def example_parallel_2d():
# ------------------ Declare Parameters ------------------
# Volume Parameters:
volume_size = 64
volume_shape = [volume_size, volume_size]
volume_spacing = [1, 1]
# Detector Parameters:
detector_shape = 100
detector_spacing = 1
# Trajectory Parameters:
number_of_projections = 90
angular_range = np.pi
# create Geometry class
geometry = GeometryParallel2D(volume_shape, volume_spacing, detector_shape, detector_spacing, number_of_projections, angular_range)
geometry.set_trajectory(circular_trajectory.circular_trajectory_2d(geometry))
# Get Phantom
phantom = shepp_logan.shepp_logan_enhanced(volume_shape).astype(np.float32)
# Add required batch dimension
phantom = np.expand_dims(phantom, axis=0)
sino = parallel_projection2d(phantom,geometry)
@tf.function
def test_func_proj(x):
return parallel_projection2d(x,geometry)
@tf.function
def test_func_reco(x):
return parallel_backprojection2d(x,geometry)
proj_theoretical, proj_numerical = tf.test.compute_gradient(test_func_proj, [sino])
reco_theoretical, reco_numerical = tf.test.compute_gradient(test_func_reco, [sino])
def bench_parallel_backward(phantom, det_count, num_angles, warmup, repeats):
geometry = create_parallel_geometry(det_count, num_angles)
sino = parallel_projection2d(phantom, geometry)
f = lambda x: parallel_backprojection2d(x, geometry)
return benchmark(f, sino, warmup, repeats)
def forwardproject(self, imgs: TensorLike):
imgs_tf = tf.convert_to_tensor(imgs, dtype=tf.float32)
# This row is needed if rank of output is not changed in parallel_projection2d gradient.
imgs_tf = tf.reshape(imgs_tf, shape=tf.shape(imgs_tf)[:3])
sinos_3D = parallel_projection2d(
imgs_tf, self._geometry) # removes channeldimension
sinos_4D = tf.expand_dims(sinos_3D, axis=-1)
return sinos_4D
def bench_parallel_backward(batch, size, det_count, num_angles, *bench_args):
with tf.device('/GPU:0'):
phantom = tf.random.normal((batch, size, size))
geometry = create_parallel_geometry(size, det_count, num_angles)
sino = parallel_projection2d(phantom, geometry)
def f(x):
return parallel_backprojection2d(x, geometry)
return benchmark(f, sino, *bench_args)
def example_parallel_2d():
# ------------------ Declare Parameters ------------------
# Volume Parameters:
volume_size = 256
volume_shape = [volume_size, volume_size]
volume_spacing = [1, 1]
# Detector Parameters:
detector_shape = 800
detector_spacing = 1
# Trajectory Parameters:
number_of_projections = 360
angular_range = 2 * np.pi
# create Geometry class
geometry = GeometryParallel2D(volume_shape, volume_spacing, detector_shape,
detector_spacing, number_of_projections,
angular_range)
geometry.set_ray_vectors(
circular_trajectory.circular_trajectory_2d(geometry))
# Get Phantom
phantom = shepp_logan.shepp_logan_enhanced(volume_shape)
phantom = np.expand_dims(phantom, axis=0)
# ------------------ Call Layers ------------------
with tf.Session() as sess:
result = parallel_projection2d(phantom, geometry)
sinogram = result.eval()
#sinogram = sinogram + np.random.normal(
# loc=np.mean(np.abs(sinogram)), scale=np.std(sinogram), size=sinogram.shape) * 0.02
reco_filter = filters.ram_lak_2D(geometry)
sino_freq = np.fft.fft(sinogram, axis=1)
sino_filtered_freq = np.multiply(sino_freq, reco_filter)
sinogram_filtered = np.fft.ifft(sino_filtered_freq, axis=1)
result_back_proj = parallel_backprojection2d(sinogram_filtered,
geometry)
reco = result_back_proj.eval()
plt.figure()
plt.imshow(np.squeeze(reco), cmap=plt.get_cmap('gist_gray'))
plt.axis('off')
plt.savefig('2d_par_reco.png',
dpi=150,
transparent=False,
bbox_inches='tight')
def example_parallel_2d():
# ------------------ Declare Parameters ------------------
# Volume Parameters:
volume_size = 256
volume_shape = [volume_size, volume_size]
volume_spacing = [1, 1]
# Detector Parameters:
detector_shape = 800
detector_spacing = 1
# Trajectory Parameters:
number_of_projections = 360
angular_range = 2 * np.pi
# create Geometry class
geometry = GeometryParallel2D(volume_shape, volume_spacing, detector_shape,
detector_spacing, number_of_projections,
angular_range)
geometry.set_trajectory(
circular_trajectory.circular_trajectory_2d(geometry))
# Get Phantom
phantom = shepp_logan.shepp_logan_enhanced(volume_shape)
# Add required batch dimension
phantom = np.expand_dims(phantom, axis=0)
# ------------------ Call Layers ------------------
sinogram = parallel_projection2d(phantom, geometry)
#sinogram = sinogram + np.random.normal(
# loc=np.mean(np.abs(sinogram)), scale=np.std(sinogram), size=sinogram.shape) * 0.02
reco_filter = filters.ram_lak_2D(geometry)
sino_freq = tf.signal.fft(tf.cast(sinogram, dtype=tf.complex64))
sino_filtered_freq = tf.multiply(sino_freq,
tf.cast(reco_filter, dtype=tf.complex64))
sinogram_filtered = tf.math.real(tf.signal.ifft(sino_filtered_freq))
reco = parallel_backprojection2d(sinogram_filtered, geometry)
plt.figure()
plt.imshow(np.squeeze(reco), cmap=plt.get_cmap('gist_gray'))
plt.axis('off')
plt.savefig('2d_par_reco.png',
dpi=150,
transparent=False,
bbox_inches='tight')
angular_range_inpainted)
geometry_inpainted.set_trajectory(
circular_trajectory.circular_trajectory_2d(geometry_inpainted))
dataset_paths = [x[2] for x in os.walk('./train_dataset')]
dataset_paths = [args.tfolder + s for s in dataset_paths[0]]
current_it = 0
for epoch in range(number_epochs):
for dataset_path in dataset_paths:
#Get and prepare training array
X = prepare_training_array(dataset_path, img_size)
for idx in range(X.shape[0]):
if idx == 0:
inpainted_sinogram_0 = parallel_projection2d(
np.expand_dims(X[0], 0), geometry_inpainted)
else:
inpainted_sinogram_0 = parallel_projection2d(
np.expand_dims(X[idx - 1], 0), geometry_inpainted)
inpainted_sinogram_0 = tf.keras.layers.LayerNormalization()(
inpainted_sinogram_0)
inpainted_sinogram_0 = tf.expand_dims(inpainted_sinogram_0,
axis=-1)
inpainted_sinogram_1 = parallel_projection2d(
np.expand_dims(X[idx], axis=0), geometry_inpainted)
inpainted_sinogram_1 = tf.keras.layers.LayerNormalization()(
inpainted_sinogram_1)
inpainted_sinogram_1 = tf.expand_dims(inpainted_sinogram_1,
axis=-1)
def model(self, input_volume):
self.updated_reco = tf.add(input_volume, self.reco)
self.current_sino = projection_2d.parallel_projection2d(self.updated_reco, self.geometry)
return self.current_sino, self.reco
def test_func_proj(x):
return parallel_projection2d(x,geometry)
def call(self, x):
self.updated_reco = tf.add(x, self.reco)
self.current_sino = projection_2d.parallel_projection2d(
self.updated_reco, self.geometry)
return self.current_sino, self.reco
circular_trajectory.circular_trajectory_2d(geometry_inpainted))
reco_filter = filters.ram_lak_2D(geometry)
epoch = 0
for dataset_path in [
args.recon,
]:
#['./test_dataset/series_5-area_0_SIRT30i_2n.h5', ]:# ['./test_dataset/20170710.h5', ] #['./test_dataset/clean4test.h5', ] ['./test_dataset/phantom_00016_recon.h5', ] ['./dataset/TIQ.h5', ]
#Get and prepare training array
X = prepare_training_array(dataset_path, img_size)
for idx in range(0, X.shape[0], 1): #for biological TEM tomo
# for idx in range(X.shape[0]):
tf.print('Proccessed_index:', idx, output_stream=sys.stdout)
if idx == 0:
original_sinogram_0 = parallel_projection2d(
np.expand_dims(X[0], axis=0), geometry)
inpainted_sinogram_0 = parallel_projection2d(
np.expand_dims(X[0], 0), geometry_inpainted)
else:
original_sinogram_0 = parallel_projection2d(
np.expand_dims(X[idx - 1], axis=0), geometry)
inpainted_sinogram_0 = parallel_projection2d(
np.expand_dims(X[idx - 1], 0), geometry_inpainted)
original_sinogram_0 = tf.keras.layers.LayerNormalization()(
original_sinogram_0)
inpainted_sinogram_0 = tf.keras.layers.LayerNormalization()(
inpainted_sinogram_0)
original_sinogram_0 = tf.expand_dims(original_sinogram_0, axis=-1)
def f(x):
return parallel_projection2d(x, geometry)
