def fuseImages(locale,inputPath=None,notif=True,outputPath=None):
if notif:
note = notify.getNotify()
note.push('Fusing images from %s' % locale)
else: print('Fusing images from %s' % locale)
if not inputPath:
inputPath = util.checkFolder('JPEGS', Input=True)
tiff = util.checkFolder(locale,path=inputPath)
tiffImages = tracker.parseFolder(tiff,findPosition=False, type='tif')
tiffPairs = tracker.findPair(tiffImages,'tif')
if not outputPath:
outputPath = util.checkFolder(locale, path=inputPath)
outputPath = util.checkFolder('Fused',path=outputPath)
try:
for i,pair in enumerate(tiffPairs):
if notif:
note.push('fusing %d :%d'%(i+1,(len(tiffPairs))))
imageVV, imageVH = satImage(pair[0],findPosition=True), satImage(pair[1],findPosition=True)
imageVV.fuseWith(imageVH,outputFolder=outputPath)
except KeyboardInterrupt:
sys.exit()
if notif:
note.push('Finished fusion')
def select(self, model, steps, index):
if model not in self.models:
raise ValueError('Model %s was not found in the model list ' %
model)
modelPath = util.checkFolder('models')
modelPath = util.checkFolder(model, path=modelPath)
folders = [
f for f in os.listdir(modelPath)
if os.path.isdir(os.path.join(modelPath, f))
]
found = False
for folder in folders:
try:
num = folder
if str(num) == str(index):
found = True
break
except:
pass
if found:
modelPath = os.path.join(modelPath, str(index))
oldStep = self.get(model, modelPath, 'steps')
if not steps:
steps = 0
return modelPath, oldStep + steps
else:
raise ValueError(
'%s model index: %d was not found in %s directory' %
(model, index, modelPath))
def colourImage(self, colour, *args):
colour = os.path.join(util.checkFolder('Colour'), colour)
if os.path.exists(colour):
output = util.checkFolder('Outputs')
output = util.checkFolder('Colour', output)
if len(args) is not 0:
oType = args[0].upper()
output = util.checkFolder(oType, output)
name = self.name.split('.')[0] + '.' + oType.lower()
oFile = os.path.join(output, name)
gdaldem = [
'gdaldem', 'colour-relief', self.path, colour, oFile, 'of',
oType
]
else:
output = checkFolder(self.name.split('.')[1], output)
name = self.name
oFile = os.path.join(output, name)
gdaldem = ['gdaldem', 'color-relief', self.path, colour, oFile]
proc = subprocess.Popen(gdaldem)
proc.wait()
print("Colour relief for %s was created" % oFile)
else:
print('Could not find colour file %s' % colour)
def organise(locale):
directory = util.checkFolder('JPEGS',Input=True)
directory = util.checkFolder(locale,path=directory)
images = tracker.parseFolder(directory,findPosition=False,type='jpg')
imageData = tracker.findImages(images,'jpg',directory)
imageData = labelImages(imageData)
stalker = tracker()
stalker.add(imageData)
stalker.saveTracker()
def polymerization(self,
image2,
mode=None,
saveImages=False,
level=None,
clean=True,
intersectOutputPath=None,
outputFolder=None):
if not intersectOutputPath:
intersectOutputPath = util.checkFolder('Fusion', Input=True)
try:
outputNames = [self.name, image2.name]
except:
outputNames = [self.name, 'IntersectOf' + self.name]
intersectState = self.createIntersectionOf(
image2,
outputNames,
saveImages=saveImages,
outputFolder=intersectOutputPath)
if not mode:
mode = 'db7'
if not level:
level = 7
if type(intersectState) is list:
array1 = intersectState[0]
array2 = intersectState[1]
# No need to delete image that weren't saved
clean = False
else:
image1 = os.path.join(intersectOutputPath, outputNames[0])
array1 = io.imread(image1, True, 'gdal')
image2 = os.path.join(intersectOutputPath, outputNames[1])
array2 = io.imread(image2, True, 'gdal')
fusedImage = self.fuseImages(array1, array2, mode, level)
if not outputFolder:
outputFolder = util.checkFolder('Fusion', Output=True)
oName = os.path.join(outputFolder, 'Fused' + self.name)
io.imsave(oName, fusedImage)
if clean:
try:
os.remove(image1)
os.remove(image2)
except FileNotFoundError:
pass
print('Created fused image :%s' % oName)
def latestPath(model):
modelPath = util.checkFolder('models')
modelPath = util.checkFolder(model, path=modelPath)
latest = 0
folders = [
f for f in os.listdir(modelPath)
if os.path.isdir(os.path.join(modelPath, f))
]
for folder in folders:
try:
num = int(folder)
latest = num
except:
pass
return util.checkFolder(str(latest), path=modelPath)
def createIntersectionOf(
self,
image,
outPutNames,
saveImages=False,
outputFolder=None): #Creates two images in the same area
src = self.dataset
# We want a section of source that matches this:
matchDs = image.dataset
image1Array, image2Array, col1, row1, col2, row2, intersectLongLat = self.getIntersect(
image)
# Output / destination
if saveImages:
image1Filename = outPutNames[0]
image2Filename = outPutNames[1]
if not outputFolder:
output = util.checkFolder("Transform", Output=True)
else:
output = outputFolder
image1Output = os.path.join(output, image1Filename)
image2Output = os.path.join(output, image2Filename)
self.createTiff(image1Array, image1Output, col1, row1)
self.createTiff(image2Array, image2Output, col2, row2)
print('Created intersection images in :%s' % output)
return True
else:
return [image1Array, image2Array]
def getFlip(filepath):
dirpath = os.path.split(filepath)[0]
dirpath = util.checkFolder('Flip',path = dirpath)
name = tracker.getBasename(filepath)
nameFlip = os.path.join(dirpath,name + 'f.jpg')
imageOG = Image.open(filepath)
return tracker.flip(nameFlip,imageOG)
def new(self, model, steps):
if model not in nets.netModel:
raise ValueError(
'Model %s was not found in the model list in nets' % model)
modelPath = util.checkFolder('models')
modelPath = util.checkFolder(model, path=modelPath)
latest = 0
folders = [
f for f in os.listdir(modelPath)
if os.path.isdir(os.path.join(modelPath, f))
]
for folder in folders:
try:
num = int(folder)
latest = num
except:
pass
latest += 1
modelPath = util.checkFolder(str(latest), path=modelPath)
self.addModel(model, modelPath)
return modelPath, steps
def loadCsv(self,name):
filePath = util.checkFolder('JPEGS',Input = True)
filePath = os.path.join(filePath,name)
images = {}
with open(filePath,'r') as imageFile:
read = csv.reader(imageFile,delimiter=';')
first = True
for row in read:
if first:
first = False
else:
basename = row.pop(0)
images.update({basename:row})
return images
def convertTo(self, type, outputPath=None):
types = {'jpeg': ['.jpeg', self.toJpeg]}
if not outputPath:
outputPath = util.checkFolder('Tiff', Output=True)
outputPath = os.path.join(outputPath, 'test.tiff')
elif os.path.exists(os.path.dirname(outputPath)):
outputName = self.name.split('.')[0] + types[type][0]
outputPath = os.path.join(outputPath, outputName)
print('Converting image to %s' % types[type][0][1:])
types[type][1](self.path, outputPath)
print('Finished image to %s' % types[type][0])
def getRotate(filepath):
images = []
dirpath = os.path.split(filepath)[0]
dirpath = util.checkFolder('Rotate',path=dirpath)
name = tracker.getBasename(filepath)
name45 = os.path.join(dirpath, name + '45' + '.jpg')
name90 = os.path.join(dirpath, name + '90' + '.jpg')
name135 = os.path.join(dirpath, name +'135' + '.jpg')
imageOG = Image.open(filepath)
images.append(tracker.rotate(name45,45,imageOG))
images.append(tracker.rotate(name90,90,imageOG))
images.append(tracker.rotate(name135,135,imageOG))
return images
def writeCsv(self,name=None):
filePath= util.checkFolder('JPEGS',Input=True)
if not name:
name = 'tracker.csv'
filePath= os.path.join(filePath,name)
with open(filePath, 'w', newline='') as imageFile:
writer = csv.writer(imageFile, delimiter=';')
headers = ['BASENAME','VH','VV','FUSED','LABEL']
writer.writerow(headers)
for basename in self.files:
tmp = [basename]
for value in self.files[basename]:
tmp.append(value)
writer.writerow(tmp)
print('finished writing')
def loadTracker(self):
trackerName = 'imageTracker.pkl'
trackerPath = util.checkFolder('trackers')
self.path = os.path.join(trackerPath,trackerName)
if os.path.isfile(self.path):
try:
with open(self.path, 'rb') as input:
track = pickle.load(input)
self.files = track.files
self.labels = track.labels
if hasattr(track,'trainList'):
self.trainList = track.trainList
if hasattr(track,'evalLIst'):
self.evalList = track.evalList
except:
return
else: return
def loadTracker(self):
trackerName = 'modelTracker.pkl'
trackerPath = util.checkFolder('trackers')
self.path = os.path.join(trackerPath, trackerName)
self.types = {
'new': self.new,
'latest': self.latest,
'select': self.select
}
if os.path.isfile(self.path):
try:
with open(self.path, 'rb') as input:
track = pickle.load(input)
self.models = track.models
self.modelData = track.modelData
except:
return
else:
return
def fuseWith(self, image2, outputFolder=None, mode=None, level=None):
array1 = self.array()
array2 = image2.array()
if not mode:
mode = 'db7'
if not level:
level = 7
try:
fusedImage = self.fuseImages(array1, array2, mode, level)
except MemoryError:
print('Unable to fuse %s' % self.name)
raise
return
if not outputFolder:
outputFolder = util.checkFolder('Fusion', Output=True)
oName = os.path.join(outputFolder, 'Fused' + self.name)
io.imsave(oName, fusedImage)
print('Created fused image :%s' % oName)
def toJpeg(locale,inputPath =None,outputPath=None, fusion = False):
directory = util.checkFolder('Proccessed', Output=True)
if not inputPath:
inputPath = util.checkFolder(locale,path=directory)
if fusion:
inputPath = util.checkFolder('Fused',path=inputPath)
tiffImages = tracker.parseFolder(inputPath,findPosition=False, type='tif')
if not outputPath:
outputPath = util.checkFolder('JPEGS',path =directory)
outputPath = util.checkFolder(locale,path=outputPath)
if fusion:
outputPath = util.checkFolder('Fused',path=outputPath)
satImage(tiffImages[0]).convertTo('jpeg',outputPath=outputPath)
def evalNN(networkName='incept',
status='latest',
batchSize=35,
index=None,
numEvals=None,
numClones=1,
cloneCpu=False,
tasks=0,
workerReplias=1,
numPSTasks=0,
preprocessThread=2):
modelStalker = modTracker()
stalker = tracker()
trainDir, fullSteps = modelStalker.load(networkName,
status=status,
index=index)
datalist = modelStalker.modelData[trainDir][1]
with tf.Graph().as_default():
#LOOK INTO MODEL CLASS****
# Use for parallelism, cant beat google
deployConfig = modelDeploy.DeploymentConfig(num_clones=1,
clone_on_cpu=cloneCpu,
replica_id=tasks,
num_replicas=workerReplias,
num_ps_tasks=numPSTasks)
with tf.device(deployConfig.variables_device()):
globalStep = slim.create_global_step()
network = getNetFunc(networkName,
numClasses=stalker.numLabels(),
isTraining=False)
fileQueue, counter = dataset.getFileQueue(datalist, [100, 90])
tmp = ''
sampleSize = 0
for l in counter:
value = counter[l]
sampleSize += value
tmp += ' Label %d , Value %d' % (l, value)
modelName = networkName + ': ' + os.path.basename(trainDir)
checkpoint = tf.train.latest_checkpoint(trainDir)
note.push('Running %s' % modelName)
note.push('Dataset size : %s' % tmp)
image, label, path = dataset.readFile(fileQueue)
height, width = netsList[networkName].defaultImageSize, netsList[
networkName].defaultImageSize
image = tf.image.resize_images(image, [height, width], method=0)
# image = preprocess.preprocessImage(image,netsList[networkName].defaultImageSize,netsList[networkName].defaultImageSize,isTraining=True)
images, labels = tf.train.batch([image, label],
batch_size=batchSize,
num_threads=preprocessThread,
capacity=2 * batchSize)
logits, _ = network(images)
restoreVars = slim.get_variables_to_restore()
prediction = tf.argmax(logits, 1)
names_to_values, names_to_updates = slim.metrics.aggregate_metric_map({
'Accuracy':
slim.metrics.streaming_accuracy(prediction, labels),
})
for name, value in names_to_values.items():
summary_name = 'eval/%s' % name
op = tf.summary.scalar(summary_name, value, collections=[])
op = tf.Print(op, [value], summary_name)
tf.add_to_collection(tf.GraphKeys.SUMMARIES, op)
if not numEvals:
numEvals = math.ceil(len(datalist) / float(batchSize))
evalDir = util.checkFolder('eval', path=trainDir)
note.push('Starting for %d evals' % numEvals)
slim.evaluation.evaluate_once(master='',
checkpoint_path=checkpoint,
logdir=evalDir,
num_evals=numEvals,
eval_op=list(names_to_updates.values()),
variables_to_restore=restoreVars)
note.push('Finished eval')