Replaced whole deepmedic source from Theano to Tensorflow version.

deepmedic/dataManagement/samplingType.py

@@ -6,12 +6,11 @@
 # or read the terms at https://opensource.org/licenses/BSD-3-Clause.
 
 from __future__ import absolute_import, print_function, division
-from six.moves import xrange
 import numpy as np
 
 class SamplingType(object) :
-    def __init__(self, myLogger, samplingType, numberOfClassesInclBackgr):
-        self.logger = myLogger
+    def __init__(self, log, samplingType, numberOfClassesInclBackgr):
+        self.log = log
         # foreBackgr=0 , uniform=1, fullImage=2, targettedPerClass=3
         self.samplingType = samplingType
 
@@ -26,13 +25,13 @@ def __init__(self, myLogger, samplingType, numberOfClassesInclBackgr):
             self.arrayWithStringPerCategoryToSample = ["WholeImage"]
         elif self.samplingType == 3 :
             self.stringOfSamplingType = "Per-Class"
-            self.arrayWithStringPerCategoryToSample = ["Class-" + str(i) for i in xrange(numberOfClassesInclBackgr) ]
+            self.arrayWithStringPerCategoryToSample = ["Class-" + str(i) for i in range(numberOfClassesInclBackgr) ]
         else :
-            self.logger.print3("ERROR: Tried to create a SamplingType instance, but the samplingType passed was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
+            self.log.print3("ERROR: Tried to create a SamplingType instance, but the samplingType passed was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
 
     def setPercentOfSamplesPerCategoryToSample(self, percentageOfSamplesPerCategoryOfSampling) :
         if self.samplingType in [0,3] and len(percentageOfSamplesPerCategoryOfSampling) != self.getNumberOfCategoriesToSample() :
-            self.logger.print3("ERROR: In class [SamplingType], the list percentageOfSamplesPerCategoryOfSampling had [" + str(len(percentageOfSamplesPerCategoryOfSampling)) + "] elements. In this case of [" + self.stringOfSamplingType + "], it should have [" + str(self.getNumberOfCategoriesToSample()) + "]! Exiting!"); exit(1)
+            self.log.print3("ERROR: In class [SamplingType], the list percentageOfSamplesPerCategoryOfSampling had [" + str(len(percentageOfSamplesPerCategoryOfSampling)) + "] elements. In this case of [" + self.stringOfSamplingType + "], it should have [" + str(self.getNumberOfCategoriesToSample()) + "]! Exiting!"); exit(1)
 
         if self.samplingType == 0 :
             self.percentOfSamplesPerCategoryToSample = self.normalizePercentages(percentageOfSamplesPerCategoryOfSampling)
@@ -43,7 +42,7 @@ def setPercentOfSamplesPerCategoryToSample(self, percentageOfSamplesPerCategoryO
         elif self.samplingType == 3 :
             self.percentOfSamplesPerCategoryToSample = self.normalizePercentages(percentageOfSamplesPerCategoryOfSampling)
         else :
-            self.logger.print3("ERROR: in [SamplingType]. self.samplingType was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
+            self.log.print3("ERROR: in [SamplingType]. self.samplingType was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
 
     def normalizePercentages(self, listOfWeights) :
         arrayOfWeights = np.asarray(listOfWeights, dtype="float32")
@@ -84,10 +83,10 @@ def logicDecidingAndGivingFinalSamplingMapsForEachCategory( self,
             if providedWeightMapsToSampleForEachCategory : #Both weight maps should be provided currently.
                 numOfProvidedWeightMaps = len(arrayWithWeightMapsWhereToSampleForEachCategory)
                 if numOfProvidedWeightMaps != self.getNumberOfCategoriesToSample() :
-                    self.logger.print3("ERROR: For SamplingType = Fore/Background(0), [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! Two (fore/back) were expected! Exiting!"); exit(1)        
+                    self.log.print3("ERROR: For SamplingType = Fore/Background(0), [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! Two (fore/back) were expected! Exiting!"); exit(1)        
                 finalWeightMapsToSampleFromPerCategoryForSubject = arrayWithWeightMapsWhereToSampleForEachCategory
             elif not providedGtLabelsBool:
-                self.logger.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], if weighted-maps are not provided, at least Ground Truth labels should be given to extract foreground! Exiting!"); exit(1)
+                self.log.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], if weighted-maps are not provided, at least Ground Truth labels should be given to extract foreground! Exiting!"); exit(1)
             elif providedRoiMaskBool : # and providedGtLabelsBool
                 maskForForegroundSampling = (gtLabelsImage>0).astype(int)
                 maskForBackgroundSampling_roiMinusGtLabels = (roiMask>0) * (maskForForegroundSampling==0)
@@ -100,7 +99,7 @@ def logicDecidingAndGivingFinalSamplingMapsForEachCategory( self,
             if providedWeightMapsToSampleForEachCategory :
                 numOfProvidedWeightMaps = len(arrayWithWeightMapsWhereToSampleForEachCategory)
                 if numOfProvidedWeightMaps != 1 :
-                    self.logger.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! One was expected! Exiting!"); exit(1)
+                    self.log.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! One was expected! Exiting!"); exit(1)
                 finalWeightMapsToSampleFromPerCategoryForSubject = arrayWithWeightMapsWhereToSampleForEachCategory #Should be an array with dim1==1 already.
             elif providedRoiMaskBool :
                 finalWeightMapsToSampleFromPerCategoryForSubject = [ roiMask ] #Be careful to not change either of the two arrays later or there'll be a problem.
@@ -110,7 +109,7 @@ def logicDecidingAndGivingFinalSamplingMapsForEachCategory( self,
             if providedWeightMapsToSampleForEachCategory :
                 numOfProvidedWeightMaps = len(arrayWithWeightMapsWhereToSampleForEachCategory)
                 if numOfProvidedWeightMaps != 1 :
-                    self.logger.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! One was expected! Exiting!"); exit(1)
+                    self.log.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! One was expected! Exiting!"); exit(1)
                 finalWeightMapsToSampleFromPerCategoryForSubject = arrayWithWeightMapsWhereToSampleForEachCategory #Should be an array with dim1==1 already.
             elif providedRoiMaskBool :
                 finalWeightMapsToSampleFromPerCategoryForSubject = [ roiMask ] #Be careful to not change either of the two arrays later or there'll be a problem.
@@ -120,16 +119,16 @@ def logicDecidingAndGivingFinalSamplingMapsForEachCategory( self,
             if providedWeightMapsToSampleForEachCategory :
                 numOfProvidedWeightMaps = len(arrayWithWeightMapsWhereToSampleForEachCategory)
                 if numOfProvidedWeightMaps != self.getNumberOfCategoriesToSample() :
-                    self.logger.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! As many as the classes [" + str(self.getNumberOfCategoriesToSample()) + "] (incl Background) were expected! Exiting!"); exit(1)
+                    self.log.print3("ERROR: For SamplingType=[" + self.stringOfSamplingType + "], [" + str(numOfProvidedWeightMaps) + "] weight maps were provided! As many as the classes [" + str(self.getNumberOfCategoriesToSample()) + "] (incl Background) were expected! Exiting!"); exit(1)
                 finalWeightMapsToSampleFromPerCategoryForSubject = arrayWithWeightMapsWhereToSampleForEachCategory #Should have as many entries as classes (incl backgr).
             elif providedGtLabelsBool :
                 finalWeightMapsToSampleFromPerCategoryForSubject = []
-                for cat_i in xrange( self.getNumberOfCategoriesToSample() ) : # Should be the same number as the number of actual classes, including background.
+                for cat_i in range( self.getNumberOfCategoriesToSample() ) : # Should be the same number as the number of actual classes, including background.
                     finalWeightMapsToSampleFromPerCategoryForSubject.append( (gtLabelsImage == cat_i).astype(int) )
             else :
-                self.logger.print3("ERROR: For SamplingType=TargettedPerClass(3), either weightMaps for each class or GT labels should be given! Exiting!"); exit(1)
+                self.log.print3("ERROR: For SamplingType=TargettedPerClass(3), either weightMaps for each class or GT labels should be given! Exiting!"); exit(1)
         else :
-            self.logger.print3("ERROR: Sampling-type-number passed in [logicDecidingAndGivingFinalSamplingMapsForEachCategory] was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
+            self.log.print3("ERROR: Sampling-type-number passed in [logicDecidingAndGivingFinalSamplingMapsForEachCategory] was invalid. Should be [0,1,2,3]. Exiting!"); exit(1)
 
         return finalWeightMapsToSampleFromPerCategoryForSubject
 

deepmedic/frontEnd/configParsing/config.py

@@ -0,0 +1,35 @@
+# Copyright (c) 2016, Konstantinos Kamnitsas
+# All rights reserved.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the BSD license. See the accompanying LICENSE file
+# or read the terms at https://opensource.org/licenses/BSD-3-Clause.
+
+from __future__ import absolute_import, print_function, division
+import os
+
+class Config(object):
+
+    def __init__(self, abs_path_to_cfg):
+        self._configStruct = {}
+        self._abs_path_to_cfg = abs_path_to_cfg # for printing later.
+        print("Given configuration file: ", self._abs_path_to_cfg)
+        exec(open(self._abs_path_to_cfg).read(), self._configStruct)
+        self._check_for_deprecated_cfg()
+
+    def __getitem__(self, key): # overriding the [] operator.
+        return self.get(key)
+
+    def get(self, string1) :
+        return self._configStruct[string1] if string1 in self._configStruct else None
+
+    def get_abs_path_to_cfg(self):
+        return self._abs_path_to_cfg
+
+    def _check_for_deprecated_cfg(self):
+        pass
+
+    def override_file_cfg_with_cmd_line_cfg(self, log, args):
+        pass
+
+

deepmedic/frontEnd/configParsing/modelConfig.py

@@ -0,0 +1,85 @@
+# Copyright (c) 2016, Konstantinos Kamnitsas
+# All rights reserved.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the BSD license. See the accompanying LICENSE file
+# or read the terms at https://opensource.org/licenses/BSD-3-Clause.
+
+from __future__ import absolute_import, print_function, division
+import os
+
+from deepmedic.frontEnd.configParsing.utils import *
+from deepmedic.frontEnd.configParsing.config import Config
+
+
+class ModelConfig(Config):
+
+    #Optional but highly suggested.
+    MODEL_NAME = "modelName"
+    #[REQUIRED] Output:
+    FOLDER_OUTP = "folderForOutput" #MUST BE GIVEN
+
+    #================ MODEL PARAMETERS =================
+    NUM_CLASSES = "numberOfOutputClasses"
+    NUM_INPUT_CHANS = "numberOfInputChannels"
+
+    #===Normal pathway===
+    N_FMS_NORM = "numberFMsPerLayerNormal"
+    KERN_DIM_NORM = "kernelDimPerLayerNormal"
+    RESID_CONN_LAYERS_NORM = "layersWithResidualConnNormal"
+    LOWER_RANK_LAYERS_NORM = "lowerRankLayersNormal"
+
+    #==Subsampled pathway==
+    USE_SUBSAMPLED = "useSubsampledPathway"
+    #The below should be mirroring the pathway, otherwise let them specify them but throw warning all around that receptive field should stay the same!
+    N_FMS_SUBS = "numberFMsPerLayerSubsampled"
+    KERN_DIM_SUBS = "kernelDimPerLayerSubsampled"
+    SUBS_FACTOR = "subsampleFactor"
+    RESID_CONN_LAYERS_SUBS = "layersWithResidualConnSubsampled"
+    LOWER_RANK_LAYERS_SUBS = "lowerRankLayersSubsampled"
+
+    #==Extra hidden FC Layers. Final Classification layer is not included in here.
+    N_FMS_FC = "numberFMsPerLayerFC"
+    KERN_DIM_1ST_FC = "kernelDimFor1stFcLayer"
+    RESID_CONN_LAYERS_FC = "layersWithResidualConnFC"
+
+    #Size of Image Segments
+    SEG_DIM_TRAIN = "segmentsDimTrain"
+    SEG_DIM_VAL = "segmentsDimVal"
+    SEG_DIM_INFER = "segmentsDimInference"
+
+    #==Batch Sizes===
+    #Required.
+    BATCH_SIZE_TR = "batchSizeTrain"
+    BATCH_SIZE_VAL = "batchSizeVal"
+    BATCH_SIZE_INFER = "batchSizeInfer"
+
+    #Dropout Rates:
+    DROP_NORM = "dropoutRatesNormal"
+    DROP_SUBS = "dropoutRatesSubsampled"
+    DROP_FC = "dropoutRatesFc"
+
+    #Initialization method of the kernel weights.
+    CONV_W_INIT = "convWeightsInit"
+    #Activation Function for all convolutional layers:
+    ACTIV_FUNC = "activationFunction"
+
+    #Batch Normalization
+    BN_ROLL_AV_BATCHES = "rollAverageForBNOverThatManyBatches"
+
+
+    def __init__(self, abs_path_to_cfg):
+        Config.__init__(self, abs_path_to_cfg)
+
+    # Called from parent constructor
+    def _check_for_deprecated_cfg(self):
+        msg_part1 = "ERROR: Deprecated input to the config: ["
+        msg_part2 = "]. Please update config and use the new corresponding variable "
+        msg_part3 = "]. Exiting."
+        if self.get("initializeClassic0orDelving1") is not None:
+            print(msg_part1 + "initializeClassic0orDelving1" + msg_part2 + "convWeightsInit" + msg_part3); exit(1)
+        if self.get("relu0orPrelu1") is not None:
+            print(msg_part1 + "relu0orPrelu1" + msg_part2 + "activationFunction" + msg_part3); exit(1)
+
+
+