]
model.addMetrics('Accuracy')
model.convActivation = "relu"
model.fcActivation = "leakyrelu"
model.regularizationCoefficient = 0.005
model.regularizationType = "l2"
model.build()
trainer = Trainer(model)
trainer.batchSize = 32
trainer.workingDir = "vggDogCat"
img = imageOps.readImage(["test1.jpg", "test2.jpg"])
img = imageOps.resizeImageWithAspectRatio(img, imageSize)
img = np.array(img).reshape([-1, imageSize, imageSize, 3])
print(img.shape)
test = {"x": img}
import sys
sys.path.insert(0, '../')
from DataQueue import DataQueue
from Model import Vgg16
from Trainer import Trainer
import numpy as np
from Datasets import ClassificationImageGenerator
from ImageOps import ImageOps
imageOps = ImageOps()
batchsize = 32
imageSize = 64
trainGen = ClassificationImageGenerator(
'/media/batman/ent/datasets/kagglecatsanddogs_3367a/images',
batchsize,
imageSize=64)
testGen = ClassificationImageGenerator(
'/media/batman/ent/datasets/kagglecatsanddogs_3367a/PetImages',
batchsize,
imageSize=64,
batchType="sequential")
testGen.init()
test = {'x': [], 'y': []}
for _ in range(20):
x, y = testGen.batchProcessor(*tuple(testGen.batchGenerator()))
test['x'].append(np.array(x))
test['y'].append(np.array(y))
test['x'] = np.concatenate(test['x'], axis=0)