def Save(self, p_sFileName):
oData = {
"FileFormat": "TALOS008",
"IsBinary": self.IsBinary,
"EpochNumber": self.EpochNumber,
"FileNames": self.FileNames,
"Accuracy": self.Accuracy,
"Recall": self.Recall,
"Precision": self.Precision,
"F1Score": self.F1Score,
"CrossF1Score": self.CrossF1Score,
"ObjectiveF1Score": self.ObjectiveF1Score,
"PositiveF1Score": self.PositiveF1Score,
"BestEpochs": self.BestEpochs,
"BestPoints": self.BestPoints,
"BestRecall": self.BestRecall,
"BestPrecision": self.BestPrecision,
"BestF1Score": self.BestF1Score,
"BestCrossF1Score": self.BestCrossF1Score,
"BestObjectiveF1Score": self.BestObjectiveF1Score,
"BestPositiveF1Score": self.BestPositiveF1Score,
"DiscardedEpochs": self.DiscardedEpochs,
"BestRecallEpochs": self.BestRecallEpochs,
"BestPrecisionEpochs": self.BestPrecisionEpochs,
"BestF1ScoreEpochs": self.BestF1ScoreEpochs,
"BestCrossF1ScoreEpochs": self.BestCrossF1ScoreEpochs,
"BestObjectiveF1ScoreEpochs": self.BestObjectiveF1ScoreEpochs,
"BestPositiveScoreEpochs": self.BestPositiveScoreEpochs
}
Storage.SerializeObjectToFile(p_sFileName,
oData,
p_bIsOverwritting=True)
def Save(self, p_sFileName):
oData = {
"FileFormat": "TALOS008",
"Kind": self.Kind,
"IDs": self.IDs,
"Actual": self.ActualClasses,
"Predicted": self.PredictedClasses,
"PredictedProbsTop": self.PredictedProbsTop,
"TopKappa": self.TopKappa,
"Accuracy": self.Accuracy,
"TopKAccuracy": self.TopKAccuracy,
"AveragePrecision": self.AveragePrecision,
"AverageRecall": self.AverageRecall,
"AverageF1Score": self.AverageF1Score,
"AverageSupport": self.AverageSupport
#,"Top1Error" : None
#,"Top5Error" : None
,
"ClassPrecision": self.Precision,
"ClassRecall": self.Recall,
"ClassF1Score": self.F1Score,
"ClassSupport": self.Support,
"ConfusionMatrix": self.ConfusionMatrix
}
Storage.SerializeObjectToFile(p_sFileName, oData)
def __saveClassesToDisk(self):
oData = {
"FormatVersion": "TALOS10",
"ClassCodes": self.ClassCodes,
"ClassDescr": self.ClassDescr,
"ClassFoldersTrain": self.Train.ClassFolders,
"ClassFoldersVal": self.Validation.ClassFolders,
"ClassFoldersTest": self.Testing.ClassFolders,
"ClassSamplesAvailableTrain": self.Train.ClassSamplesAvailable,
"ClassSamplesAvailableVal": self.Validation.ClassSamplesAvailable,
"ClassSamplesAvailableTest": self.Testing.ClassSamplesAvailable,
"HasTrain": self.Train.IsActive,
"HasVal": self.Validation.IsActive,
"HasTest": self.Testing.IsActive,
"CaltechClassDescr": self.CaltechClassDescr,
"ImageNetClassID": self.ImageNetClassID,
"ImageNetClassCodes": self.ImageNetClassCodes,
"ImageNetClassDescr": self.ImageNetClassDescr,
"TrainSamplesPerClass": self.TrainSamplesPerClass,
"PageSize": self.PageSize
}
Storage.SerializeObjectToFile(self.DataSetFolder.ClassesFile,
oData,
p_bIsOverwritting=True)
def __writeCounter(self, p_nNumber):
self.Counter["RunCounter"] = p_nNumber
Storage.SerializeObjectToFile(self.CountersFileName, self.Counter,
True)
def Save(self, p_nImageDimensions):
oTrain = self.Train.PageIterator(self.PageSize)
for nPageIndex, oPage in enumerate(oTrain):
# [sPageFileName, nIDs, sSampleFiles, nTargets]
sPageFileName = oPage[0]
nIDs = oPage[1]
sSampleFiles = oPage[2]
nTargets = oPage[3]
nSamples = np.zeros((len(sSampleFiles), p_nImageDimensions[0],
p_nImageDimensions[1], 3),
dtype=np.uint8)
print("%d/%d samples:" % (nPageIndex + 1, oTrain.EstimatedPages),
nSamples.shape)
if not Storage.IsExistingFile(sPageFileName):
for nIndex, sFileName in enumerate(sSampleFiles):
#img = timg.LoadImageAndCropToSize(sFileName, p_tSize=p_nImageDimensions)
#nSamples[nIndex,:,:,:]=img[:,:,:]
img = timg.LoadImageAndMakeAugmentedSquare(
sFileName, p_tSize=p_nImageDimensions)
# Place the RGB properties in the 4th dimension of the tensor in order to be Tensorflow ready
nSamples[nIndex, :, :, :] = img[0][:, :, :]
oData = {"IDs": nIDs, "Samples": nSamples, "Targets": nTargets}
Storage.SerializeObjectToFile(
sPageFileName,
oData,
p_nExtraLabel="%d/%d" %
(nPageIndex + 1, oTrain.EstimatedPages))
else:
print(" {%d/%d} Exists %s" %
(nPageIndex + 1, oTrain.EstimatedPages, sPageFileName))
oVal = self.Validation.PageIterator(self.PageSize)
for nPageIndex, oPage in enumerate(oVal):
# [sPageFileName, nIDs, sSampleFiles, nTargets]
sPageFileName = oPage[0]
nIDs = oPage[1]
sSampleFiles = oPage[2]
nTargets = oPage[3]
nSamples = np.zeros((len(sSampleFiles), p_nImageDimensions[0],
p_nImageDimensions[1], 3),
dtype=np.uint8)
print("%d/%d samples:" % (nPageIndex + 1, oVal.EstimatedPages),
nSamples.shape)
if not Storage.IsExistingFile(sPageFileName):
for nIndex, sFileName in enumerate(sSampleFiles):
img = timg.LoadImageAndMakeAugmentedSquare(
sFileName, p_tSize=p_nImageDimensions)
nSamples[nIndex, :, :, :] = img[0][:, :, :]
oData = {"IDs": nIDs, "Samples": nSamples, "Targets": nTargets}
Storage.SerializeObjectToFile(
sPageFileName,
oData,
p_nExtraLabel="%d/%d" %
(nPageIndex + 1, oVal.EstimatedPages))
else:
print(" {%d/%d} Exists %s" %
(nPageIndex + 1, oVal.EstimatedPages, sPageFileName))
oTrain = self.Testing.PageIterator(self.PageSize)
for nPageIndex, oPage in enumerate(oTrain):
# [sPageFileName, nIDs, sSampleFiles, nTargets]
sPageFileName = oPage[0]
nIDs = oPage[1]
sSampleFiles = oPage[2]
nTargets = oPage[3]
nSamples = np.zeros((len(sSampleFiles), p_nImageDimensions[0],
p_nImageDimensions[1], 3),
dtype=np.uint8)
print("%d/%d samples:" % (nPageIndex + 1, oTrain.EstimatedPages),
nSamples.shape)
if not Storage.IsExistingFile(sPageFileName):
for nIndex, sFileName in enumerate(sSampleFiles):
img = timg.LoadImageAndMakeAugmentedSquare(
sFileName, p_tSize=p_nImageDimensions)
nSamples[nIndex, :, :, :] = img[0][:, :, :]
oData = {"IDs": nIDs, "Samples": nSamples, "Targets": nTargets}
Storage.SerializeObjectToFile(
sPageFileName,
oData,
p_nExtraLabel="%d/%d" %
(nPageIndex + 1, oTrain.EstimatedPages))
else:
print(" {%d/%d} Exists %s" %
(nPageIndex + 1, oTrain.EstimatedPages, sPageFileName))