First commit

AddLayers_To3DVol.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+
+import sys
+import argparse
+import time 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Concatenating layer to 3D volume')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--input', type=str, required=True,
+						  help='3D TIFF file (multi-layer)')
+	required.add_argument('--layer1', type=str, required=True,
+						  help='2D TIFF file (single layer)')
+	required.add_argument('--output', type=str, required=True,
+						  help='Combined TIFF file (multi-layer)')
+	options = parser.add_argument_group('Options')
+	required.add_argument('--layer2', type=str, required=True,
+						  help='2D TIFF file (single layer)')
+	options.add_argument('--repeat', type=int, default=15,
+						  help='Repeat tile number (default 15)')
+	return parser
+
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgInput_name = args.input
+	imgLayer1_name = args.layer1
+	imgLayer2_name = args.layer2
+	output_name = args.output
+	repeatNb = args.repeat
+	
+	# Read Combined image - all layers
+	print("\nReading input file - 3D volume...")
+	imgInput = imageio.mimread(imgInput_name,memtest=False)
+	imgInput = np.array(imgInput)
+	# print('\nimgInput type: ', imgInput.dtype)
+	print('\t imgInput shape: ', imgInput.shape)
+
+	# Read Layer1 image
+	print("Reading Layer1 file...")
+	imgLayer1 = imageio.imread(imgLayer1_name)
+	# print('imgLayer1 type: ', imgLayer1.dtype)
+	print('\t imgLayer1 shape: ', imgLayer1.shape)
+
+	# Resample layer1 (repeating values, to match input size)
+	imgLayer1_repeat0 = np.repeat(imgLayer1, repeatNb, axis=0)
+	imgLayer1_repeat = np.repeat(imgLayer1_repeat0, repeatNb, axis=1)
+	imgLayer1_repeat = np.expand_dims(imgLayer1_repeat, axis=0)
+	print('\t imgLayer1_repeat shape: ', imgLayer1_repeat.shape)
+
+	# Stack layer to imgInput, to generate one 3D volume
+	print("Adding 2D layer to 3D volume...")
+	imgAll = np.concatenate((imgInput,imgLayer1_repeat), axis = 0)
+
+	if args.layer2:
+		print("Reading Layer2 file...")
+		imgLayer2 = imageio.imread(imgLayer2_name)
+		# print('imgLayer2 type: ', imgLayer2.dtype)
+		print('\t imgLayer2 shape: ', imgLayer2.shape)
+
+		imgLayer2_repeat0 = np.repeat(imgLayer2, repeatNb, axis=0)
+		imgLayer2_repeat = np.repeat(imgLayer2_repeat0, repeatNb, axis=1)
+		imgLayer2_repeat = np.expand_dims(imgLayer2_repeat, axis=0)
+		print('\t imgLayer2_repeat shape: ', imgLayer2_repeat.shape)
+
+		# Stack layer to imgInput, to generate one 3D volume
+		print("Adding 2D layer to 3D volume...")
+		imgAll = np.concatenate((imgAll,imgLayer2_repeat), axis = 0)
+
+	print("Saving output volume...")
+	print('\t imgAll shape: ', imgAll.shape)
+	imageio.mimwrite(output_name,imgAll)
+
+	Time2 = time.time()
+	TimeDiff = Time2 - Time1
+	print("Execution Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Compute_Density.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+from sklearn.metrics import mean_squared_error
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Data analysis - density computation')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--pred', type=str, required=True,
+		help='Prediction TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+		help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--adjtilesdim', type=int, required=True,
+		help='Adjacent tiles dimensions (e.g. 1, 3 or 5) to exclude NaN border')
+	required.add_argument('--output', type=str, required=True,
+		help='CSV file')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--inclusionmask', type=str, 
+						  help='save inclusion mask as output (diff < 2 pixels)')
+	options.add_argument('--exclusionmask', type=str, 
+						  help='save exclusion mask as output (diff >= 2 pixels)')
+	options.add_argument('--threshold', type=float, default = 2.0,
+		help='threshold for inclusion / exclusion mask')
+	options.add_argument('--verbose', action="store_true",
+						  help='verbose mode')
+	return parser
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+	# print(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgPred_name = args.pred
+	imgPred_basename = os.path.basename(imgPred_name)
+	AdjacentTilesDim = args.adjtilesdim
+	imgGT_name = args.groundtruth
+	output_name = args.output
+	inclusion_mask_name = args.inclusionmask
+	exclusion_mask_name = args.exclusionmask
+	threshold = args.threshold
+	
+
+	# Read Pred and GroundTruth images
+	#print("Reading Pred file...")
+	imgPred = imageio.imread(imgPred_name)
+
+
+	#print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+
+	# Remove NaN border when needed	
+	imgPred_Crop = imgPred
+	imgGT_Crop = imgGT
+	if (AdjacentTilesDim == 3):
+		Border = 1
+		imgPred_Crop = imgPred[Border:-Border,Border:-Border]
+		imgGT_Crop = imgGT[Border:-Border,Border:-Border]
+	elif (AdjacentTilesDim == 5):
+		Border = 2
+		imgPred_Crop = imgPred[Border:-Border,Border:-Border]
+		imgGT_Crop = imgGT[Border:-Border,Border:-Border]
+
+	# Quality control - NaN
+	TestNaN_imgPred_Crop = np.any(np.isnan(imgPred_Crop))
+	TestNaN_imgGT_Crop = np.any(np.isnan(imgGT_Crop))
+
+	# #  List indices with Nan values
+	# ListNaN_imgPred_Crop = np.argwhere(np.isnan(imgPred_Crop))
+	# print('ListNaN_imgPred_Crop: ', ListNaN_imgPred_Crop)
+
+	# Verbose mode
+	if args.verbose:
+		print('imgPred type: ', imgPred.dtype)
+		print('imgPred shape: ', imgPred.shape)		
+		print('imgGT type: ', imgGT.dtype)
+		print('imgGT shape: ', imgGT.shape)
+		print('imgPred_Crop shape: ', imgPred_Crop.shape)
+		print('imgGT_Crop shape: ', imgGT_Crop.shape)
+		print('TestNaN_imgPred_Crop: ', TestNaN_imgPred_Crop)
+		print('TestNaN_imgGT_Crop: ', TestNaN_imgGT_Crop)
+
+	# Compute Image difference
+	np_diff = np.abs(imgPred_Crop - imgGT_Crop)
+	print('np_diff type: ', np_diff.dtype)
+	print('np_diff shape: ', np_diff.shape)		
+		
+	# Generate output mask
+	np_exclusionmask = np.uint8(np.where(np_diff >= threshold, 1, 0))
+	np_inclusionmask = np.uint8(np.where(np_diff < threshold, 1, 0))
+	print('np_inclusionmask type: ', np_inclusionmask.dtype)
+	print('np_inclusionmask shape: ', np_inclusionmask.shape)		
+	
+	# Compute density
+	density = np.sum(np_inclusionmask) / (np_inclusionmask.shape[0] * np_inclusionmask.shape[1])
+	print('density: ',density)
+
+	QC_data = np.array([[imgPred_basename,density]])
+	Columns = ['FileName','Density']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	print(df.head())
+	df.to_csv(output_name, index=False)
+
+	if inclusion_mask_name is not None:
+		print('\t\t Writing output inclusion mask - imageio...')
+		imageio.imwrite(inclusion_mask_name, np_inclusionmask)
+
+	if inclusion_mask_name is not None:
+		print('\t\t Writing output exclusion mask - imageio...')
+		imageio.imwrite(exclusion_mask_name, np_exclusionmask)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Compute_DispLMA_Mask.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Data analysis - NaN computation')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--disp_lma', type=str, required=True,
+		help='Prediction TIFF file (single layer)')
+	required.add_argument('--output', type=str, required=True,
+		help='Output CSV file')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--verbose', action="store_true",
+						  help='verbose mode')
+	options.add_argument('--mask', type=str,
+						  help='verbose mode')
+	return parser
+
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	DispLMA_name = args.disp_lma
+	output_name = args.output
+
+	# Read Pred and GroundTruth images
+	#print("Reading Pred file...")
+	imgDispLMA = imageio.imread(DispLMA_name)
+
+	# Verbose mode
+	if args.verbose:
+		print('imgDispLMA type: ', imgDispLMA.dtype)
+		print('imgDispLMA shape: ', imgDispLMA.shape)		
+		# print('imgGT type: ', imgGT.dtype)
+
+	# Compute NaN
+	Bool_NaN = np.isnan(imgDispLMA)
+	Nb_NaN = np.count_nonzero(Bool_NaN)
+	Nb_NotNaN = np.count_nonzero(~Bool_NaN)
+	Nb_Total = imgDispLMA.shape[0] * imgDispLMA.shape[1]
+	NaN_Percent = Nb_NaN / Nb_Total
+	
+	if args.verbose:
+		print('Nb_NaN: ', Nb_NaN)
+		print('Nb_NotNaN: ', Nb_NotNaN)
+		print('Nb_Total: ', Nb_Total)
+		print('NaN_Percent: ', NaN_Percent)
+
+
+	# Mask
+	DispLMA_Mask = np.uint8(np.reshape(~Bool_NaN, imgDispLMA.shape))
+	if args.verbose:
+		print('DispLMA_Mask type: ', DispLMA_Mask.dtype)
+		print('DispLMA_Mask shape: ', DispLMA_Mask.shape)		
+	
+
+	QC_data = np.array([[DispLMA_name, Nb_NaN, Nb_NotNaN, Nb_Total, NaN_Percent]])
+	Columns = ['FileName','Nb_NaN', 'Nb_NotNaN', 'Nb_Total', 'NaN_Percent']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	print(df.head())
+	df.to_csv(output_name, index=False)
+
+	if (args.mask is not None):
+		Mask_FileName = args.mask
+		imageio.imwrite(Mask_FileName, DispLMA_Mask * 255)
+
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Compute_RMSE.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+from sklearn.metrics import mean_squared_error
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Data analysis - RMSE computation')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--pred', type=str, required=True,
+		help='Prediction TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+		help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--adjtilesdim', type=int, required=True,
+		help='Adjacent tiles dimensions (e.g. 1, 3 or 5) to exclude NaN border')
+	required.add_argument('--output', type=str, required=True,
+		help='CSV file')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--verbose', action="store_true",
+						  help='verbose mode')
+	return parser
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgPred_name = args.pred
+	imgPred_basename = os.path.basename(imgPred_name)
+	AdjacentTilesDim = args.adjtilesdim
+	imgGT_name = args.groundtruth
+	output_name = args.output
+
+	# Read Pred and GroundTruth images
+	#print("Reading Pred file...")
+	imgPred = imageio.imread(imgPred_name)
+
+
+	#print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+
+	# Remove NaN border when needed	
+	imgPred_Crop = imgPred
+	imgGT_Crop = imgGT
+	if (AdjacentTilesDim == 3):
+		Border = 1
+		imgPred_Crop = imgPred[Border:-Border,Border:-Border]
+		imgGT_Crop = imgGT[Border:-Border,Border:-Border]
+	elif (AdjacentTilesDim == 5):
+		Border = 2
+		imgPred_Crop = imgPred[Border:-Border,Border:-Border]
+		imgGT_Crop = imgGT[Border:-Border,Border:-Border]
+
+	# Quality control - NaN
+	TestNaN_imgPred_Crop = np.any(np.isnan(imgPred_Crop))
+	TestNaN_imgGT_Crop = np.any(np.isnan(imgGT_Crop))
+
+	# #  List indices with Nan values
+	# ListNaN_imgPred_Crop = np.argwhere(np.isnan(imgPred_Crop))
+	# print('ListNaN_imgPred_Crop: ', ListNaN_imgPred_Crop)
+
+	# Verbose mode
+	if args.verbose:
+		# print('imgPred type: ', imgPred.dtype)
+		print('imgPred shape: ', imgPred.shape)		
+		# print('imgGT type: ', imgGT.dtype)
+		print('imgGT shape: ', imgGT.shape)
+		print('imgPred_Crop shape: ', imgPred_Crop.shape)
+		print('imgGT_Crop shape: ', imgGT_Crop.shape)
+		print('TestNaN_imgPred_Crop: ', TestNaN_imgPred_Crop)
+		print('TestNaN_imgGT_Crop: ', TestNaN_imgGT_Crop)
+
+	# Compute RMSE
+	rmse = mean_squared_error(imgGT_Crop, imgPred_Crop, squared=False)
+	print('rmse: ',rmse)
+
+	QC_data = np.array([[imgPred_basename,rmse]])
+	Columns = ['FileName','RMSE']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	print(df.head())
+	df.to_csv(output_name, index=False)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Compute_RMSE_WithFiltering.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+from sklearn.metrics import mean_squared_error
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Data analysis - RMSE computation')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--pred', type=str, required=True,
+		help='Prediction TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+		help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--confidence', type=str, required=True,
+		help='Confidence TIFF file (single layer)')
+	required.add_argument('--disp_lma', type=str, required=True,
+		help='LMA disparity TIFF file (single layer)')
+	required.add_argument('--adjtilesdim', type=int, required=True,
+		help='Adjacent tiles dimensions (e.g. 1, 3 or 5) to exclude NaN border')
+	required.add_argument('--output', type=str, required=True,
+		help='CSV file')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--threshold', type=float, default = 0.15,
+						  help='threshold on confidence map')
+	options.add_argument('--output_mask', type=str,
+		help='output mask image (TIFF file)')
+	options.add_argument('--verbose', action="store_true",
+						  help='verbose mode')
+	return parser
+
+# Remove NaN border
+def cropping(img, border):
+	if border == 0:
+		return img
+	else:
+		img_cropped = img[border:-border,border:-border]
+		return img_cropped
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+	
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgPred_name = args.pred
+	imgPred_basename = os.path.basename(imgPred_name)
+	AdjacentTilesDim = args.adjtilesdim
+	imgGT_name = args.groundtruth
+	imgDispLMA_name = args.disp_lma
+	imgConfidence_name = args.confidence
+	output_name = args.output
+	outputmask_name = args.output_mask
+	threshold = args.threshold
+
+	# Read images
+	#print("Reading Pred file...")
+	imgPred = imageio.imread(imgPred_name)
+
+	#print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+
+	#print("Reading Confidence file...")
+	imgConfidence = imageio.imread(imgConfidence_name)
+
+	#print("Reading DispLMA file...")
+	imgDispLMA = imageio.imread(imgDispLMA_name)
+
+	# Remove NaN border when needed	
+	if (AdjacentTilesDim == 3):
+		Border = 1	
+	elif (AdjacentTilesDim == 5):
+		Border = 2
+	else:
+		Border = 0
+	imgPred_Crop = cropping(imgPred, Border)
+	imgGT_Crop = cropping(imgGT, Border)
+	imgConfidence_Crop = cropping(imgConfidence, Border)
+	imgDispLMA_Crop = cropping(imgDispLMA, Border)
+
+	# Quality control - NaN
+	TestNaN_imgPred_Crop = np.any(np.isnan(imgPred_Crop))
+	TestNaN_imgGT_Crop = np.any(np.isnan(imgGT_Crop))
+
+	# #  List indices with Nan values
+	# ListNaN_imgPred_Crop = np.argwhere(np.isnan(imgPred_Crop))
+	# print('ListNaN_imgPred_Crop: ', ListNaN_imgPred_Crop)
+
+	# Verbose mode
+	if args.verbose:
+		# print('imgPred type: ', imgPred.dtype)
+		print('imgPred shape: ', imgPred.shape)		
+		# print('imgGT type: ', imgGT.dtype)
+		print('imgGT shape: ', imgGT.shape)
+		print('imgPred_Crop shape: ', imgPred_Crop.shape)
+		print('imgGT_Crop shape: ', imgGT_Crop.shape)
+		print('TestNaN_imgPred_Crop: ', TestNaN_imgPred_Crop)
+		print('TestNaN_imgGT_Crop: ', TestNaN_imgGT_Crop)
+
+
+	# Define sample_weight using Confidence and DispLMA maps
+	imgDispLMAMask_Crop = np.uint8(np.reshape(~np.isnan(imgDispLMA_Crop), imgDispLMA_Crop.shape))
+	imgConfidenceMask_Crop = np.uint8(np.where(imgConfidence_Crop >= threshold, 1, 0))
+	imgSampleWeight_Crop = np.uint8(np.logical_and(imgDispLMAMask_Crop, imgConfidenceMask_Crop))
+
+	if outputmask_name is not None:
+		imageio.imwrite(outputmask_name, imgSampleWeight_Crop * 255)
+
+	# Compute RMSE
+	rmse = mean_squared_error(imgGT_Crop, imgPred_Crop, sample_weight=imgSampleWeight_Crop, squared=False)
+	print('rmse: ',rmse)
+
+	QC_data = np.array([[imgPred_basename,rmse]])
+	Columns = ['FileName','RMSE']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	#print(df.head())
+
+	print('Saving CSV file...')
+	df.to_csv(output_name, index=False)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Compute_RMSE_WithThreshold.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+from sklearn.metrics import mean_squared_error
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Data analysis - RMSE computation')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--pred', type=str, required=True,
+		help='Prediction TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+		help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--adjtilesdim', type=int, required=True,
+		help='Adjacent tiles dimensions (e.g. 1, 3 or 5) to exclude NaN border')
+	required.add_argument('--output', type=str, required=True,
+		help='CSV file')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--threshold', type=float, default = 2.0,
+						  help='threshold on image difference')
+	options.add_argument('--output_mask', type=str,
+		help='output mask image (TIFF file)')
+	options.add_argument('--verbose', action="store_true",
+						  help='verbose mode')
+	return parser
+
+# Remove NaN border
+def cropping(img, border):
+	if border == 0:
+		return img
+	else:
+		img_cropped = img[border:-border,border:-border]
+		return img_cropped
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+	
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgPred_name = args.pred
+	imgPred_basename = os.path.basename(imgPred_name)
+	AdjacentTilesDim = args.adjtilesdim
+	imgGT_name = args.groundtruth
+	output_name = args.output
+	outputmask_name = args.output_mask
+	threshold = args.threshold
+
+	# Read images
+	#print("Reading Pred file...")
+	imgPred = imageio.imread(imgPred_name)
+
+	#print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+
+	# Remove NaN border when needed	
+	if (AdjacentTilesDim == 3):
+		Border = 1	
+	elif (AdjacentTilesDim == 5):
+		Border = 2
+	else:
+		Border = 0
+	imgPred_Crop = cropping(imgPred, Border)
+	imgGT_Crop = cropping(imgGT, Border)
+	
+	# Quality control - NaN
+	TestNaN_imgPred_Crop = np.any(np.isnan(imgPred_Crop))
+	TestNaN_imgGT_Crop = np.any(np.isnan(imgGT_Crop))
+
+	# #  List indices with Nan values
+	# ListNaN_imgPred_Crop = np.argwhere(np.isnan(imgPred_Crop))
+	# print('ListNaN_imgPred_Crop: ', ListNaN_imgPred_Crop)
+
+	# Verbose mode
+	if args.verbose:
+		# print('imgPred type: ', imgPred.dtype)
+		print('imgPred shape: ', imgPred.shape)		
+		# print('imgGT type: ', imgGT.dtype)
+		print('imgGT shape: ', imgGT.shape)
+		print('imgPred_Crop shape: ', imgPred_Crop.shape)
+		print('imgGT_Crop shape: ', imgGT_Crop.shape)
+		print('TestNaN_imgPred_Crop: ', TestNaN_imgPred_Crop)
+		print('TestNaN_imgGT_Crop: ', TestNaN_imgGT_Crop)
+
+
+	# Define sample_weight, where absolute difference is below threshold (2 pixels), to remove outliers
+	imgAbsDiff_Crop = np.absolute(imgPred_Crop - imgGT_Crop)
+	imgSampleWeight_Crop = np.uint8(np.where(imgAbsDiff_Crop < threshold, 1, 0))
+	
+	if outputmask_name is not None:
+		imageio.imwrite(outputmask_name, imgSampleWeight_Crop * 255)
+
+	# Compute RMSE
+	rmse = mean_squared_error(imgGT_Crop, imgPred_Crop, sample_weight=imgSampleWeight_Crop, squared=False)
+	print('rmse: ',rmse)
+
+	QC_data = np.array([[imgPred_basename,rmse]])
+	Columns = ['FileName','RMSE_WithThreshold']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	#print(df.head())
+
+	print('Saving CSV file...')
+	df.to_csv(output_name, index=False)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

Preprocessing_CombinedImages.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+
+import sys
+import argparse
+import time 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Combine images to generate 3D volume')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--corr', type=str, required=True,
+						  help='2D Corr TIFF file (multi-layer)')
+	required.add_argument('--targetdisp', type=str, required=True,
+						  help='Target Disparity TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+						  help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--confidence', type=str, required=True,
+						  help='Confidence TIFF file (single layer)')
+	required.add_argument('--disp_lma', type=str, required=True,
+						  help='LMA Disparity TIFF file (single layer)')
+	required.add_argument('--output', type=str, required=True,
+						  help='Combined TIFF file (multi-layer)')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--repeat', type=int, default=15,
+						  help='Repeat tile number (default 15)')
+	return parser
+
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	img2DCorr_name = args.corr
+	imgTargetDisp_name = args.targetdisp
+	imgGT_name = args.groundtruth
+	imgConfidence_name = args.confidence
+	imgDispLMA_name = args.disp_lma
+	output_name = args.output
+	repeatNb = args.repeat
+	
+	# Read 2dcorr image - all layers
+	print("Reading 2dcorr file...")
+	img2DCorr = imageio.mimread(img2DCorr_name,memtest=False)
+	img2DCorr = np.array(img2DCorr)
+	# print('\nimg2DCorr type: ', img2DCorr.dtype)
+	# print('img2DCorr shape: ', img2DCorr.shape)
+
+	# Read TargetDisp and GroundTruth images
+	print("Reading TargetDisp file...")
+	imgTargetDisp = imageio.imread(imgTargetDisp_name)
+	# print('imgTargetDisp type: ', imgTargetDisp.dtype)
+	# print('imgTargetDisp shape: ', imgTargetDisp.shape)
+
+	print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+	# print('imgGT type: ', imgGT.dtype)
+	# print('imgGT shape: ', imgGT.shape)
+
+	print("Reading Confidence file...")
+	imgConfidence = imageio.imread(imgConfidence_name)
+	# print('imgConfidence type: ', imgConfidence.dtype)
+	# print('imgConfidence shape: ', imgConfidence.shape)
+
+	print("Reading DispLMA file...")
+	imgDispLMA = imageio.imread(imgDispLMA_name)
+	# print('imgDispLMA type: ', imgDispLMA.dtype)
+	# print('imgDispLMA shape: ', imgDispLMA.shape)
+
+	# - - - - - - - - - - -
+	print("Generating combined image...")
+	# Resample TargetDisp and imgTG (repeating values, to match img2DCor size)
+	imgTargetDisp_repeat0 = np.repeat(imgTargetDisp, repeatNb, axis=0)
+	imgTargetDisp_repeat = np.repeat(imgTargetDisp_repeat0, repeatNb, axis=1)
+	imgTargetDisp_repeat = np.expand_dims(imgTargetDisp_repeat, axis=0)
+
+	imgGT_repeat0 = np.repeat(imgGT, repeatNb, axis=0)
+	imgGT_repeat = np.repeat(imgGT_repeat0, repeatNb, axis=1)
+	imgGT_repeat = np.expand_dims(imgGT_repeat, axis=0)
+	# print('imgTargetDisp_repeat shape: ', imgTargetDisp_repeat.shape)
+	# print('imgGT_repeat shape: ', imgGT_repeat.shape)
+
+	imgConfidence_repeat0 = np.repeat(imgConfidence, repeatNb, axis=0)
+	imgConfidence_repeat = np.repeat(imgConfidence_repeat0, repeatNb, axis=1)
+	imgConfidence_repeat = np.expand_dims(imgConfidence_repeat, axis=0)
+	print('\t imgConfidence_repeat shape: ', imgConfidence_repeat.shape)
+
+	imgDispLMA_repeat0 = np.repeat(imgDispLMA, repeatNb, axis=0)
+	imgDispLMA_repeat = np.repeat(imgDispLMA_repeat0, repeatNb, axis=1)
+	imgDispLMA_repeat = np.expand_dims(imgDispLMA_repeat, axis=0)
+	print('\t imgDispLMA_repeat shape: ', imgDispLMA_repeat.shape)
+
+	# Stack layers to img2dCorr, to generate one 3D volume
+	imgAll = np.concatenate((img2DCorr,imgTargetDisp_repeat), axis = 0)
+	imgAll = np.concatenate((imgAll,imgGT_repeat), axis = 0)
+	imgAll = np.concatenate((imgAll,imgConfidence_repeat), axis = 0)
+	imgAll = np.concatenate((imgAll,imgDispLMA_repeat), axis = 0)
+	
+	imageio.mimwrite(output_name,imgAll)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	print("Reading Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

QualityControl_Confidence.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Quality control')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--confidence', type=str, required=True,
+						  help='2D confidence TIFF file (single layer)')
+	required.add_argument('--output', type=str, required=True,
+						  help='CSV file')
+	return parser
+
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	imgConfidence_name = args.confidence
+	imgConfidence_basename = os.path.basename(imgConfidence_name)
+	output_name = args.output
+
+	# Read confidence image - all layers
+	#print("Reading confidence file...")
+	imgConfidence = imageio.imread(imgConfidence_name)
+	print('\nimgConfidence type: ', imgConfidence.dtype)
+	print('imgConfidence shape: ', imgConfidence.shape)
+
+	Bool_NaN = np.isnan(imgConfidence)
+	Nb_NaN = np.count_nonzero(Bool_NaN)
+	imgConfidence_NotNaN = imgConfidence[~np.isnan(imgConfidence)]
+	print('\nNb_NaN: ', Nb_NaN)
+	print('\nimgConfidence_NotNaN type: ', imgConfidence_NotNaN.dtype)
+	print('imgConfidence_NotNaN shape: ', imgConfidence_NotNaN.shape)
+
+	# Compute Image info: mean,min,max values
+	imgConfidence_mean = np.mean(imgConfidence_NotNaN)
+	imgConfidence_min = np.min(imgConfidence_NotNaN)
+	imgConfidence_max = np.max(imgConfidence_NotNaN)	
+	print('\nimgConfidence_mean : ', imgConfidence_mean)
+	print('imgConfidence_min : ', imgConfidence_min)
+	print('imgConfidence_max : ', imgConfidence_max)
+	
+	# Save information to CSV file
+	QC_data = np.array([[imgConfidence_basename,imgConfidence.shape[0],imgConfidence.shape[1], \
+		Nb_NaN, imgConfidence_mean, imgConfidence_min, imgConfidence_max]
+	])
+	Columns = ['FileName','imgConfidence_Shape0','imgConfidence_Shape1', \
+	'Nb_NaN','imgConfidence_mean','imgConfidence_min','imgConfidence_max']
+	df = pd.DataFrame(QC_data,columns=Columns)
+	print(df.head())
+	df.to_csv(output_name, index=False)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

QualityControl_Images.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+import pandas as pd
+
+import sys
+import argparse
+import time 
+import os 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Quality control')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--corr', type=str, required=True,
+						  help='2D Corr TIFF file (multi-layer)')
+	required.add_argument('--targetdisp', type=str, required=True,
+						  help='Target Disparity TIFF file (single layer)')
+	required.add_argument('--groundtruth', type=str, required=True,
+						  help='Ground Truth TIFF file (single layer)')
+	required.add_argument('--output', type=str, required=True,
+						  help='CSV file')
+	return parser
+
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	img2DCorr_name = args.corr
+	img2DCorr_basename = os.path.basename(img2DCorr_name)
+	imgTargetDisp_name = args.targetdisp
+	imgGT_name = args.groundtruth
+	output_name = args.output
+
+	# Read 2dcorr image - all layers
+	#print("Reading 2dcorr file...")
+	img2DCorr = imageio.mimread(img2DCorr_name,memtest=False)
+	img2DCorr = np.array(img2DCorr)
+	# print('\nimg2DCorr type: ', img2DCorr.dtype)
+	#print('img2DCorr shape: ', img2DCorr.shape)
+
+	# Read TargetDisp and GroundTruth images
+	#print("Reading TargetDisp file...")
+	imgTargetDisp = imageio.imread(imgTargetDisp_name)
+	# print('imgTargetDisp type: ', imgTargetDisp.dtype)
+	#print('imgTargetDisp shape: ', imgTargetDisp.shape)
+
+	#print("Reading GroundTruth file...")
+	imgGT = imageio.imread(imgGT_name)
+	# print('imgGT type: ', imgGT.dtype)
+	#print('imgGT shape: ', imgGT.shape)
+
+	# Compute Image info: mean,min,max values
+	img2DCorr_mean = np.mean(img2DCorr)
+	img2DCorr_min = np.min(img2DCorr)
+	img2DCorr_max = np.max(img2DCorr)
+	
+	imgTargetDisp_mean = np.mean(imgTargetDisp)
+	imgTargetDisp_min = np.min(imgTargetDisp)
+	imgTargetDisp_max = np.max(imgTargetDisp)
+	
+	imgGT_mean = np.mean(imgGT)
+	imgGT_min = np.min(imgGT)
+	imgGT_max = np.max(imgGT)
+	# print('\n img2DCorr pixel info:')
+	# print('\t img2DCorr_mean: ', img2DCorr_mean)
+	# print('\t img2DCorr_min: ', img2DCorr_min)
+	# print('\t img2DCorr_max: ', img2DCorr_max)
+
+	# Compute image difference between TargetDisp and GroundTruth images
+	Diff = np.abs(imgTargetDisp - imgGT)
+	Diff_mean = np.mean(Diff)
+	Diff_min = np.min(Diff)
+	Diff_max = np.max(Diff)
+	# print('Image difference:')
+	# print('\t Diff_mean:',Diff_mean )
+	# print('\t Diff_min:', Diff_min)
+	# print('\t Diff_max:', Diff_max)
+	
+
+	# Save information to CSV file
+	QC_data = np.array([[img2DCorr_basename,img2DCorr.shape[0],img2DCorr.shape[1], img2DCorr.shape[2], \
+		imgTargetDisp.shape[0], imgTargetDisp.shape[1], imgGT.shape[0], imgGT.shape[1], \
+		img2DCorr_mean, img2DCorr_min, img2DCorr_max, \
+		imgTargetDisp_mean, imgTargetDisp_min, imgTargetDisp_max, \
+		imgGT_mean, imgGT_min, imgGT_max, \
+		Diff_mean, Diff_min, Diff_max]
+	])
+	Columns = ['FileName','img2DCorr_Shape0','img2DCorr_Shape1','img2DCorr_Shape2',\
+	'imgTargetDisp_Shape0','imgTargetDisp_Shape1',\
+	'imgGT_Shape0','imgGT_Shape1',
+	'img2DCorr_mean','img2DCorr_min','img2DCorr_max',\
+	'imgTargetDisp_mean','imgTargetDisp_min','imgTargetDisp_max',\
+	'imgGT_mean','imgGT_min','imgGT_max',\
+	'Diff_mean','Diff_min','Diff_max',\
+	]
+	df = pd.DataFrame(QC_data,columns=Columns)
+	print(df.head())
+	df.to_csv(output_name, index=False)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	#print("Computing Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+

README.txt

+This project contains various python scripts used for disparity map analysis
+
+
+## 1. Data pre-processing stage
+
+Note: Data preprocessing stage to generate input files for AI analysis via ir-tp-net
+
+### 1.1. Extracting images of interest from multi-layer 3D TIFF files
+
+Example:
+''' 
+python3 ./Preprocessing_2DPhaseCorrelation.py --crop_border --input $i --targetdisp $TargetDispImage \
+		--groundtruth $GroundTruthImage --confidence $ConfidenceImage --disp_lma $DispLMAImage --corr $CorrImage --verbose
+'''
+
+Notes:
+ - Input file is a special multi-layer TIFF file
+ - Output CorrImage is a 3D TIFF file with 120 layers
+ - Other output files are 2D images
+
+### 1.2. Generating 3D TIFF files as direct input files to neural network ir-tp-net
+
+Example:
+''' 
+python3 ./Preprocessing_CombinedImages.py --corr $CorrImage --targetdisp $TargetDispImage \
+		--groundtruth $GroundTruthImage --confidence $ConfidenceImage --disp_lma $DispLMAImage --output $CombinedImage
+''' 
+Notes: 
+ - Output file is a 3D image with 124 layers
+  - scaled to the size of the correlation image
+
+## 2. AI analysis stage
+
+Please see "ir-tp-net" project for neural network training and testing to generate predicted disparity map
+
+
+## 3. Data post-processing stage
+
+Note: Data postprocessing stage to analyze outputs from AI analysis
+
+###  3.1. Density analysis
+
+Examples:
+''' 
+python3 ./Compute_Density.py --pred $i --groundtruth $GroundTruthImage --adjtilesdim 1 --output $Density_CSVFile --inclusionmask $InclusionMask_File --exclusionmask $ExclusionMask_File --threshold 2.0 --verbose
+''' 
+
+Note: input file is the predicted disparity map
+
+###  3.2. RMSE analysis
+
+Example:
+''' 
+python3 ./Compute_RMSE_WithThreshold.py --pred $i --groundtruth $GroundTruthImage --adjtilesdim $AdjTilesDim --threshold $RMSE_Threshold --output $RMSE_CSVFile
+			
+python3 ./Compute_RMSE_WithFiltering.py --pred $i --groundtruth $GroundTruthImage --confidence $ConfidenceImage --disp_lma $DispLMAImage --adjtilesdim 1 --threshold $Threshold --output $RMSE_CSVFile --output_mask $RMSEFiltering_MaskFile
+''' 
+
+Note: input file is the predicted disparity map
+
+
+## 4. Data Quality control
+
+Quality control stage generating multi-layer 3D tiff file
+
+''' 
+python3 ./Compute_Inference_QCImage.py --pred $i --groundtruth $GroundTruthImage --targetdisp $TargetDispImage --mask $DataFilteringMask --threshold 2.0 --output $InferenceQC_File --verbose
+''' 
+
+Notes: 
+ - input file is the predicted disparity map
+ - output file includes predicted disparity map, ground truth map, target disparity map and mask map

RandomNoiseLevel_ImageCreation.py

+#!/usr/bin/env python3
+
+
+import imageio 
+import numpy as np
+
+import sys
+import argparse
+import time 
+
+def arg_parser():
+	parser = argparse.ArgumentParser(description='Generate combined image with random noise level')
+	required = parser.add_argument_group('Required')
+	required.add_argument('--template', type=str, required=True,
+						  help='Combined TIFF file (multi-layer)')
+	required.add_argument('--offsetTemplate', type=float, required=True,
+						  help='Template noise offset level')
+	required.add_argument('--noiseLevel', type=float, required=True,
+						  help='Noise offset level')
+	required.add_argument('--output', type=str, required=True,
+						  help='Output combined TIFF file (multi-layer)')
+	options = parser.add_argument_group('Options')
+	options.add_argument('--tileSize', type=int, default=15,
+						  help='Tile Size')
+	options.add_argument('--outputTargetDisp', type=str,
+						  help='Output TargetDisp (2D image)')
+	options.add_argument('--outputGroundTruth', type=str,
+						  help='Output GroundTruth (2D image)')
+	return parser
+
+
+# Steps
+# Option to define offset noise level
+# Assess offset images based on noise level
+# Load all images as list
+# for each tile, pick among available images
+# Save image
+
+#MAIN
+def main(args=None):
+	args = arg_parser().parse_args(args)
+
+	Time1 = time.time()
+	#print(bcolors.BOLDWHITE+"Time1: "+str(Time1)+bcolors.ENDC)
+
+	template_name = args.template
+	offset_template = args.offsetTemplate
+	noise_level = args.noiseLevel
+	output_name = args.output
+	tile_size = args.tileSize
+	output_TargetDisp = args.outputTargetDisp
+	output_GroundTruth = args.outputGroundTruth
+
+	# Offsets
+	List_AvailableOffsets = np.array([-5.000, -4.003, -3.116, -2.341, -1.676, -1.122, -0.679, -0.346, -0.125, -0.014, \
+		0.014, 0.125, 0.346, 0.679, 1.122, 1.676, 2.341, 3.116, 4.003, 5.000])
+	#print('List_AvailableOffsets',List_AvailableOffsets)
+
+	# Filter offsets based on offsetLevel
+	List_Offsets = List_AvailableOffsets[np.logical_and(List_AvailableOffsets >= -noise_level,List_AvailableOffsets <= noise_level)]
+	print('\n List_Offsets',List_Offsets)
+
+	# Nb images
+	nb_images = len(List_Offsets)
+	print('nb_images',nb_images)
+
+	# Read template combined image
+	print("\n Reading template file...")
+	imgTemplate_layerList = imageio.mimread(template_name,memtest=False)
+	imgTemplate = np.stack(imgTemplate_layerList, axis=0)
+	print('\t imgTemplate type: ', imgTemplate.dtype)
+	print('\t imgTemplate shape: ', imgTemplate.shape)
+
+	print("\n Creating image stack...")
+	image_stack = np.zeros((nb_images, imgTemplate.shape[0], imgTemplate.shape[1], imgTemplate.shape[2]), dtype = imgTemplate.dtype)
+	# print('image_stack type: ', image_stack.dtype)
+	# print('image_stack shape: ', image_stack.shape)
+
+	# List_ImageNames
+	List_ImageNames = []
+	for i, offset in enumerate(List_Offsets):
+		print('\t Reading Image {} with offset {:.3f} ...'.format(i,offset))
+		offset_string = "{:.3f}".format(offset)
+		FileName_current = template_name.replace(str(offset_template),offset_string)
+		List_ImageNames.append(FileName_current)
+		img_current_layerList = imageio.mimread(FileName_current,memtest=False)
+		img_current = np.stack(img_current_layerList, axis=0)
+		image_stack[i,...] = img_current
+	#print('List_ImageNames',List_ImageNames)
+
+
+	# Generating image stack
+	# image_stack = np.stack(List_Images, axis=0)
+	print('\t image_stack type: ', image_stack.dtype)
+	print('\t image_stack shape: ', image_stack.shape)
+
+	# Output image
+	print("\n Generating output image by random input patch selection...")
+	imgOutput = np.zeros(imgTemplate.shape, dtype = imgTemplate.dtype)
+	print('\t imgOutput type: ', imgOutput.dtype)
+	print('\t imgOutput shape: ', imgOutput.shape)
+
+	i = j = 0
+	for i in range(0,imgTemplate.shape[1], tile_size):
+		for j in range(0,imgTemplate.shape[2], tile_size):
+			random_selection = np.random.randint(nb_images)
+			#print('i j rnd: {} {} {}'.format(i,j,random_selection))
+			#imgOutput[:,i:i+tile_size,j:j+tile_size] = tile_size * patchNb
+			imgOutput[:,i:i+tile_size,j:j+tile_size] = image_stack[random_selection,:,i:i+tile_size,j:j+tile_size]
+
+	imgOutput_TargetDisp = imgOutput[-2,:,:]
+	imgOutput_TargetDisp = imgOutput_TargetDisp[::tile_size,::tile_size]
+	print('\t imgOutput_TargetDisp shape: ', imgOutput_TargetDisp.shape)
+
+	imgOutput_GroundTruth = imgOutput[-1,:,:]
+	imgOutput_GroundTruth = imgOutput_GroundTruth[::tile_size,::tile_size]
+	print('\t imgOutput_GroundTruth shape: ', imgOutput_GroundTruth.shape)
+
+	# Write output combined image (multi-layer)
+	print("\n Writing output file - combined image...")
+	imageio.mimwrite(output_name,imgOutput)
+
+	# Write output TargetDisp image (2D)
+	if (output_TargetDisp is not None):
+		print("\nWriting output file - TargetDisp image...")
+		imageio.imwrite(output_TargetDisp,imgOutput_TargetDisp)
+
+	# Write output GroundTruth image (2D)
+	if (output_GroundTruth is not None):
+		print("\nWriting output file - GroundTruth image...")
+		imageio.imwrite(output_GroundTruth,imgOutput_GroundTruth)
+
+	Time2 = time.time()
+	#print(bcolors.BOLDWHITE+"Time2: "+str(Time2)+bcolors.ENDC)
+	TimeDiff = Time2 - Time1
+	print("Reading Time: "+str(TimeDiff))
+
+if __name__ == '__main__':
+	sys.exit(main(sys.argv[1:]))
+