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.'
# Append array
HDF5Datas.append(DataPatch)
HDF5Labels.append(LabelPatch)
# List type to array numpy
HDF5Datas = np.asarray(HDF5Datas).reshape(-1, PatchSize, PatchSize,
NormalizedT1wImage = T1wImage/MaxValue
# Shave region outside
print 'Remove the region outside the brain with the value of ', args.thresholdvalue
darkRegionValue = args.thresholdvalue
darkRegionBox = np.where(NormalizedT1wImage>darkRegionValue)
border = ((np.min(darkRegionBox[0]),np.max(darkRegionBox[0])),
(np.min(darkRegionBox[1]),np.max(darkRegionBox[1])),
(np.min(darkRegionBox[2]),np.max(darkRegionBox[2])))
DatasT1wImage = NormalizedT1wImage[border[0][0]:border[0][1],
border[1][0]:border[1][1],
border[2][0]:border[2][1]]
# Extract 3D patches
DatasT1wPatch = array_to_patches(DatasT1wImage,
patch_shape=(PatchSize,PatchSize,PatchSize),
extraction_step = args.stride ,
normalization=False)
print 'for patches of training phase.'
# n-dimensional Caffe supports data's form : [numberOfBatches,channels,heigh,width,depth]
# Add channel axis !
DatasT1wPatch = DatasT1wPatch[:,np.newaxis,:,:,:]
# Rearrange
RandomOrder = np.random.permutation(DatasT1wPatch.shape[0])
DatasT1wPatch = DatasT1wPatch[RandomOrder,:,:,:,:]
# Crop data to desired number of samples
if args.samples :
DatasT1wPatch = DatasT1wPatch[:args.samples ,...]