def stack(folderPath,sampleNumber,sobelLevels,brushMasks,superPixMethod='combined',features='combined',triGrown=True,sobelType='combined'):
FOLDER_PATH = folderPath
SAMPLE_NUMBER = sampleNumber
#trainRatio = int(opts['<trainRatio>'])
#path = './palstaves2/2013T482_Lower_Hardres_Canterbury/Axe1/'
levels = sobelLevels
path = FOLDER_PATH
outputFilename = os.path.join(os.path.dirname(path),'trainingData_'+str(sobelLevels)+'_'+\
'brush'+str(brushMasks)+'_'+str(superPixMethod)+'_'+str(features)+'_'+'grown'+str(triGrown)+'.npz')
if features=='RGB':
wholeXArray = np.zeros([0,3])#([0,8+3])#np.zeros([0,levels*3+3])#+3 is for RGB non sobelised
elif features=='sobel':
wholeXArray = np.zeros([0,levels*3+3])
elif features=='entropy':
wholeXArray = np.zeros([0,7+3])
elif features=='combinedEntSob':
wholeXArray = np.zeros([0,levels*3+7+3])
elif features =='dwt':
wholeXArray = np.zeros([0,8+3])
#wholeXArray = np.zeros([0,8+3])
elif features=='combinedDwtSob':
wholeXArray = np.zeros([0,levels*3+8+3])
#wholeXArray = np.zeros([0,levels*3+3])
elif features =='sobelSansRGB':
wholeXArray = np.zeros([0,levels*3])
elif features =='sobelHandv':
wholeXArray = np.zeros([0,levels*6+3])
else:
print ("Error selecting type of features")
l=lp#breakpoint
wholeyArray = np.zeros([0])
numberStacked = 0
numberSuccessStacked = 0
imageSetSize = 0
trainRatio = 10
get_ipython().magic('run testing_sobel')
get_ipython().magic('run sobelise')
#a=np.load('./palAxe1arrays/JAIMG_3517.npz')
#a=np.asarray(a)
#print(a.shape)
for filepath in glob.glob(os.path.join(path, '*.jpg')):
imageSetSize += 1
trainSetSize = int(imageSetSize/trainRatio+1)
print('Image set size = '+str(imageSetSize))
print('Training set size = '+str(trainSetSize))
print('Sampling rate = '+str(SAMPLE_NUMBER))
imageNames = glob.glob(os.path.join(path, '*.jpg'))
#random.shuffle(imageNames)
imageNameHolder = []
counter = 0
for filepath in imageNames:
#print('in')
if counter % 1 == 0:
if numberSuccessStacked >= trainSetSize or numberStacked == imageSetSize:
#used to be break
a=1
numberStacked += 1
numberSuccessStacked += 1
#print('blah')
fileNameStringWithExtension = os.path.basename(filepath)
fileNameString = os.path.splitext(fileNameStringWithExtension)[0]
if brushMasks == False:
maskPath = os.path.join(os.path.dirname(folderPath), 'trainMasks/'+fileNameString+'_mask')##normally 'masks/'
else:
maskPath = os.path.join(os.path.dirname(folderPath), 'brushMasks/'+fileNameString+'_mask')##normally 'masks/'
# loading 1/4 sized images
#all levels concatenated together
#print(maskPath)
try:
maskRaw = Image.open(maskPath+'.jpg').convert(mode='L')#convert is new
imageNameHolder.append(fileNameString)
except IOError:
print('Image '+fileNameString+' has no corresponding mask, it has been skipped')
numberSuccessStacked -= 1
continue
alreadyDone = sobelise.process_image(filepath,levels,features)
totalSob = testing_sobel.concatSob(filepath,levels,features)
im = Image.open(filepath)
im = np.asarray(im)
#im = ndimage.gaussian_filter(im, 3)
'''
sx0 = ndimage.sobel(im[...,...,0], axis=0, mode='constant')
sy0 = ndimage.sobel(im[...,...,0], axis=1, mode='constant')
#sob0 = np.hypot(sx0, sy0)
sx1 = ndimage.sobel(im[...,...,1], axis=0, mode='constant')
sy1 = ndimage.sobel(im[...,...,1], axis=1, mode='constant')
#sob1 = np.hypot(sx1, sy1)
sx2 = ndimage.sobel(im[...,...,2], axis=0, mode='constant')
sy2 = ndimage.sobel(im[...,...,2], axis=1, mode='constant')
#sob2 = np.hypot(sx2, sy2)
sobx = np.dstack([sx0,sx1,sx2])
soby = np.dstack([sy0,sy1,sy2])
sobx_blurred0 = ndimage.gaussian_filter(sobx, 8)
soby_blurred0 = ndimage.gaussian_filter(soby, 8)
#sob_blurred1 = ndimage.gaussian_filter(sob1_3D, 8)
#sob_blurred2 = ndimage.gaussian_filter(sob2_3D, 8)
#sob_blurred = sob_blurred0+sob_blurred1+sob_blurred2
#sob_blurred2 = ndimage.gaussian_filter(sob_blurred, 8)
#sob_blurred3 = ndimage.gaussian_filter(sob_blurred2, 8)
imWithSobBlurred0 = np.dstack([im,sobx,soby,sobx_blurred0,soby_blurred0])
'''
#im = rescale(im,0.25)
im = rescale(im,0.25)#0.125)
if features=='RGB'or features=='entropy'or features=='dwt':
imArray = im*255
elif features=='sobel' or features=='sobelHandv'or features=='combinedEntSob'or features=='combinedDwtSob':
imArray = np.asarray(totalSob)
imArray = np.dstack([imArray,im*255])
elif features =='sobelSansRGB':
imArray = np.asarray(totalSob)
if features =='entropy'or features =='dwt' or features =='combinedDwtSob' or features =='combinedEntSob':
dwtFeature = dwtSlide(filepath,4,features)
flatIm = im.reshape(im.shape[0]*im.shape[1],-1)
#print(np.max(dwtFeature))
print(np.max(im))
#print(np.mean(dwtFeature))
print(np.mean(im))
#l=lp
#l=lp
#imArray = im
maskArray = np.asarray(maskRaw) #not all 255 or 0 because of compression, may need to threshold
## mod the maskArray here
if triGrown == True:
###maskArray = resize(maskArray,[totalSob.shape[0],totalSob.shape[1]])
maskArray = resize(maskArray,[im.shape[0],im.shape[1]])
##maskArray *= 255
a=water_test.watershedFunc2(filepath,superPixMethod,trainingSeg=True)
#print('between')
featureMap=maskArray
classifier = 1
b,totClassified,totMask2,segmentOutlines,totMask=water_test.superPix(im,a,featureMap,classifier,100,alreadyClassified=True,thresh=0.2)
#print(b.shape)
maskArray=totMask.reshape([(totClassified.shape[0]),(totClassified.shape[1])])
else:
maskArray = resize(maskArray,[im.shape[0],im.shape[1]])
###maskArray = ((maskArray+featureMap)/2)
#plt.imshow(b, interpolation='nearest')
#plt.show()
#print(maskArray.shape)
path2 = os.path.dirname(path)
newpath=os.path.join(path2,'postTrainingMasks')
print(newpath)
if not os.path.exists(newpath):
os.makedirs(newpath)
greyRegion=((maskArray > 0.1) & (maskArray < 0.9)).astype(int) #(a > 1) & (a < 5)
print(np.max(greyRegion))
maskImage = Image.fromarray((((maskArray*255).astype(np.uint8))))
maskImage = maskImage.convert('RGB')
print(np.array(maskImage).shape)
#print(np.array(maskImage)[:,:,0])
#print(np.array(maskImage)[:,:,0])
maskImage = np.array(maskImage)
print(np.max(maskImage[...,0]))
print(np.max(maskImage[...,1]))
print(np.max(maskImage[...,2]))
#greyRegion=(maskArray == 0.2*255).astype(int)*1.0
print(np.max(greyRegion))
#l=sf
greyRegionImage = Image.fromarray((((greyRegion*255).astype(np.uint8))))
greyRegionImage = greyRegionImage.convert('RGB')
print( np.sum(greyRegion))
greyRegionImage = np.array(greyRegionImage)
greyRegionImage[:,:,2:3]=0
maskImage[:,:,1:3] = 0
print('hereBrB')
print( np.sum(greyRegion))
print( np.max(maskImage))
print( np.max(greyRegion))
print(greyRegion.shape)
print(type(greyRegion))
#grey255 = greyRegion*255.0
maskImage2 = (greyRegionImage)+maskImage
#52 shows red, 51 doesnt
#superMask2 = (superMask < (a)).astype('int')
print(np.max(maskImage))
maskImage = Image.fromarray((((maskImage).astype(np.uint8))))
print maskImage2.shape
maskImage2 = Image.fromarray((((maskImage2).astype(np.uint8))))
#maskImage = maskImage.convert('RGB')
origImage = Image.fromarray((im*255).astype(np.uint8))
#blend2 = Image.blend(maskImage2,origImage,0.5)
#maskImage2.save(os.path.join(newpath,fileNameString+'_mask2_training.jpg'))
greyImage = Image.fromarray((greyRegion*255).astype(np.uint8))
print im.shape
#greyImage.save(os.path.join(newpath,fileNameString+'_grey_training.jpg'))
blend = Image.blend(maskImage2,origImage,0.5)
blend.save(os.path.join(newpath,fileNameString+'_mask_training.jpg'))
#Image.fromarray((maskArray*255).astype(np.uint8)).save(os.path.join(newpath,fileNameString+'_mask_training.jpg'))
maskArray = maskArray*255
flatMaskArray = maskArray.reshape(maskArray.shape[0]*maskArray.shape[1])
flatImArray = imArray.reshape(imArray.shape[0]*imArray.shape[1],imArray.shape[2])
#flatIm = np.zeros((flatIm.shape[0],flatIm.shape[1])) # for dwt only testing
print flatIm.shape
#print dwtFeature.shape
if features=='combinedEntSob'or features=='combinedDwtSob' or features=='entropy' or features=='dwt':
flatImArray = np.hstack([flatImArray,dwtFeature])
'''
foreGround = (flatMaskArray>=64)
backGround = (flatMaskArray<64)
'''
print( np.max(maskArray))
foreGround = (flatMaskArray>=100)#values depend on the gray used
backGround = (flatMaskArray<10)
foreGroundSamples = flatImArray[foreGround,...]
backGroundSamples = flatImArray[backGround,...]
print(foreGroundSamples.shape[0])
maxSampleCount = min(foreGroundSamples.shape[0],backGroundSamples.shape[0])
outputSampleCount = min(maxSampleCount,int(SAMPLE_NUMBER))
foreGroundIndices = np.random.choice(foreGroundSamples.shape[0],replace=False,size=outputSampleCount)
backGroundIndices = np.random.choice(backGroundSamples.shape[0],replace=False,size=outputSampleCount)
X = np.vstack([foreGroundSamples[foreGroundIndices,...],backGroundSamples[backGroundIndices,...]])
y = np.concatenate([np.ones(outputSampleCount),np.zeros(outputSampleCount)])
#wholeArray = np.array(joinedArray.shape[0],joinedArray.shape[1])
#print(wholeArray.shape)
#a=joinedArray[0:1000000].reshape(4,1000000)
#print(a.shape)
wholeXArray = np.concatenate((wholeXArray,X),axis=0)
wholeyArray = np.concatenate((wholeyArray,y),axis=0)
header = {'images':imageNameHolder}
print('here')
np.savez_compressed(outputFilename,X=wholeXArray,y=wholeyArray,S=int(SAMPLE_NUMBER),R=trainRatio,shuffled=imageNames,header=header)
#print('Stacked image '+fileNameString+ '; number '+str(numberSuccessStacked)+' out of '+str(trainSetSize))
counter += 1
import water_test
import numpy as np
import sobelise
import testing_sobel
from skimage.transform import rescale, resize
sobelise.process_image('C:\Python34\palstaves2\\2013T482_Lower_Hardres_Canterbury\Axe1\IMG_3527.JPG',5)
totalSob = testing_sobel.concatSob('C:\Python34\palstaves2\\2013T482_Lower_Hardres_Canterbury\Axe1\IMG_3527.JPG',5)
from PIL import Image #to here done
im = Image.open('C:\Python34\palstaves2\\2013T482_Lower_Hardres_Canterbury\Axe1\IMG_3527.JPG')
im = np.asarray(im)
im = rescale(im,0.25)
imArray = np.asarray(totalSob)
imArray = np.dstack([imArray,im])#to here done
featureMap = imArray
import pickle
pickleFile = open('C:\Python34\palstaves2\\2013T482_Lower_Hardres_Canterbury\Axe1\Tree.pickle','rb')
classifier = pickle.load(pickleFile)
a=water_test.watershedFunc2('C:\Python34\palstaves2\\2013T482_Lower_Hardres_Canterbury\Axe1\IMG_3527.JPG')
b=water_test.superPix(im,a,featureMap,classifier,100)
imageNameHolder = []
counter = 0
for filepath in imageNames:
if counter % 10 == 0:
if numberSuccessStacked >= trainSetSize or numberStacked == imageSetSize:
break
numberStacked += 1
numberSuccessStacked += 1
fileNameStringWithExtension = os.path.basename(filepath)
fileNameString = os.path.splitext(fileNameStringWithExtension)[0]
maskPath = os.path.join(path, 'masks/'+fileNameString+'_mask')##normally 'masks/'
sobelise.process_image(filepath,levels)
totalSob = testing_sobel.concatSob(filepath,levels) # loading 1/4 sized images
#all levels concatenated together
try:
maskRaw = Image.open(maskPath+'.jpg').convert(mode='L')#convert is new
imageNameHolder.append(fileNameString)
except IOError:
print('Image '+fileNameString+' has no corresponding mask, it has been skipped')
numberSuccessStacked -= 1
continue
im = Image.open(filepath)
im = np.asarray(im)
#im = ndimage.gaussian_filter(im, 3)
'''
sx0 = ndimage.sobel(im[...,...,0], axis=0, mode='constant')
def useClassifier():
path = FILE_PATH
levels = 5
#trainingPath = os.path.join(path,)
training = np.load(path)
shuffled = training['shuffled']
trainRatio = training['R']
#trainingImages = training['header'](images)
newpath = os.path.join(DIR_PATH,'predictedMasks3')
if not os.path.exists(newpath):
os.makedirs(newpath)
print('Sampling rate = '+str(training['S'])+', trainingRatio = '+str(trainRatio))
#print(training.item())
header = training['header'][()]['images']
#print(type(a))
#print((a[()])['images'])
#print(header)
print('Images used as training: '+ str(header))
if CLASSIFIER_TYPE == 'LinearSVC':
try:
pickleFile = open(os.path.join(DIR_PATH,'linear-svm.pickle'), 'rb')
except IOError:
print('Classifier not trained '+'\n'+'##'+'\n'+'##'+'\n'+'##'+'\n'+'##')
print('##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>')
classifier = pickle.load(pickleFile)
elif CLASSIFIER_TYPE == 'Tree':
try:
pickleFile = open(os.path.join(DIR_PATH,'Tree.pickle'), 'rb')
except IOError:
print('Classifier not trained '+'\n'+'##'+'\n'+'##'+'\n'+'##'+'\n'+'##')
print('##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>')
classifier = pickle.load(pickleFile)
else:
print('Classifier requested has not been recognised')
totalError = 0
totTestingError = 0
totTrainingError = 0
imageSetSize = shuffled.shape[0]
numberPredicted = 0
imageIndex = 0
missingTest = 0
missingTrain = 0
for filepath in shuffled: #glob.glob(os.path.join(DIR_PATH, '*.jpg')):
'''
if imageIndex == int(shuffled.shape[0]/trainRatio+1):
averageErrorTraining = totalError/numberPredicted
print('Average error for training set of '+str(int(shuffled.shape[0]/trainRatio+1))+' images is '+ str(averageErrorTraining))
totalError = 0
realTrainSetSize = numberPredicted
'''
print('imageIndex')
print(imageIndex)
print('out of')
print(shuffled.shape[0])
fileNameStringWithExtension = os.path.basename(filepath)
fileNameString = os.path.splitext(fileNameStringWithExtension)[0]
maskPath = os.path.join(DIR_PATH, 'masks/'+fileNameString+'_mask')
sobelise.process_image(filepath,levels)
totalSob = testing_sobel.concatSob(filepath,levels)
try:
maskRaw = Image.open(maskPath+'.jpg')
except IOError:
print('Image '+fileNameString+' has no corresponding mask, it has been skipped')
if imageIndex % trainRatio == 0:
missingTrain +=1
else:
missingTest +=1
imageIndex += 1
continue
im = Image.open(filepath)
im = np.asarray(im)
'''
#im = ndimage.gaussian_filter(im, 3)
sx0 = ndimage.sobel(im[...,...,0], axis=0, mode='constant')
sy0 = ndimage.sobel(im[...,...,0], axis=1, mode='constant')
#sob0 = np.hypot(sx0, sy0)
sx1 = ndimage.sobel(im[...,...,1], axis=0, mode='constant')
sy1 = ndimage.sobel(im[...,...,1], axis=1, mode='constant')
#sob1 = np.hypot(sx1, sy1)
sx2 = ndimage.sobel(im[...,...,2], axis=0, mode='constant')
sy2 = ndimage.sobel(im[...,...,2], axis=1, mode='constant')
#sob2 = np.hypot(sx2, sy2)
sobx = np.dstack([sx0,sx1,sx2])
soby = np.dstack([sy0,sy1,sy2])
sobx_blurred0 = ndimage.gaussian_filter(sobx, 8)
soby_blurred0 = ndimage.gaussian_filter(soby, 8)
#sob_blurred1 = ndimage.gaussian_filter(sob1_3D, 8)
#sob_blurred2 = ndimage.gaussian_filter(sob2_3D, 8)
#sob_blurred = sob_blurred0+sob_blurred1+sob_blurred2
#sob_blurred2 = ndimage.gaussian_filter(sob_blurred, 8)
#sob_blurred3 = ndimage.gaussian_filter(sob_blurred2, 8)
imWithSobBlurred0 = np.dstack([im,sobx,soby,sobx_blurred0,soby_blurred0])
'''
im = rescale(im,0.25)
imArray = np.asarray(totalSob)
imArray = np.dstack([imArray,im])
#new
featureMap = imArray
a=water_test.watershedFunc2(filepath)
b,totClassified,totMask2=water_test.superPix(im,a,featureMap,classifier,100)
#new end
#imArray = im
maskArray = np.asarray(maskRaw)
maskArray = resize(maskArray,[totalSob.shape[0],totalSob.shape[1]])
maskArray *= 255
flatMaskArray = maskArray.reshape(maskArray.shape[0]*maskArray.shape[1],1)
flatImArray = imArray.reshape(imArray.shape[0]*imArray.shape[1],imArray.shape[2])
#predictedMask = classifier.predict(flatImArray)#for superpix
numberPredicted += 1
pixelCount = flatImArray.shape[0]
outputSampleCount = int(1*pixelCount)
#indices = np.random.choice(pixelCount,replace=False,size=outputSampleCount)
X = flatImArray#flatImArray[indices,...]
y = flatMaskArray#flatMaskArray[indices,...]
#yPrime = predictedMask.astype(np.int)#for superpix
yPrime = b #new line for superpix
yPrime = np.asarray(yPrime)
totMask2 = np.asarray(totMask2)
totMask2 = np.reshape(totMask2,(totalSob.shape[0],totalSob.shape[1]))
totMask2 = rescale(totMask2,4,preserve_range=True)
totMask2 *= 255
totMask2 = totMask2.astype(np.uint8)
yPrime = np.reshape(yPrime, (-1, 1)) # -1 means make it whatever it needs to be
print(yPrime.shape)
print(np.max(yPrime))
yPrimeForMaskSave = np.reshape(yPrime,(totalSob.shape[0],totalSob.shape[1]))
print(np.max(yPrimeForMaskSave))
yPrimeForMaskSave = rescale(yPrimeForMaskSave,4,preserve_range=True)
print(np.max(yPrimeForMaskSave))
print(np.max(yPrimeForMaskSave))
yPrimeForMaskSave *= 255
yPrimeForMaskSave = yPrimeForMaskSave.astype(np.uint8)
print(os.path.join(newpath,fileNameString+'_mask'))
Image.fromarray(totMask2).save(os.path.join(newpath,fileNameString+'_ratio_mask.jpg'))
Image.fromarray(yPrimeForMaskSave).save(os.path.join(newpath,fileNameString+'_mask.jpg'))
Image.fromarray((totClassified*255).astype(np.uint8)).save(os.path.join(newpath,fileNameString+'_basic_mask.jpg'))
#yPrime = (yPrime>64).astype(np.int)
y = (y>64).astype(np.int)
absError = (np.absolute(y-yPrime)).sum()
print('Error from image '+fileNameString+ ' is '+str(absError))
if imageIndex % trainRatio == 0:
print('Training Image')
totTrainingError = totTrainingError+absError
else:
totTestingError = totTestingError+absError
#totalError = totalError+absError
imageIndex += 1
'''
if imageIndex == int(shuffled.shape[0]/trainRatio):
averageErrorTraining = totalError/numberPredicted
print('Average error for training set of '+str(int(shuffled.shape[0]/trainRatio))+' images is '+ str(averageErrorTraining))
totalError = 0
realTrainSetSize = numberPredicted - 1
averageErrorTest = totalError/(numberPredicted-realTrainSetSize)
print('Average error for testing set of '+str(imageSetSize-shuffled.shape[0]/trainRatio)+' images is '+ str(averageErrorTest))
'''
print('Number Predicted = ' + str(numberPredicted) +' out of '+str(shuffled.shape[0]))
averageErrorTraining = totTrainingError/(len(header)-missingTrain)
print('Average error for training set (predicted only) of '+str(int((shuffled.shape[0]/trainRatio+1)-missingTrain))+' images is '+ str(averageErrorTraining))
averageErrorTest = totTestingError/(shuffled.shape[0]-len(header)-missingTest)
print('Average error for testing set (predicted only) of '+str((shuffled.shape[0]-len(header)-missingTest))+' images is '+ str(averageErrorTest))
def useClassifier(FILE_PATH,levels,CLASSIFIER_TYPE,trainSample,superPixMethod,brushMasks,features,triGrown):
path = FILE_PATH
#levels = 5
#trainingPath = os.path.join(path,)
training = np.load(os.path.join(path,'trainingData_'+str(levels)+'_'+\
'brush'+str(brushMasks)+'_'+str(superPixMethod)+'_'+str(features)+'_'+'grown'+str(triGrown)+'.npz'))
shuffled = training['shuffled']
trainRatio = training['R']
#trainingImages = training['header'](images)
newpath = os.path.join(path,'predictedMasks')
k=0
header = training['header'][()]['images']
searchingFolders = True
#masksInfo = (levels,CLASSIFIER_TYPE,trainSample,superPixMethod)
#pickle.dump( masksInfo, open(os.path.join(newpath,"maskInfo.pickle"), "wb" ) )
while searchingFolders == True:
#print((json.load(open(os.path.join(newpath,"maskInfo.json"), 'rb'))==
#(levels,CLASSIFIER_TYPE,trainSample,superPixMethod)))
newpath=os.path.join(path,'predictedMasks')+str(k)
if os.path.exists(newpath):
File = open(os.path.join(newpath,"maskInfo.json"), 'r')
loadInfo = json.load(File)
#a=pickle.load(open(os.path.join(newpath,"maskInfo.pickle")))
#print( (loadInfo))
#print( [levels,CLASSIFIER_TYPE,trainSample,superPixMethod])
#print((loadInfo==[levels,CLASSIFIER_TYPE,trainSample,superPixMethod]))
#print((loadInfo))
print((levels,CLASSIFIER_TYPE,trainSample,superPixMethod))
#print loadInfo
if not os.path.exists(newpath):
searchingFolders = False
os.makedirs(newpath)
masksInfo = {'levels':levels,'CLASSIFIER_TYPE':CLASSIFIER_TYPE,
'trainSample':trainSample,'superPixMethod':superPixMethod
,'brushMasks':brushMasks,'features':features,'triGrown':triGrown}
json.dump( masksInfo, open(os.path.join(newpath,"maskInfo.json"), "w" ) )
print('first')
json.dump(header,open(os.path.join(newpath,"trainInfo.json"), "w" ))
elif (loadInfo=={'levels':levels,'CLASSIFIER_TYPE':CLASSIFIER_TYPE,
'trainSample':trainSample,'superPixMethod':superPixMethod
,'brushMasks':brushMasks,'features':features,'triGrown':triGrown}):
searchingFolders = False
print('here')
#l=lp
'''
elif (loadInfo!=(levels,CLASSIFIER_TYPE,trainSample,superPixMethod)):
#newpath=os.path.join(path,'predictedMasks')+str(k)
if not os.path.exists(newpath):
searchingFolders = False
os.makedirs(newpath)
masksInfo = (levels,CLASSIFIER_TYPE,trainSample,superPixMethod)
json.dump( masksInfo, open(os.path.join(newpath,"maskInfo.json"), "w" ) )
elif (loadInfo==(levels,CLASSIFIER_TYPE,trainSample,superPixMethod)):
searchingFolders = False
elif (loadInfo==(levels,CLASSIFIER_TYPE,trainSample,superPixMethod)):
searchingFolders = False
'''
k +=1
print('Sampling rate = '+str(training['S'])+', trainingRatio = '+str(trainRatio))
#print(training.item())
header = training['header'][()]['images']
#print(type(a))
#print((a[()])['images'])
#print(header)
print('Images used as training: '+ str(header))
if CLASSIFIER_TYPE == 'LinearSVC':
try:
pickleFile = open(os.path.join(path,'LinearSVC'+'_'+str(levels)+'_'+\
'brush'+str(brushMasks)+'_'+str(superPixMethod)+'_'+str(features)+'_'+'grown'+str(triGrown)+'.pickle'), 'rb')
except IOError:
print('Classifier not trained '+'\n'+'##'+'\n'+'##'+'\n'+'##'+'\n'+'##')
print('##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>')
classifier = pickle.load(pickleFile)
elif CLASSIFIER_TYPE == 'Tree':
try:
pickleFile = open(os.path.join(path,'Tree'+'_'+str(levels)+'_'+\
'brush'+str(brushMasks)+'_'+str(superPixMethod)+'_'+str(features)+'_'+'grown'+str(triGrown)+'.pickle'), 'rb')
except IOError:
print('Classifier not trained '+'\n'+'##'+'\n'+'##'+'\n'+'##'+'\n'+'##')
print('##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>'+'\n'+'##>>>>>>>>')
classifier = pickle.load(pickleFile)
#print(os.path.join(path,'Tree'+'_'+str(levels)+'_'+str(trainSample)+'.pickle'))
else:
print('Classifier requested has not been recognised')
totalError = 0
totTestingError = 0
totTrainingError = 0
imageSetSize = shuffled.shape[0]
numberPredicted = 0
imageIndex = 0
missingTest = 0
missingTrain = 0
for filepath in shuffled: #glob.glob(os.path.join(DIR_PATH, '*.jpg')):
'''
if imageIndex == int(shuffled.shape[0]/trainRatio+1):
averageErrorTraining = totalError/numberPredicted
print('Average error for training set of '+str(int(shuffled.shape[0]/trainRatio+1))+' images is '+ str(averageErrorTraining))
totalError = 0
realTrainSetSize = numberPredicted
'''
#print(imageIndex)
#print('out of')
#print(shuffled.shape[0])
fileNameStringWithExtension = os.path.basename(filepath)
fileNameString = os.path.splitext(fileNameStringWithExtension)[0]
maskPath = os.path.join(path, 'masks/'+fileNameString+'_mask')
brushMaskPath = os.path.join(path, 'brushMasks/'+fileNameString+'_mask'+'.jpg')
trainMaskPath = os.path.join(path, 'trainMasks/'+fileNameString+'_mask'+'.jpg')
procTrain = False
if not os.path.exists(os.path.join(newpath,fileNameString+'_mask.jpg')):
print('Image '+str(imageIndex+1)+' out of '+str(shuffled.shape[0]))
sobelise.process_image(filepath,levels,features)
totalSob = testing_sobel.concatSob(filepath,levels,features)
maskMissing = False
try:
maskRaw = Image.open(maskPath+'.jpg')
maskMissing = False
print 'harpy'
if os.path.exists(brushMaskPath) and brushMasks==True:
procTrain = True
if os.path.exists(trainMaskPath) and brushMasks==False:
procTrain = True
imageIndex += 1
except IOError:
print('Image '+fileNameString+' has no corresponding mask, therefore error cannot be calculated')
if os.path.exists(brushMaskPath) and brushMasks==True:#imageIndex % trainRatio == 0:
missingTrain +=1
procTrain = True
print('exists 0')
elif os.path.exists(trainMaskPath) and brushMasks==False:#imageIndex % trainRatio == 0:
missingTrain +=1
procTrain = True
print('exists 0')
else:
missingTest +=1
imageIndex += 1
maskMissing = True
#continue
im = Image.open(filepath)
im = np.asarray(im)
'''
#im = ndimage.gaussian_filter(im, 3)
sx0 = ndimage.sobel(im[...,...,0], axis=0, mode='constant')
sy0 = ndimage.sobel(im[...,...,0], axis=1, mode='constant')
#sob0 = np.hypot(sx0, sy0)
sx1 = ndimage.sobel(im[...,...,1], axis=0, mode='constant')
sy1 = ndimage.sobel(im[...,...,1], axis=1, mode='constant')
#sob1 = np.hypot(sx1, sy1)
sx2 = ndimage.sobel(im[...,...,2], axis=0, mode='constant')
sy2 = ndimage.sobel(im[...,...,2], axis=1, mode='constant')
#sob2 = np.hypot(sx2, sy2)
sobx = np.dstack([sx0,sx1,sx2])
soby = np.dstack([sy0,sy1,sy2])
sobx_blurred0 = ndimage.gaussian_filter(sobx, 8)
soby_blurred0 = ndimage.gaussian_filter(soby, 8)
#sob_blurred1 = ndimage.gaussian_filter(sob1_3D, 8)
#sob_blurred2 = ndimage.gaussian_filter(sob2_3D, 8)
#sob_blurred = sob_blurred0+sob_blurred1+sob_blurred2
#sob_blurred2 = ndimage.gaussian_filter(sob_blurred, 8)
#sob_blurred3 = ndimage.gaussian_filter(sob_blurred2, 8)
imWithSobBlurred0 = np.dstack([im,sobx,soby,sobx_blurred0,soby_blurred0])
'''
#im = rescale(im,0.25)
im = rescale(im,0.25)#125)
if features=='RGB'or features=='entropy'or features=='dwt':
imArray = im*255 # normalising
elif features=='sobel' or features=='sobelHandv' or features=='combinedEntSob'or features=='combinedDwtSob':
imArray = np.asarray(totalSob)
imArray = np.dstack([imArray,im*255])
elif features =='sobelSansRGB':
imArray = np.asarray(totalSob)
if features =='entropy'or features =='dwt' or features =='combinedDwtSob' or features =='combinedEntSob':
dwtFeature = dwtSlide(filepath,4,features)
'''
abc = dwtFeature[:,0].reshape(im.shape[0],im.shape[1])
abc = abc/np.max(abc)
b = dwtFeature[:,1].reshape(im.shape[0],im.shape[1])
b = b/np.max(b)
abc2 = dwtFeature[:,2].reshape(im.shape[0],im.shape[1])
abc2 = abc2/np.max(abc2)
b2 = dwtFeature[:,3].reshape(im.shape[0],im.shape[1])
b2 = b2/np.max(b2)
abc3 = dwtFeature[:,4].reshape(im.shape[0],im.shape[1])
abc3 = abc3/np.max(abc3)
#b3 = dwtFeature[:,7].reshape(im.shape[0],im.shape[1])
#b3 = b3/np.max(b)
abc = np.hstack([abc,b,abc2,b2,abc3])
'''
#abc = dwtFeature[:,0].reshape(im.shape[0],im.shape[1])
#abc = abc/np.max(abc)
#b = dwtFeature[:,1].reshape(im.shape[0],im.shape[1])
#b = b/np.max(b)
#abc = np.hstack([abc,b])
flatIm = im.reshape(im.shape[0]*im.shape[1],-1)
#flatIm = np.zeros((flatIm.shape[0],flatIm.shape[1])) #for dwt testing
##flatImArray = np.hstack([flatIm,dwtFeature])
flatImArray = imArray.reshape(imArray.shape[0]*imArray.shape[1],imArray.shape[2])
if features=='combinedEntSob'or features=='combinedDwtSob' or features=='entropy' or features=='dwt':
flatImArray = np.hstack([flatImArray,dwtFeature])
featureMap = flatImArray
#print('here b4')
a=water_test.watershedFunc2(filepath,superPixMethod)
#print('between')
b,totClassified,totMask2,segmentOutlines,totMask=water_test.superPix(im,a,featureMap,classifier,100)
if superPixMethod == 'None':
b=totClassified
#print('here after')
print(np.unique((segmentOutlines*255).astype(np.uint8)))
#new end
#imArray = im
#maskArray = np.asarray(maskRaw)
#maskArray = resize(maskArray,[totalSob.shape[0],totalSob.shape[1]])
#maskArray *= 255
#flatMaskArray = maskArray.reshape(maskArray.shape[0]*maskArray.shape[1],1)
###flatImArray = imArray.reshape(imArray.shape[0]*imArray.shape[1],imArray.shape[2])
#predictedMask = classifier.predict(flatImArray)#for superpix
numberPredicted += 1
pixelCount = flatImArray.shape[0]
outputSampleCount = int(1*pixelCount)
#indices = np.random.choice(pixelCount,replace=False,size=outputSampleCount)
X = flatImArray#flatImArray[indices,...]
#y = flatMaskArray#flatMaskArray[indices,...]
#yPrime = predictedMask.astype(np.int)#for superpix
yPrime = b #new line for superpix
yPrime = np.asarray(yPrime)
'''
totMask2 = np.asarray(totMask2)
totMask2 = np.reshape(totMask2,(totalSob.shape[0],totalSob.shape[1]))
totMask2 = rescale(totMask2,4,preserve_range=True)
totMask2 *= 255
totMask2 = totMask2.astype(np.uint8)
'''
print yPrime.shape
yPrime = np.reshape(yPrime, (-1, 1)) # -1 means make it whatever it needs to be
#print(yPrime.shape)
#print(np.max(yPrime))
print yPrime.shape
print im.shape
yPrimeForMaskSave = np.reshape(yPrime,(im.shape[0],im.shape[1]))
yPrimeForMaskSaveCopy = np.reshape(yPrime,(im.shape[0],im.shape[1]))
#print(np.max(yPrimeForMaskSave))
yPrimeForMaskSave = rescale(yPrimeForMaskSave,8,preserve_range=True,order=0)#order was 1
#print(np.max(yPrimeForMaskSave))
#print(np.max(yPrimeForMaskSave))
yPrimeForMaskSave *= 255
##yPrimeForMaskSaveCopy = yPrimeForMaskSaveCopy.astype(np.uint8)
##yPrimeForMaskSaveCopy *= 255
yPrimeForMaskSave = yPrimeForMaskSave.astype(np.uint8)
##yPrimeForMaskSaveCopy = yPrimeForMaskSaveCopy.astype(np.uint8)
#print(os.path.join(newpath,fileNameString+'_mask'))
if not os.path.exists(os.path.join(path,'preMasks'+str(k-1))):
os.makedirs(os.path.join(path,'preMasks'+str(k-1)))
basicPath = os.path.join(path,'preMasks'+str(k-1))
print(np.max(totMask2))
print(np.min(totMask2))
#Image.fromarray((abc*255).astype(np.uint8)).save(os.path.join(basicPath,fileNameString+'abc_mask.jpg'))
Image.fromarray(np.uint8(cm.afmhot(totMask2)*255)).save(os.path.join(basicPath,fileNameString+'_ratio_mask.jpg'))
Image.fromarray(yPrimeForMaskSave).save(os.path.join(newpath,fileNameString+'_mask.jpg'))
Image.fromarray((totClassified*255).astype(np.uint8)).save(os.path.join(basicPath,fileNameString+'_basic_mask.jpg'))
segImage = Image.fromarray((((segmentOutlines*255).astype(np.uint8))))
segImage=segImage.convert('RGB')
yPrimeForMaskSaveImage = Image.fromarray((((yPrimeForMaskSaveCopy*255).astype(np.uint8))))
yPrimeForMaskSaveImage=yPrimeForMaskSaveImage.convert('RGB')
yPrimeForMaskSaveImage2 = np.array(yPrimeForMaskSaveImage)
yPrimeForMaskSaveImage2[:,:,1:3]=0
print np.max(yPrimeForMaskSaveImage2)
print np.max(yPrimeForMaskSaveCopy)
#l=lp
yPrimeForMaskSaveImage2 = Image.fromarray((((yPrimeForMaskSaveImage2).astype(np.uint8))))
print(np.max(im))
origImage = Image.fromarray((im*255).astype(np.uint8))
#np.array(segImage)[...,1:3]=0
#segImage = Image.fromarray((segmentOutlines).astype(np.uint8))
print(np.array(segImage).shape)
print(np.array(origImage).shape)
blend = Image.blend(segImage,origImage,0.7)
blend.save(os.path.join(basicPath,fileNameString+'_segment_mask.jpg'))
#print(yPrimeForMaskSaveCopy.shape)
print(im.shape)
print(segmentOutlines.shape)
print(yPrimeForMaskSaveCopy.shape)
blend2 = Image.blend((yPrimeForMaskSaveImage2),origImage,0.7)
blend2.save(os.path.join(basicPath,fileNameString+'_final_mask.jpg'))
#Image.fromarray((((segmentOutlines*255).astype(np.uint8)))).save(os.path.join(basicPath,fileNameString+'_segment_mask.jpg'))
#yPrime = (yPrime>64).astype(np.int)
if maskMissing == False:
maskArray = np.asarray(maskRaw)
maskArray = resize(maskArray,[im.shape[0],im.shape[1]])
maskArray *= 255
flatMaskArray = maskArray.reshape(maskArray.shape[0]*maskArray.shape[1],1)
y = flatMaskArray
y = (y>64).astype(np.int)
absError = np.float((np.absolute(y-yPrime)).sum())/(y.shape[0]*y.shape[1])
print('Error from image '+fileNameString+ ' is '+str(absError))
if procTrain==True:#os.path.exists(brushMaskPath):
#print('Training Image')
print('exists 1')
totTrainingError = totTrainingError+absError
else:
totTestingError = totTestingError+absError
#totalError = totalError+absError
else:
print('Image '+str(imageIndex+1)+' out of '+str(shuffled.shape[0])+' already processed')
numberPredicted+=1
try:
maskRaw = Image.open(maskPath+'.jpg')
maskMissing = False
imageIndex += 1
if os.path.exists(brushMaskPath) and brushMasks==True:
procTrain = True
if os.path.exists(trainMaskPath) and brushMasks==False:
procTrain = True
except IOError:
print('Image '+fileNameString+' has no corresponding mask, therefore error cannot be calculated')
if os.path.exists(brushMaskPath) and brushMasks==True:
procTrain = True
if os.path.exists(trainMaskPath) and brushMasks==False:
procTrain = True
if procTrain==True:#os.path.exists(brushMaskPath):#imageIndex % trainRatio == 0:
missingTrain +=1
print('exists 2')
else:
missingTest +=1
imageIndex += 1
maskMissing = True
continue # was commented
imgLoad = np.asarray(Image.open(os.path.join(newpath,fileNameString+'_mask.jpg')))
yPrime = resize(imgLoad,[imgLoad.shape[0]/4,imgLoad.shape[1]/4])
#yPrime = resize(imgLoad,[imgLoad.shape[0]*imgLoad.shape[1]])
if (np.max(yPrime) <= 1):
yPrime *= 255
flatYPrime = yPrime.reshape(yPrime.shape[0]*yPrime.shape[1])
flatYPrime = (flatYPrime>64).astype(np.int)
flatYPrime = np.reshape(flatYPrime, (-1, 1))
if maskMissing == False:
maskArray = np.asarray(maskRaw)
maskArray = resize(maskArray,[imgLoad.shape[0]/4,imgLoad.shape[1]/4])
maskArray *= 255
flatMaskArray = maskArray.reshape(maskArray.shape[0]*maskArray.shape[1],1)
y = flatMaskArray
#print(y.shape)
#print(flatYPrime.shape)
y = (y>64).astype(np.int)
absError = np.float((np.absolute(y-flatYPrime)).sum())/(y.shape[0]*y.shape[1])
print('Error from image '+fileNameString+ ' is '+str(absError))
if procTrain==True:#os.path.exists(brushMaskPath):
print('Training Image')
totTrainingError = totTrainingError+absError
else:
totTestingError = totTestingError+absError
#totalError = totalError+absError
#imageIndex += 1
'''
if imageIndex == int(shuffled.shape[0]/trainRatio):
averageErrorTraining = totalError/numberPredicted
print('Average error for training set of '+str(int(shuffled.shape[0]/trainRatio))+' images is '+ str(averageErrorTraining))
totalError = 0
realTrainSetSize = numberPredicted - 1
averageErrorTest = totalError/(numberPredicted-realTrainSetSize)
print('Average error for testing set of '+str(imageSetSize-shuffled.shape[0]/trainRatio)+' images is '+ str(averageErrorTest))
'''
if len(header)-missingTrain>0:
print('Number Predicted = ' + str(numberPredicted) +' out of '+str(shuffled.shape[0]))
averageErrorTraining = totTrainingError/(len(header)-missingTrain)
print'tot training error'
print totTrainingError
print('Average error for training set (predicted only) of '+str(int((shuffled.shape[0]/trainRatio+1)-missingTrain))+' images is '+ str(averageErrorTraining))
averageErrorTest = totTestingError/(shuffled.shape[0]-len(header)-missingTest)
print('Average error for testing set (predicted only) of '+str((shuffled.shape[0]-len(header)-missingTest))+' images is '+ str(averageErrorTest))
performance = {'size':(shuffled.shape[0]-len(header)-missingTest),'error':str(averageErrorTest)}
json.dump( performance, open(os.path.join(path,"performance_"+str(levels)+'_'+\
'brush'+str(brushMasks)+'_'+str(superPixMethod)+'_'+str(features)+'_'+'grown'+str(triGrown)+".json"), "w" ) )
else:
print('Could not calculate error as there were no true masks')