def getSuperPixelTrainingData(msrcDataDirectory,
nbSuperPixels,
superPixelCompactness,
scale,
trainSplit=0.6,
validationSplit=0.2,
testSplit=0.2):
# Should probably make this a call to pomio in case the ordering changes in the future...
voidClassLabel = pomio.getVoidIdx()
# These could be user-specified
if scale == None:
scale = 0.05 # default to 10% of data
msrcData = pomio.msrc_loadImages(msrcDataDirectory, None)
print "Now generating superpixel classifier for MSRC data"
# splitData = pomio.splitInputDataset_msrcData(msrcData, datasetScale=scale, keepClassDistForTraining=True, trainSplit, validationSplit, testSplit )
splitData = pomio.splitInputDataset_msrcData(msrcData, scale, True,
trainSplit, validationSplit,
testSplit)
# prepare superpixel training data
trainingMsrcImages = splitData[0]
# Just use the above function to get superpixel features and labels for training data
return SuperPixelClassifier.getSuperPixelData(trainingMsrcImages,
nbSuperPixels,
superPixelCompactness)
def evaluatePrediction(predictLabels, gtLabels, imageName):
assert np.shape(predictLabels) == np.shape(gtLabels) , "Predict image and ground truth image are not the same size..."
rows = np.shape(predictLabels)[1]
cols = np.shape(gtLabels)[0]
print "Evaluating image of size = [" , rows, " ," , cols, " ]"
voidLabel = pomio.getVoidIdx()
allPixels = 0
voidGtPixels = 0
correctPixels = 0
incorrectPixels = 0
# for each pixel, do a comparision of index
for r in range(0,rows):
for c in range(cols):
allPixels = allPixels + 1
gtLabel = gtLabels[c][r]
predictLabel = predictLabels[c][r]
if gtLabel == voidLabel:
voidGtPixels = voidGtPixels + 1
else:
# only compare if GT isnt void
if (predictLabel != voidLabel) and (predictLabels[c][r] == gtLabels[c][r]):
correctPixels = correctPixels + 1
else:
incorrectPixels = incorrectPixels + 1
assert allPixels == (rows * cols) , "Total iterated pixels != (rows * cols) num pixels!"
assert allPixels == (voidGtPixels + correctPixels + incorrectPixels) , "Some mismatch on pixel counts:: all" + str(allPixels) + " void=" + str(voidGtPixels) + " correct=" + str(correctPixels) + " incorrect=" + str(incorrectPixels)
validGtPixels = allPixels - voidGtPixels
percentage = float(correctPixels) / float(validGtPixels) * 100.0
if percentage == 0 or percentage == 0.0:
print "WARNING:: " + str(imageName) + " accuracy is 0%"
data = "ImageName = " + str(imageName) + "\n\tTotal pixels =" + str(allPixels) + "\n\tVOID pixels = " + str(voidGtPixels) + "\n\tCorrect pixels = " + str(correctPixels) + "\n\tIncorrect pixels=" + str(incorrectPixels) + "\n"
#logFile.write(data)
#zeroListFile.write(imageName + "\n")
print "Pecentage accuracy = " + str( float(correctPixels) / float(validGtPixels) * 100.0 ) + str("%")
return [int(correctPixels), int(validGtPixels), int(voidGtPixels), int(allPixels)]
def countClassAdjacencies( self, nbClasses, allSPClassLabels ):
counts = np.zeros( ( nbClasses, nbClasses ) )
voidLabel = pomio.getVoidIdx()
adjVoidCount = 0
adjCount = len(self.m_edges)
for (ei, ej) in self.m_edges:
ci = allSPClassLabels[ ei ]
cj = allSPClassLabels[ ej ]
# Not doing stats for void
if ci != voidLabel and cj != voidLabel:
counts[ ci, cj ] += 1
counts[ cj, ci ] += 1
else:
adjVoidCount += 1
return (counts, adjVoidCount, adjCount)
def countClassAdjacencies(self, nbClasses, allSPClassLabels):
counts = np.zeros((nbClasses, nbClasses))
voidLabel = pomio.getVoidIdx()
adjVoidCount = 0
adjCount = len(self.m_edges)
for (ei, ej) in self.m_edges:
ci = allSPClassLabels[ei]
cj = allSPClassLabels[ej]
# Not doing stats for void
if ci != voidLabel and cj != voidLabel:
counts[ci - 1, cj - 1] += 1
counts[cj - 1, ci - 1] += 1
else:
adjVoidCount += 1
return (counts, adjVoidCount, adjCount)
def getSuperPixelLabelledImage(image, superPixelMask, superPixelLabels):
superPixels = np.unique(superPixelMask)
numSuperPixels = np.shape(superPixels)[0]
numSuperPixelLabels = np.shape(superPixelLabels)[0]
assert numSuperPixels == numSuperPixelLabels , "The number of superpixels in mask != number superpixel labels:: " + str(numSuperPixels) + " vs. " + str(numSuperPixelLabels)
labelImage = np.zeros( np.shape(image[:,:,0]) )
for idx in range(0, numSuperPixels):
# Assume a consistent ordering between unique superpixel values and superpixel labels
labelImage = labelImage + ( superPixelLabels[idx] * (superPixelMask==superPixels[idx]).astype(int) )
assert np.unique(labelImage).all() == np.unique(superPixelLabels).all() , "List of unique class labels in image != image labels:: " + str(np.unique(labelImage)) + " vs. " + str(superPixelLabels)
assert pomio.getVoidIdx() not in np.unique(superPixelLabels) , "The set of predicted labels for the image contains the void label. It shouldn't."
return labelImage
def evaluatePrediction(predictLabels, gtLabels, gtimageName, dbg=False):
assert np.shape(predictLabels) == np.shape(
gtLabels
), "Predict image and ground truth image are not the same size..."
rows = np.shape(predictLabels)[1]
cols = np.shape(gtLabels)[0]
if dbg:
print "Evaluating image of size = [", rows, " ,", cols, " ]"
voidLabel = pomio.getVoidIdx()
allPixels = 0
voidGtPixels = 0
correctPixels = 0
incorrectPixels = 0
# for each pixel, do a comparision of index
allPixels = rows * cols
voidGtPixels = np.count_nonzero(gtLabels == voidLabel)
correctPixels = np.count_nonzero(
np.logical_and(gtLabels != voidLabel, predictLabels == gtLabels))
validGtPixels = allPixels - voidGtPixels
incorrectPixels = validGtPixels - correctPixels
assert allPixels == (
rows * cols), "Total iterated pixels != (rows * cols) num pixels!"
assert allPixels == (
voidGtPixels + correctPixels +
incorrectPixels), "Some mismatch on pixel counts:: all" + str(
allPixels) + " void=" + str(voidGtPixels) + " correct=" + str(
correctPixels) + " incorrect=" + str(incorrectPixels)
percentage = float(correctPixels) / float(validGtPixels) * 100.0
if percentage == 0 or percentage == 0.0:
print "WARNING:: " + str(gtimageName) + " accuracy is 0%"
if dbg:
print "Pecentage accuracy = " + str(
float(correctPixels) / float(validGtPixels) * 100.0) + str("%")
return np.array([correctPixels, validGtPixels, voidGtPixels, allPixels],
dtype=int)
def evaluateConfusionMatrix(predictedImg, gtImg):
assert np.shape(predictedImg) == np.shape(
gtImg), "Predict image and ground truth image are not the same size..."
numClasses = pomio.getNumClasses()
confusionMatrix = np.zeros([numClasses, numClasses], int)
# rows are actual, cols are predicted
for cl in range(numClasses):
clMask = (gtImg == cl)
# It's easy, just histogram those values
vals = predictedImg[clMask]
assert np.all(np.logical_and(0 <= vals, vals < numClasses)), vals.max()
confusionMatrix[cl, :] = np.histogram(vals, range(numClasses + 1))[0]
assert confusionMatrix.sum() == np.count_nonzero(
gtImg != pomio.getVoidIdx())
return confusionMatrix
def getSuperPixelTrainingData(msrcDataDirectory, nbSuperPixels,superPixelCompactness, scale, trainSplit=0.6, validationSplit=0.2, testSplit=0.2):
# Should probably make this a call to pomio in case the ordering changes in the future...
voidClassLabel = pomio.getVoidIdx()
# These could be user-specified
if scale == None:
scale = 0.05 # default to 10% of data
msrcData = pomio.msrc_loadImages(msrcDataDirectory, None)
print "Now generating superpixel classifier for MSRC data"
# splitData = pomio.splitInputDataset_msrcData(msrcData, datasetScale=scale, keepClassDistForTraining=True, trainSplit, validationSplit, testSplit )
splitData = pomio.splitInputDataset_msrcData(msrcData, scale, True, trainSplit, validationSplit, testSplit )
# prepare superpixel training data
trainingMsrcImages = splitData[0]
# Just use the above function to get superpixel features and labels for training data
return SuperPixelClassifier.getSuperPixelData(trainingMsrcImages,nbSuperPixels,superPixelCompactness)
def computeSuperPixelLabels( gtImage, superPixelObj ):
n = superPixelObj.getNumSuperPixels()
res = pomio.getVoidIdx() * np.ones( (n,), dtype=int )
c = 0
for sp in range( n ):
try:
res[sp] = assignClassLabelToSuperPixel( superPixelObj.getLabelImage()==sp, gtImage )
except:
# Keep a count of the undecided pixels. Default label is void
c += 1
if c > 0:
print 'WARNING: %d of %d superpixels have conflicting GT class labels. Setting to void.' % (c,len(res))
if showDodgySPs:
# Display for debug purposes. set to false to turn it off.
superPixels.displayImage( superPixels.generateImageWithSuperPixelBoundaries( pomio.msrc_convertLabelsToRGB(gtImage),\
superPixelObj.m_labels ), \
imgTitle="Bad image graph", orientation="lower" )
pdb.set_trace()
plt.waitforbuttonpress()
return res
def assignClassLabelToSuperPixel(superPixelValueMask, imagePixelLabels):
"""This function provides basic logic for setting the overall class label for a superpixel"""
voidIdx = pomio.getVoidIdx()
# just adopt the most frequently occurring class label as the superpixel label
superPixelConstituentLabels = imagePixelLabels[superPixelValueMask]
labelCount = stats.itemfreq(superPixelConstituentLabels)
maxLabelFreq = np.max(labelCount[:, 1])
maxLabelIdx = (labelCount[:, 1] == maxLabelFreq)
maxLabel = labelCount[maxLabelIdx]
numSuperPixels = np.sum(labelCount[:, 1])
percentOfPixelsRequired = 0.5
requiredPixelThreshold = np.round(0.5 * numSuperPixels, 0).astype(int)
maxLabelValue = voidIdx
# what if the max count gives more than 1 match? Naughty superpixel.
# It would be nice to select the class that is least represented in the dataset so far...
if np.shape(maxLabel)[0] > 1:
for idx in range(0, np.shape(maxLabel)[0]):
print "\t\tINFO:: max label:", maxLabel[idx, :]
if maxLabel[idx, 0] != voidIdx and maxLabel[
idx, 1] >= requiredPixelThreshold:
maxLabelValue = maxLabel[idx, 0]
break
# If no match, maxLabelValues is equal to void, and will be skipped in training
else:
maxLabelValue = maxLabel[0, 0]
return maxLabelValue
def assignClassLabelToSuperPixel(superPixelValueMask, imagePixelLabels):
"""This function provides basic logic for setting the overall class label for a superpixel"""
voidIdx = pomio.getVoidIdx()
# just adopt the most frequently occurring class label as the superpixel label
superPixelConstituentLabels = imagePixelLabels[superPixelValueMask]
labelCount = stats.itemfreq(superPixelConstituentLabels)
maxLabelFreq = np.max(labelCount[:, 1])
maxLabelIdx = (labelCount[:,1] == maxLabelFreq)
maxLabel = labelCount[maxLabelIdx]
numSuperPixels = np.sum(labelCount[:,1])
percentOfPixelsRequired = 0.5
requiredPixelThreshold = np.round(0.5*numSuperPixels , 0).astype(int)
maxLabelValue = voidIdx
# what if the max count gives more than 1 match? Naughty superpixel.
# It would be nice to select the class that is least represented in the dataset so far...
if np.shape(maxLabel)[0] > 1:
for idx in range(0, np.shape(maxLabel)[0]):
print "\t\tINFO:: max label:" , maxLabel[idx,:]
if maxLabel[idx,0] != voidIdx and maxLabel[idx,1] >= requiredPixelThreshold:
maxLabelValue = maxLabel[idx,0]
break
# If no match, maxLabelValues is equal to void, and will be skipped in training
else:
maxLabelValue = maxLabel[0,0]
return maxLabelValue
def getSuperPixelLabelledImage(image, superPixelMask, superPixelLabels):
superPixels = np.unique(superPixelMask)
numSuperPixels = np.shape(superPixels)[0]
numSuperPixelLabels = np.shape(superPixelLabels)[0]
assert numSuperPixels == numSuperPixelLabels, "The number of superpixels in mask != number superpixel labels:: " + str(
numSuperPixels) + " vs. " + str(numSuperPixelLabels)
labelImage = np.zeros(np.shape(image[:, :, 0]))
for idx in range(0, numSuperPixels):
# Assume a consistent ordering between unique superpixel values and superpixel labels
labelImage = labelImage + (
superPixelLabels[idx] *
(superPixelMask == superPixels[idx]).astype(int))
assert np.unique(labelImage).all() == np.unique(superPixelLabels).all(
), "List of unique class labels in image != image labels:: " + str(
np.unique(labelImage)) + " vs. " + str(superPixelLabels)
assert pomio.getVoidIdx() not in np.unique(
superPixelLabels
), "The set of predicted labels for the image contains the void label. It shouldn't."
return labelImage
def evaluatePrediction(predictLabels, gtLabels, imageName):
assert np.shape(predictLabels) == np.shape(
gtLabels
), "Predict image and ground truth image are not the same size..."
rows = np.shape(predictLabels)[1]
cols = np.shape(gtLabels)[0]
print "Evaluating image of size = [", rows, " ,", cols, " ]"
voidLabel = pomio.getVoidIdx()
allPixels = 0
voidGtPixels = 0
correctPixels = 0
incorrectPixels = 0
# for each pixel, do a comparision of index
for r in range(0, rows):
for c in range(cols):
allPixels = allPixels + 1
gtLabel = gtLabels[c][r]
predictLabel = predictLabels[c][r]
if gtLabel == voidLabel:
voidGtPixels = voidGtPixels + 1
else:
# only compare if GT isnt void
if (predictLabel != voidLabel) and (predictLabels[c][r]
== gtLabels[c][r]):
correctPixels = correctPixels + 1
else:
incorrectPixels = incorrectPixels + 1
assert allPixels == (
rows * cols), "Total iterated pixels != (rows * cols) num pixels!"
assert allPixels == (
voidGtPixels + correctPixels +
incorrectPixels), "Some mismatch on pixel counts:: all" + str(
allPixels) + " void=" + str(voidGtPixels) + " correct=" + str(
correctPixels) + " incorrect=" + str(incorrectPixels)
validGtPixels = allPixels - voidGtPixels
percentage = float(correctPixels) / float(validGtPixels) * 100.0
if percentage == 0 or percentage == 0.0:
print "WARNING:: " + str(imageName) + " accuracy is 0%"
data = "ImageName = " + str(imageName) + "\n\tTotal pixels =" + str(
allPixels) + "\n\tVOID pixels = " + str(
voidGtPixels) + "\n\tCorrect pixels = " + str(
correctPixels) + "\n\tIncorrect pixels=" + str(
incorrectPixels) + "\n"
#logFile.write(data)
#zeroListFile.write(imageName + "\n")
print "Pecentage accuracy = " + str(
float(correctPixels) / float(validGtPixels) * 100.0) + str("%")
return [
int(correctPixels),
int(validGtPixels),
int(voidGtPixels),
int(allPixels)
]
def getSuperPixelData(msrcImages,numberSuperPixels, superPixelCompactness):
# Should probably make this a call to pomio in case the ordering changes in the future...
voidClassLabel = pomio.getVoidIdx()
numberImages = len(msrcImages)
# for each image:
# determine superpixel label (discard if void)
# compute superpixel features of valid superpixels
# append features to cumulative array of all super pixel features
# append label to array of all labels
superPixelFeatures = None
superPixelLabels = np.array([], int) # used for superpixel labels
numberVoidSuperPixels = 0 # keep track of void superpixels
nbClasses = pomio.getNumClasses()
classAdjCounts = np.zeros( (nbClasses, nbClasses) )
adjCountsTotal = 0
adjVoidCountsTotal = 0
for imgIdx in range(0, numberImages):
superPixelIgnoreList = np.array([], int) # this is used to skip over the superpixel in feature processing
print "\n**Processing Image#" , (imgIdx + 1) , " of" , numberImages
# get raw image and ground truth labels
img = msrcImages[imgIdx].m_img
imgPixelLabels = msrcImages[imgIdx].m_gt
# create superpixel map and graph for image
spgraph = SuperPixels.computeSuperPixelGraph( img, 'slic', [numberSuperPixels, superPixelCompactness] )
imgSuperPixelMask = spgraph.m_labels
imgSuperPixels = spgraph.m_nodes
numberImgSuperPixels = spgraph.getNumSuperPixels()
# create superpixel exclude list & superpixel label array
allSPClassLabels = []
for spIdx in range(0, numberImgSuperPixels):
superPixelValue = imgSuperPixels[spIdx]
#print "\tINFO: Processing superpixel =", superPixelValue , " of" , numberImgSuperPixels, " in image"
# Assume superpixel labels are sequence of integers
superPixelValueMask = (imgSuperPixelMask == superPixelValue ) # Boolean array for indexing superpixel-pixels
superPixelLabel = assignClassLabelToSuperPixel(superPixelValueMask, imgPixelLabels)
allSPClassLabels.append( superPixelLabel)
if(superPixelLabel == voidClassLabel):
# add to ignore list, increment void count & do not add to superpixel label array
superPixelIgnoreList = np.append(superPixelIgnoreList, superPixelValue)
numberVoidSuperPixels = numberVoidSuperPixels + 1
else:
superPixelLabels = np.append(superPixelLabels, superPixelLabel)
assert len(allSPClassLabels) == numberImgSuperPixels
(theseClassAdjCounts,adjVoidCount,adjCount) = spgraph.countClassAdjacencies( nbClasses, allSPClassLabels )
classAdjCounts += theseClassAdjCounts
adjCountsTotal += adjCount
adjVoidCountsTotal += adjVoidCount
# Now we have the superpixel labels, and an ignore list of void superpixels - time to get the features!
imgSuperPixelFeatures = FeatureGenerator.generateSuperPixelFeatures(img, imgSuperPixelMask, excludeSuperPixelList=superPixelIgnoreList)
if superPixelFeatures == None:
superPixelFeatures = imgSuperPixelFeatures;
else:
# stack the superpixel features into a single list
superPixelFeatures = np.vstack( [ superPixelFeatures, imgSuperPixelFeatures ] )
assert np.shape(superPixelFeatures)[0] == np.shape(superPixelFeatures)[0] , "Number of samples != number labels"
print "\n**Processed total of" , numberImages, "images"
print " %d out of %d adjacencies were ignored due to void (%.2f %%)" % \
(adjVoidCountsTotal, adjCountsTotal, \
100.0*adjVoidCountsTotal/adjCountsTotal)
# Now return the results
return [ superPixelFeatures, superPixelLabels, classAdjCounts ]
def getSuperPixelData(msrcImages, numberSuperPixels, superPixelCompactness):
# Should probably make this a call to pomio in case the ordering changes in the future...
voidClassLabel = pomio.getVoidIdx()
numberImages = len(msrcImages)
# for each image:
# determine superpixel label (discard if void)
# compute superpixel features of valid superpixels
# append features to cumulative array of all super pixel features
# append label to array of all labels
superPixelFeatures = None
superPixelLabels = np.array([], int) # used for superpixel labels
numberVoidSuperPixels = 0 # keep track of void superpixels
nbClasses = pomio.getNumClasses()
classAdjCounts = np.zeros((nbClasses, nbClasses))
adjCountsTotal = 0
adjVoidCountsTotal = 0
for imgIdx in range(0, numberImages):
superPixelIgnoreList = np.array(
[], int
) # this is used to skip over the superpixel in feature processing
print "\n**Processing Image#", (imgIdx + 1), " of", numberImages
# get raw image and ground truth labels
img = msrcImages[imgIdx].m_img
imgPixelLabels = msrcImages[imgIdx].m_gt
# create superpixel map and graph for image
spgraph = SuperPixels.computeSuperPixelGraph(
img, 'slic', [numberSuperPixels, superPixelCompactness])
imgSuperPixelMask = spgraph.m_labels
imgSuperPixels = spgraph.m_nodes
numberImgSuperPixels = spgraph.getNumSuperPixels()
# create superpixel exclude list & superpixel label array
allSPClassLabels = []
for spIdx in range(0, numberImgSuperPixels):
superPixelValue = imgSuperPixels[spIdx]
#print "\tINFO: Processing superpixel =", superPixelValue , " of" , numberImgSuperPixels, " in image"
# Assume superpixel labels are sequence of integers
superPixelValueMask = (
imgSuperPixelMask == superPixelValue
) # Boolean array for indexing superpixel-pixels
superPixelLabel = assignClassLabelToSuperPixel(
superPixelValueMask, imgPixelLabels)
allSPClassLabels.append(superPixelLabel)
if (superPixelLabel == voidClassLabel):
# add to ignore list, increment void count & do not add to superpixel label array
superPixelIgnoreList = np.append(superPixelIgnoreList,
superPixelValue)
numberVoidSuperPixels = numberVoidSuperPixels + 1
else:
superPixelLabels = np.append(superPixelLabels, superPixelLabel)
assert len(allSPClassLabels) == numberImgSuperPixels
(theseClassAdjCounts, adjVoidCount,
adjCount) = spgraph.countClassAdjacencies(nbClasses, allSPClassLabels)
classAdjCounts += theseClassAdjCounts
adjCountsTotal += adjCount
adjVoidCountsTotal += adjVoidCount
# Now we have the superpixel labels, and an ignore list of void superpixels - time to get the features!
imgSuperPixelFeatures = FeatureGenerator.generateSuperPixelFeatures(
img, imgSuperPixelMask, excludeSuperPixelList=superPixelIgnoreList)
if superPixelFeatures == None:
superPixelFeatures = imgSuperPixelFeatures
else:
# stack the superpixel features into a single list
superPixelFeatures = np.vstack(
[superPixelFeatures, imgSuperPixelFeatures])
assert np.shape(superPixelFeatures)[0] == np.shape(
superPixelFeatures)[0], "Number of samples != number labels"
print "\n**Processed total of", numberImages, "images"
print " %d out of %d adjacencies were ignored due to void (%.2f %%)" % \
(adjVoidCountsTotal, adjCountsTotal, \
100.0*adjVoidCountsTotal/adjCountsTotal)
# Now return the results
return [superPixelFeatures, superPixelLabels, classAdjCounts]
def createAndSaveFeatureLabelData(
msrcData, outfileBase, dataType, outfileType, nbSuperPixels, superPixelCompactness, ftype, aggtype, verbose
):
if verbose:
print "Creating feature set ", outfileBase
if dataType == None or dataType == "":
outfileFtrs = "%s_ftrs.%s" % (outfileBase, outfileType)
outfileLabs = "%s_labs.%s" % (outfileBase, outfileType)
else:
outfileFtrs = "%s_%s_ftrs.%s" % (outfileBase, dataType, outfileType)
outfileLabs = "%s_%s_labs.%s" % (outfileBase, dataType, outfileType)
# Check can write these files.
f = open(outfileFtrs, "w")
f.close()
f = open(outfileLabs, "w")
f.close()
if verbose:
print " - computing superpixels"
allSuperPixels = superPixels.computeSuperPixelGraphMulti(
[z.m_img for z in msrcData], "slic", [nbSuperPixels, superPixelCompactness], nbCores=args.nbCores
)
if verbose:
print " - computing features"
superPixelFeatures = features.computeSuperPixelFeaturesMulti(
[z.m_img for z in msrcData], allSuperPixels, ftype, aggtype, asMatrix=True, nbCores=args.nbCores
)
if verbose:
print " - extracting labels"
superPixelLabels = classification.computeSuperPixelLabelsMulti([z.m_gt for z in msrcData], allSuperPixels)
assert np.all(np.isfinite(superPixelFeatures))
# Don't save features with void labels.
good = superPixelLabels != pomio.getVoidIdx()
if not np.all(good):
if verbose:
print " - discarding %d superpixels with void labels" % np.count_nonzero(np.logical_not(good))
superPixelLabels = superPixelLabels[good]
superPixelFeatures = superPixelFeatures[good, :]
if verbose:
print " - writing %d feature vectors of dimension %d to output files" % (
superPixelFeatures.shape[0],
superPixelFeatures.shape[1],
)
# Output
if outfileType == "pkl":
pomio.pickleObject(superPixelFeatures, outfileFtrs)
pomio.pickleObject(superPixelLabels, outfileLabs)
elif outfileType == "csv":
pomio.writeMatToCSV(superPixelFeatures, outfileFtrs)
pomio.writeMatToCSV(superPixelLabels, outfileLabs)
else:
assert False, "unknown output file format " + outfileType
print "Output written to file ", outfileFtrs, " and ", outfileLabs
def createAndSaveFeatureLabelData(msrcData, outfileBase, dataType, outfileType,
nbSuperPixels, superPixelCompactness, ftype,
aggtype, verbose):
if verbose:
print 'Creating feature set ', outfileBase
if dataType == None or dataType == "":
outfileFtrs = '%s_ftrs.%s' % (outfileBase, outfileType)
outfileLabs = '%s_labs.%s' % (outfileBase, outfileType)
else:
outfileFtrs = '%s_%s_ftrs.%s' % (outfileBase, dataType, outfileType)
outfileLabs = '%s_%s_labs.%s' % (outfileBase, dataType, outfileType)
# Check can write these files.
f = open(outfileFtrs, 'w')
f.close()
f = open(outfileLabs, 'w')
f.close()
if verbose:
print ' - computing superpixels'
allSuperPixels = superPixels.computeSuperPixelGraphMulti( \
[ z.m_img for z in msrcData ],
'slic',
[nbSuperPixels, superPixelCompactness], nbCores=args.nbCores )
if verbose:
print ' - computing features'
superPixelFeatures = features.computeSuperPixelFeaturesMulti(
[z.m_img for z in msrcData],
allSuperPixels,
ftype,
aggtype,
asMatrix=True,
nbCores=args.nbCores)
if verbose:
print ' - extracting labels'
superPixelLabels = classification.computeSuperPixelLabelsMulti( \
[z.m_gt for z in msrcData], allSuperPixels )
assert np.all(np.isfinite(superPixelFeatures))
# Don't save features with void labels.
good = (superPixelLabels != pomio.getVoidIdx())
if not np.all(good):
if verbose:
print ' - discarding %d superpixels with void labels' % np.count_nonzero(
np.logical_not(good))
superPixelLabels = superPixelLabels[good]
superPixelFeatures = superPixelFeatures[good, :]
if verbose:
print ' - writing %d feature vectors of dimension %d to output files' % \
(superPixelFeatures.shape[0], superPixelFeatures.shape[1])
# Output
if outfileType == 'pkl':
pomio.pickleObject(superPixelFeatures, outfileFtrs)
pomio.pickleObject(superPixelLabels, outfileLabs)
elif outfileType == 'csv':
pomio.writeMatToCSV(superPixelFeatures, outfileFtrs)
pomio.writeMatToCSV(superPixelLabels, outfileLabs)
else:
assert False, 'unknown output file format ' + outfileType
print 'Output written to file ', outfileFtrs, ' and ', outfileLabs
