def _varconvolve_2d_parametrized(image, parameters, kernel_name, sampler=None, queue=None,
out=None, block=False):
"""Variable convolution of *image* with *parameters*, use OpoenCL kernel *kernel_name*,
*sampler*, *queue*, *out* and wait if *block* is True. Return *out*.
"""
if queue is None:
queue = cfg.OPENCL.queue
if out is None:
out = cl.array.Array(queue, image.shape, dtype=cfg.PRECISION.np_float)
if sampler is None:
sampler = cl.Sampler(queue.context, False, cl.addressing_mode.CLAMP_TO_EDGE,
cl.filter_mode.NEAREST)
if not isinstance(parameters, cl_array.Array):
params_host = np.empty(parameters[0].shape, dtype=cfg.PRECISION.vfloat2)
params_host['y'] = g_util.get_host(parameters[0])
params_host['x'] = g_util.get_host(parameters[1])
parameters = cl_array.to_device(queue, params_host)
if parameters.shape != image.shape:
raise ValueError("Parameters shape '{}' differs from image shape '{}'".
format(parameters.shape, image.shape))
image = g_util.get_image(image, queue=queue)
args = (image, out.data, sampler, cl_array.vec.make_int2(0, 0), parameters.data)
varconvolve(kernel_name, image.shape[::-1], args, queue=queue, block=block)
return out
def _varconvolve_2d_parametrized(
image, parameters, kernel_name, sampler=None, queue=None, out=None, block=False
):
"""Variable convolution of *image* with *parameters*, use OpoenCL kernel *kernel_name*,
*sampler*, *queue*, *out* and wait if *block* is True. Return *out*.
"""
if queue is None:
queue = cfg.OPENCL.queue
if out is None:
out = cl.array.Array(queue, image.shape, dtype=cfg.PRECISION.np_float)
if sampler is None:
sampler = cl.Sampler(
queue.context, False, cl.addressing_mode.CLAMP_TO_EDGE, cl.filter_mode.NEAREST
)
if not isinstance(parameters, cl_array.Array):
params_host = np.empty(parameters[0].shape, dtype=cfg.PRECISION.vfloat2)
params_host["y"] = g_util.get_host(parameters[0])
params_host["x"] = g_util.get_host(parameters[1])
parameters = cl_array.to_device(queue, params_host)
if parameters.shape != image.shape:
raise ValueError(
"Parameters shape '{}' differs from image shape '{}'".format(
parameters.shape, image.shape
)
)
image = g_util.get_image(image, queue=queue)
args = (image, out.data, sampler, cl_array.vec.make_int2(0, 0), parameters.data)
varconvolve(kernel_name, image.shape[::-1], args, queue=queue, block=block)
return out
def get_image(self,
photons,
shot_noise=True,
amplifier_noise=True,
psf=True,
queue=None):
"""Get digital counts image from incoming *photons*. The resulting image is based on the
incoming photons and dark current. We apply noise based on EMVA 1288 standard according to
which the variance :math:`\sigma_y^2 = K^2 ( \sigma_e^2 + \sigma_d^2 ) + \sigma_q^2`, where
:math:`K` is the system gain, :math:`\sigma_e^2` is the poisson- distributed shot noise
variance, :math:`\sigma_d^2` is the normal distributed electronics noise variance and
:math:`\sigma_q^2` is the quantization noise variance. If *shot_noise* is False don't apply
it. If *amplifier_noise* is False don't apply it as well. If *psf* is False don't apply the
point spread function.
"""
if self._last_input_shape != photons.shape:
self._last_input_shape = photons.shape
self._bin_factor = (photons.shape[0] / self.shape[0],
photons.shape[1] / self.shape[1])
if queue is None:
queue = cfg.OPENCL.queue
# Shot noise
# Adjust dark current for later binning and gain
dark = float(
self.dark_current) / self._bin_factor[0] / self._bin_factor[1]
electrons = dark + gutil.get_host(photons)
if self._bin_factor != (1, 1):
if psf:
sigma = (fwnm_to_sigma(self._bin_factor[0]),
fwnm_to_sigma(self._bin_factor[1]))
small = decimate(electrons,
self.shape,
sigma=sigma,
queue=queue)
else:
small = bin_image(electrons, self.shape, queue=queue)
electrons = gutil.get_host(small)
if shot_noise:
electrons = np.random.poisson(electrons)
if amplifier_noise and self.amplifier_sigma > 0:
# Add electronics noise
electrons = np.random.normal(electrons, self.amplifier_sigma)
counts = self.gain * electrons
# Cut the values beyond the maximum represented grey value given by
# bytes per pixel.
counts[counts > self.max_grey_value] = self.max_grey_value
# Apply quantization noise
return counts.astype(self.dtype)
def main():
args = parse_args()
syris.init(device_index=0)
m = 20
if args.input == 'grid':
image = make_grid(args.n, m * q.m).thickness.get()
elif args.input == 'lena':
from scipy.misc import lena
image = lena().astype(cfg.PRECISION.np_float)
if args.n != image.shape[0]:
image = gutil.get_host(ip.rescale(image, (args.n, args.n)))
n = image.shape[0]
crop_n = n - 2 * m - 2
y, x = np.mgrid[-n / 2:n / 2, -n / 2:n / 2]
# Compute a such that the disk diameter is exactly the period when distance from the middle is n
# / 2
a = m / (2 * (crop_n / 2.)**2)
radii = (a * np.sqrt(x**2 + y**2)**2 + 1e-3).astype(cfg.PRECISION.np_float)
x_param = radii
y_param = radii
result = ip.varconvolve_disk(image, (y_param, x_param)).get()
result = ip.crop(result, (m - 1, m - 1, crop_n, crop_n)).get()
radii = ip.crop(radii, (m - 1, m - 1, crop_n, crop_n)).get()
image = ip.crop(image, (m - 1, m - 1, crop_n, crop_n)).get()
if args.output:
save_image(args.output, result)
show(image, title='Original Image')
show(2 * radii, title='Blurring Disk Diameters')
show(result, title='Blurred Image')
plt.show()
def test_fft(self):
data = gpu_util.get_array(np.random.normal(100, 100,
size=(4, 4)).astype(cfg.PRECISION.np_float))
orig = gpu_util.get_host(data)
data = ip.fft_2(data)
ip.ifft_2(data)
np.testing.assert_almost_equal(orig, data.get().real, decimal=4)
# With a plan
from pyfft.cl import Plan
plan = Plan((4, 4), queue=cfg.OPENCL.queue)
data = ip.fft_2(np.copy(orig), plan=plan)
ip.ifft_2(data, plan=plan)
np.testing.assert_almost_equal(orig, data.get().real, decimal=4)
# Test double precision
syris.init(double_precision=True, device_index=0)
data = gpu_util.get_array(np.random.normal(100, 100,
size=(4, 4)).astype(cfg.PRECISION.np_float))
gt = np.fft.fft2(data.get())
data = ip.fft_2(data)
np.testing.assert_almost_equal(gt, data.get(), decimal=4)
gt = np.fft.ifft2(data.get())
data = ip.ifft_2(data)
np.testing.assert_almost_equal(gt, data.get(), decimal=4)
def main():
args = parse_args()
syris.init(device_index=0)
m = 20
if args.input == 'grid':
image = make_grid(args.n, m * q.m).thickness.get()
elif args.input == 'lena':
from scipy.misc import lena
image = lena().astype(cfg.PRECISION.np_float)
if args.n != image.shape[0]:
image = gutil.get_host(ip.rescale(image, (args.n, args.n)))
n = image.shape[0]
crop_n = n - 2 * m - 2
y, x = np.mgrid[-n / 2:n / 2, -n / 2:n / 2]
# Compute a such that the disk diameter is exactly the period when distance from the middle is n
# / 2
a = m / (2 * (crop_n / 2.) ** 2)
radii = (a * np.sqrt(x ** 2 + y ** 2) ** 2 + 1e-3).astype(cfg.PRECISION.np_float)
x_param = radii
y_param = radii
result = ip.varconvolve_disk(image, (y_param, x_param)).get()
result = ip.crop(result, (m - 1, m - 1, crop_n, crop_n)).get()
radii = ip.crop(radii, (m - 1, m - 1, crop_n, crop_n)).get()
image = ip.crop(image, (m - 1, m - 1, crop_n, crop_n)).get()
if args.output:
save_image(args.output, result)
show(image, title='Original Image')
show(2 * radii, title='Blurring Disk Diameters')
show(result, title='Blurred Image')
plt.show()
def make_motion(args):
syris.init()
n = 256
shape = (n, n)
energies = np.arange(5, 30, 1) * q.keV
bm, detector = make_devices(n, energies)
mb = create_sample(n, detector.pixel_size, velocity=20 * q.mm / q.s)
mb_2 = create_sample(n, detector.pixel_size, velocity=10 * q.mm / q.s)
mb.material = get_material('pmma_5_30_kev.mat')
mb_2.material = mb.material
cube = make_cube() / q.m * 30 * detector.pixel_size + 0.1 * detector.pixel_size
fov = detector.pixel_size * n
circle = make_circle().magnitude * fov / 30000 + fov / 2
tr = Trajectory(circle, velocity=10 * q.um / q.s)
glass = get_material('glass.mat')
mesh = Mesh(cube, tr, material=glass)
ex = Experiment([bm, mb, mb_2, mesh], bm, detector, 0 * q.m, energies)
for sample in ex.samples:
if sample != bm:
sample.trajectory.bind(detector.pixel_size)
if args.show_flat:
show(get_flat(shape, energies, detector, bm), title='Counts')
plt.show()
if args.conduct:
if args.output is not None and not os.path.exists(args.output):
os.makedirs(args.output, mode=0o755)
t_0 = 0 * q.s
if args.num_images:
t_1 = args.num_images / detector.camera.fps
else:
t_1 = ex.time
st = time.time()
mpl_im = None
for i, proj in enumerate(ex.make_sequence(t_0, t_1)):
image = get_host(proj)
if args.show:
if mpl_im is None:
plt.figure()
mpl_im = plt.imshow(image)
plt.show(False)
else:
mpl_im.set_data(image)
plt.draw()
if args.output:
path = os.path.join(args.output, 'projection_{:>05}.png').format(i)
scipy.misc.imsave(path, image)
print 'Maximum intensity:', image.max()
print 'Duration: {} s'.format(time.time() - st)
plt.show()
def get_image(self, photons, shot_noise=True, amplifier_noise=True, psf=True, queue=None):
"""Get digital counts image from incoming *photons*. The resulting image is based on the
incoming photons and dark current. We apply noise based on EMVA 1288 standard according to
which the variance :math:`\sigma_y^2 = K^2 ( \sigma_e^2 + \sigma_d^2 ) + \sigma_q^2`, where
:math:`K` is the system gain, :math:`\sigma_e^2` is the poisson- distributed shot noise
variance, :math:`\sigma_d^2` is the normal distributed electronics noise variance and
:math:`\sigma_q^2` is the quantization noise variance. If *shot_noise* is False don't apply
it. If *amplifier_noise* is False don't apply it as well. If *psf* is False don't apply the
point spread function.
"""
if self._last_input_shape != photons.shape:
self._last_input_shape = photons.shape
self._bin_factor = (photons.shape[0] / self.shape[0], photons.shape[1] / self.shape[1])
if queue is None:
queue = cfg.OPENCL.queue
# Shot noise
# Adjust dark current for later binning and gain
dark = float(self.dark_current) / self._bin_factor[0] / self._bin_factor[1]
electrons = dark + gutil.get_host(photons)
if self._bin_factor != (1, 1):
if psf:
sigma = (fwnm_to_sigma(self._bin_factor[0]), fwnm_to_sigma(self._bin_factor[1]))
small = decimate(electrons, self.shape, sigma=sigma, queue=queue)
else:
small = bin_image(electrons, self.shape, queue=queue)
electrons = gutil.get_host(small)
if shot_noise:
electrons = np.random.poisson(electrons)
if amplifier_noise and self.amplifier_sigma > 0:
# Add electronics noise
electrons = np.random.normal(electrons, self.amplifier_sigma)
counts = self.gain * electrons
# Cut the values beyond the maximum represented grey value given by
# bytes per pixel.
counts[counts > self.max_grey_value] = self.max_grey_value
# Apply quantization noise
return counts.astype(self.dtype)
def _test():
shape = 8, 4
dtypes = ['i', 'u', 'f']
lengths = [2, 4, 8]
types = [
np.dtype('{}{}'.format(dt, length))
for dt, length in itertools.product(dtypes, lengths)
]
types.append(np.dtype('i1'))
types.append(np.dtype('u1'))
types += [np.dtype('c8'), np.dtype('c16')]
for dtype in types:
np_data = np.arange(shape[0] *
shape[1]).reshape(shape).astype(dtype)
# host -> Array
cl_data = gu.get_array(np_data)
np.testing.assert_equal(np_data, cl_data.get())
# Array -> Array
res = gu.get_array(cl_data)
np.testing.assert_equal(res.get(), cl_data.get())
# Array -> host
host_data = gu.get_host(cl_data)
np.testing.assert_equal(np_data, host_data)
# host -> host
host_data = gu.get_host(np_data)
np.testing.assert_equal(np_data, host_data)
if dtype.kind != 'c':
# numpy -> Image and Image -> Array
image = gu.get_image(np_data)
back = gu.get_array(image).get()
np.testing.assert_equal(back, np_data)
# Image -> host
host_data = gu.get_host(image)
np.testing.assert_equal(host_data, np_data)
# Array -> Image
image = gu.get_image(cl_data)
back = gu.get_array(image).get()
np.testing.assert_equal(back, np_data)
# Image -> Image
image_2 = gu.get_image(image)
back = gu.get_array(image_2).get()
np.testing.assert_equal(back, np_data)
def _test():
shape = 8, 4
dtypes = ["i", "u", "f"]
lengths = [2, 4, 8]
types = [
np.dtype("{}{}".format(dt, length))
for dt, length in itertools.product(dtypes, lengths)
]
types.append(np.dtype("i1"))
types.append(np.dtype("u1"))
types += [np.dtype("c8"), np.dtype("c16")]
for dtype in types:
np_data = np.arange(shape[0] *
shape[1]).reshape(shape).astype(dtype)
# host -> Array
cl_data = gu.get_array(np_data)
np.testing.assert_equal(np_data, cl_data.get())
# Array -> Array
res = gu.get_array(cl_data)
np.testing.assert_equal(res.get(), cl_data.get())
# Array -> host
host_data = gu.get_host(cl_data)
np.testing.assert_equal(np_data, host_data)
# host -> host
host_data = gu.get_host(np_data)
np.testing.assert_equal(np_data, host_data)
if gu.are_images_supported() and dtype.kind != "c":
# numpy -> Image and Image -> Array
image = gu.get_image(np_data)
back = gu.get_array(image).get()
np.testing.assert_equal(back, np_data)
# Image -> host
host_data = gu.get_host(image)
np.testing.assert_equal(host_data, np_data)
# Array -> Image
image = gu.get_image(cl_data)
back = gu.get_array(image).get()
np.testing.assert_equal(back, np_data)
# Image -> Image
image_2 = gu.get_image(image)
back = gu.get_array(image_2).get()
np.testing.assert_equal(back, np_data)
def merge_tiles(tiles, num_tiles=None, outlier=(0, 0)):
"""Merge *tiles* which is a list to one large image. *num_tiles* is a tuple specifying the
number of tiles as (y, x) or None, meaning there is equal number of tiles in both dimensions.
The tiles must be stored in the row-major order.
"""
n, m = get_num_tiles(tiles, num_tiles=num_tiles)
tile_shape = tiles[0].shape
crop_shape = (tile_shape[0] - outlier[0], tile_shape[1] - outlier[1])
result = np.zeros((n * crop_shape[0], m * crop_shape[1]), dtype=tiles[0].dtype)
for j in range(n):
for i in range(m):
tile = g_util.get_host(tiles[j * m + i])[outlier[0] / 2:tile_shape[0] - outlier[0] / 2,
outlier[1] / 2:tile_shape[1] - outlier[1] / 2]
result[j * crop_shape[0]:(j + 1) * crop_shape[0],
i * crop_shape[1]:(i + 1) * crop_shape[1]] = tile
return result
def test_fft(self):
data = gpu_util.get_array(
np.random.normal(100, 100, size=(4, 4)).astype(cfg.PRECISION.np_float)
)
orig = gpu_util.get_host(data)
data = ip.fft_2(data)
ip.ifft_2(data)
np.testing.assert_almost_equal(orig, data.get().real, decimal=4)
# Test double precision
default_syris_init(double_precision=True)
data = gpu_util.get_array(
np.random.normal(100, 100, size=(4, 4)).astype(cfg.PRECISION.np_float)
)
gt = np.fft.fft2(data.get())
data = ip.fft_2(data)
np.testing.assert_almost_equal(gt, data.get(), decimal=4)
gt = np.fft.ifft2(data.get())
data = ip.ifft_2(data)
np.testing.assert_almost_equal(gt, data.get(), decimal=4)
def make_motion(args):
syris.init()
n = 256
shape = (n, n)
energies = np.arange(5, 30, 1) * q.keV
bm, detector = make_devices(n, energies)
mb = create_sample(n, detector.pixel_size, velocity=20 * q.mm / q.s)
mb_2 = create_sample(n, detector.pixel_size, velocity=10 * q.mm / q.s)
mb.material = get_material('pmma_5_30_kev.mat')
mb_2.material = mb.material
cube = make_cube(
) / q.m * 30 * detector.pixel_size + 0.1 * detector.pixel_size
fov = detector.pixel_size * n
circle = make_circle().magnitude * fov / 30000 + fov / 2
tr = Trajectory(circle, velocity=10 * q.um / q.s)
glass = get_material('glass.mat')
mesh = Mesh(cube, tr, material=glass)
ex = Experiment([bm, mb, mb_2, mesh], bm, detector, 0 * q.m, energies)
for sample in ex.samples:
if sample != bm:
sample.trajectory.bind(detector.pixel_size)
if args.show_flat:
show(get_flat(shape, energies, detector, bm), title='Counts')
plt.show()
if args.conduct:
if args.output is not None and not os.path.exists(args.output):
os.makedirs(args.output, mode=0o755)
t_0 = 0 * q.s
if args.num_images:
t_1 = args.num_images / detector.camera.fps
else:
t_1 = ex.time
st = time.time()
mpl_im = None
for i, proj in enumerate(ex.make_sequence(t_0, t_1)):
image = get_host(proj)
if args.show:
if mpl_im is None:
plt.figure()
mpl_im = plt.imshow(image)
plt.show(False)
else:
mpl_im.set_data(image)
plt.draw()
if args.output:
path = os.path.join(args.output,
'projection_{:>05}.png').format(i)
scipy.misc.imsave(path, image)
print 'Maximum intensity:', image.max()
print 'Duration: {} s'.format(time.time() - st)
plt.show()
st = time.time()
mpl_im = None
# make projections
for i, [data, filename] in enumerate(
ex.make_tomography(NO_OF_IMAGES,
THETA_MAX,
PAUSE,
NUM_REF_PER_BLOCK,
NUM_PROJ_PER_BLOCK,
NUM_DARK_IMG,
start_frame=START_I,
shape=shape)):
if START_I <= i:
image = get_host(data)
msg = '===== COMPUTED {}'
LOG.debug(msg.format(filename))
if PLOT_AND_PAUSE:
show(image)
plt.show()
if OUTPUT:
path_img = os.path.join(OUTPUT, filename)
tf.imsave(path_img, image.astype(np.uint16))
path_log = os.path.join(OUTPUT, 'scan.log')
ex.write_scan_log(path_log, NO_OF_IMAGES, NUM_REF_PER_BLOCK,
NUM_PROJ_PER_BLOCK, THETA_MAX, NUM_DARK_IMG)
