# Compute scale factor:
TargetShape = ReferenceImage.shape
InputShape = InputImage.shape
scale = (Targetidx / inputidx
for Targetidx, inputidx in zip(TargetShape, InputShape))
print 'With respect to scale factor x', scale
# Cubic Interpolation
if TargetShape != InputShape:
InputImage = scipy.ndimage.zoom(InputImage,
zoom=scale,
order=args.order)
# Shave border
LabelImage = shave3D(ReferenceImage, border)
DataImage = shave3D(InputImage, border)
# Extract 3D patches
DataPatch = array_to_patches(DataImage,
patch_shape=(PatchSize, PatchSize,
PatchSize),
extraction_step=args.stride,
normalization=False)
print 'for the (interpolated or not) low-resolution patches of training phase.'
LabelPatch = array_to_patches(LabelImage,
patch_shape=(PatchSize, PatchSize,
PatchSize),
extraction_step=args.stride,
normalization=False)
print 'for the reference high-resolution patches of training phase.'
LowResolutionImage = scipy.ndimage.zoom(BlurReferenceImage,
zoom = (1/float(idxScale) for idxScale in UpScale),
order = 0)
# Normalization by the max valeur of LR image
MaxValue = np.max(LowResolutionImage)
NormalizedReferenceImage = ReferenceImage/MaxValue
NormalizedLowResolutionImage = LowResolutionImage/MaxValue
# Cubic Interpolation
InterpolatedImage = scipy.ndimage.zoom(NormalizedLowResolutionImage,
zoom = UpScale,
order = args.order)
# Shave border
LabelImage = shave3D(NormalizedReferenceImage, border)
DataImage = shave3D(InterpolatedImage, border)
# Extract 3D patches
print 'Generating training patches with the resolution of ', NewResolution, ' : '
DataPatch = array_to_patches(DataImage,
patch_shape=(PatchSize,PatchSize,PatchSize),
extraction_step = args.stride ,
normalization=False)
print 'for the interpolated low-resolution patches of training phase.'
LabelPatch = array_to_patches(LabelImage,
patch_shape=(PatchSize,PatchSize,PatchSize),
extraction_step = args.stride ,
normalization=False)
print 'for the reference high-resolution patches of training phase.'
BlurReferenceImage = modcrop3D(BlurReferenceImage, scale)
ReferenceImage = modcrop3D(ReferenceImage, scale)
# Downsampling
LowResolutionImage = scipy.ndimage.zoom(
BlurReferenceImage,
zoom=(1 / float(idxScale) for idxScale in scale),
order=args.order)
# Cubic Interpolation
InterpolatedImage = scipy.ndimage.zoom(LowResolutionImage,
zoom=scale,
order=args.order)
# Shave border
LabelImage = shave3D(ReferenceImage, border)
DataImage = shave3D(InterpolatedImage, border)
# Extract 3D patches
DataPatch = array_to_patches(DataImage,
patch_shape=(PatchSize, PatchSize,
PatchSize),
extraction_step=args.stride,
normalization=False)
print 'for the interpolated low-resolution patches of training phase.'
LabelPatch = array_to_patches(LabelImage,
patch_shape=(PatchSize, PatchSize,
PatchSize),
extraction_step=args.stride,
normalization=False)
print 'for the reference high-resolution patches of training phase.'
# Loading weights
UnSynGAN_test = UnSynGAN_test(args.weights,T1toT2)
for i in range(0,len(args.test)):
# Read low-resolution image
TestFile = args.test[i]
print 'Processing testing image : ', TestFile
TestNifti = sitk.ReadImage(TestFile)
TestImage = np.swapaxes(sitk.GetArrayFromImage(TestNifti),0,2).astype('float32')
TestImageMinValue = float(np.min(TestImage))
TestImageMaxValue = float(np.max(TestImage))
TestImageNorm = TestImage/TestImageMaxValue
# Shave border
ShavedImage = shave3D(TestImage, border)
# GAN
print "Testing : ",
EstimatedImage = UnSynGAN_test.test_by_patch(ShavedImage,step=args.step)
# Padding
pad_border = [(idx,idx) for idx in border]
PaddedEstimatedImage = np.pad(EstimatedImage,pad_border,'constant')
# Synthetic image
EstimatedHRImageInverseNorm = PaddedEstimatedImage*TestImageMaxValue
EstimatedHRImageInverseNorm[EstimatedHRImageInverseNorm <= TestImageMinValue] = TestImageMinValue # Clear negative value
OutputImage = sitk.GetImageFromArray(np.swapaxes(EstimatedHRImageInverseNorm,0,2))
OutputImage.SetSpacing(TestNifti.GetSpacing())
OutputImage.SetOrigin(TestNifti.GetOrigin())
print 'Processing testing image : ', TestFile
TestNifti = sitk.ReadImage(TestFile)
TestImage = np.swapaxes(sitk.GetArrayFromImage(TestNifti),0,2).astype('float32')
TestImageMinValue = float(np.min(TestImage))
TestImageMaxValue = float(np.max(TestImage))
TestImageNorm = TestImage/TestImageMaxValue
# Check scale factor type
UpScale = tuple(itema/itemb for itema,itemb in zip(TestNifti.GetSpacing(),NewResolution))
# spline interpolation
InterpolatedImage = scipy.ndimage.zoom(TestImageNorm,
zoom = UpScale,
order = args.order)
# Shave border
ShavedInterpolatedImage = shave3D(InterpolatedImage, border)
# SRResidualCNN3D
SRReCNN3D_test = SRReCNN3D_test()
EstimatedHRImage = SRReCNN3D_test.test(ShavedInterpolatedImage,weights,
NetDepth = args.netdepth,
NetNumKernel = args.numkernel,
KernelSize = args.kernelsize,
Residual= residual)
# Padding
pad_border = [(idx,idx) for idx in border]
PaddedEstimatedHRImage = np.pad(EstimatedHRImage,pad_border,'constant')