def tst_conservation_of_counts(self): from scitbx import matrix from random import uniform, seed from dials.algorithms.profile_model.gaussian_rs import CoordinateSystem from dials.algorithms.profile_model.gaussian_rs import transform from scitbx.array_family import flex seed(0) assert (len(self.detector) == 1) s0 = self.beam.get_s0() m2 = self.gonio.get_rotation_axis() s0_length = matrix.col(self.beam.get_s0()).length() # Create an s1 map s1_map = transform.beam_vector_map(self.detector[0], self.beam, True) for i in range(100): # Get random x, y, z x = uniform(300, 1800) y = uniform(300, 1800) z = uniform(500, 600) # Get random s1, phi, panel s1 = matrix.col(self.detector[0].get_pixel_lab_coord( (x, y))).normalize() * s0_length phi = self.scan.get_angle_from_array_index(z, deg=False) panel = 0 # Calculate the bounding box bbox = self.calculate_bbox(s1, z, panel) x0, x1, y0, y1, z0, z1 = bbox # Create the coordinate system cs = CoordinateSystem(m2, s0, s1, phi) if abs(cs.zeta()) < 0.1: continue # The grid index generator step_size = self.delta_divergence / self.grid_size grid_index = transform.GridIndexGenerator(cs, x0, y0, (step_size, step_size), self.grid_size, s1_map) # Create the image #image = flex.double(flex.grid(z1 - z0, y1 - y0, x1 - x0), 1) image = gaussian((z1 - z0, y1 - y0, x1 - x0), 10.0, (z - z0, y - y0, x - x0), (2.0, 2.0, 2.0)) mask = flex.bool(flex.grid(image.all()), False) for j in range(y1 - y0): for i in range(x1 - x0): inside = False gx00, gy00 = grid_index(j, i) gx01, gy01 = grid_index(j, i + 1) gx10, gy10 = grid_index(j + 1, i) gx11, gy11 = grid_index(j + 1, i + 1) mingx = min([gx00, gx01, gx10, gx11]) maxgx = max([gx00, gx01, gx10, gx11]) mingy = min([gy00, gy01, gy10, gy11]) maxgy = max([gy00, gy01, gy10, gy11]) if (mingx >= 0 and maxgx < 2 * self.grid_size + 1 and mingy >= 0 and maxgy < 2 * self.grid_size + 1): inside = True for k in range(1, z1 - z0 - 1): mask[k, j, i] = inside # Transform the image to the grid transformed = transform.TransformForwardNoModel( self.spec, cs, bbox, 0, image.as_double(), mask) grid = transformed.profile() # Get the sums and ensure they're the same eps = 1e-7 sum_grid = flex.sum(grid) sum_image = flex.sum(flex.double(flex.select(image, flags=mask))) assert (abs(sum_grid - sum_image) <= eps) mask = flex.bool(flex.grid(image.all()), True) transformed = transform.TransformForwardNoModel( self.spec, cs, bbox, 0, image.as_double(), mask) grid = transformed.profile() # Boost the bbox to make sure all intensity is included x0, x1, y0, y1, z0, z1 = bbox bbox2 = (x0 - 10, x1 + 10, y0 - 10, y1 + 10, z0 - 10, z1 + 10) # Do the reverse transform transformed = transform.TransformReverseNoModel( self.spec, cs, bbox2, 0, grid) image2 = transformed.profile() # Check the sum of pixels are the same sum_grid = flex.sum(grid) sum_image = flex.sum(image2) assert (abs(sum_grid - sum_image) <= eps) # Do the reverse transform transformed = transform.TransformReverseNoModel( self.spec, cs, bbox, 0, grid) image2 = transformed.profile() from dials.algorithms.statistics import pearson_correlation_coefficient cc = pearson_correlation_coefficient(image.as_1d().as_double(), image2.as_1d()) assert (cc >= 0.99) # if cc < 0.99: # print cc, bbox # from matplotlib import pylab # pylab.plot(image.as_numpy_array()[(z1-z0)/2,(y1-y0)/2,:]) # pylab.show() # pylab.plot(image2.as_numpy_array()[(z1-z0)/2,(y1-y0)/2,:]) # pylab.show() # pylab.plot((image.as_double()-image2).as_numpy_array()[(z1-z0)/2,(y1-y0)/2,:]) # pylab.show() # Test passed print 'OK'
def test_forward_no_model(dials_data): sequence = load.imageset( dials_data("centroid_test_data").join("sweep.json").strpath) # Get the models beam = sequence.get_beam() detector = sequence.get_detector() gonio = sequence.get_goniometer() scan = sequence.get_scan() scan.set_image_range((0, 1000)) # Set some parameters sigma_divergence = beam.get_sigma_divergence(deg=False) mosaicity = 0.157 * math.pi / 180 n_sigma = 3 grid_size = 20 delta_divergence = n_sigma * sigma_divergence step_size = delta_divergence / grid_size delta_divergence2 = delta_divergence + step_size * 0.5 delta_mosaicity = n_sigma * mosaicity # Create the bounding box calculator calculate_bbox = BBoxCalculator3D(beam, detector, gonio, scan, delta_divergence2, delta_mosaicity) # Initialise the transform spec = transform.TransformSpec(beam, detector, gonio, scan, sigma_divergence, mosaicity, n_sigma + 1, grid_size) # tst_conservation_of_counts(self): random.seed(0) assert len(detector) == 1 s0 = beam.get_s0() m2 = gonio.get_rotation_axis() s0_length = matrix.col(beam.get_s0()).length() # Create an s1 map s1_map = transform.beam_vector_map(detector[0], beam, True) for i in range(100): # Get random x, y, z x = random.uniform(300, 1800) y = random.uniform(300, 1800) z = random.uniform(500, 600) # Get random s1, phi, panel s1 = matrix.col(detector[0].get_pixel_lab_coord( (x, y))).normalize() * s0_length phi = scan.get_angle_from_array_index(z, deg=False) panel = 0 # Calculate the bounding box bbox = calculate_bbox(s1, z, panel) x0, x1, y0, y1, z0, z1 = bbox # Create the coordinate system cs = CoordinateSystem(m2, s0, s1, phi) if abs(cs.zeta()) < 0.1: continue # The grid index generator step_size = delta_divergence / grid_size grid_index = transform.GridIndexGenerator(cs, x0, y0, (step_size, step_size), grid_size, s1_map) # Create the image # image = flex.double(flex.grid(z1 - z0, y1 - y0, x1 - x0), 1) image = gaussian((z1 - z0, y1 - y0, x1 - x0), 10.0, (z - z0, y - y0, x - x0), (2.0, 2.0, 2.0)) mask = flex.bool(flex.grid(image.all()), False) for j in range(y1 - y0): for i in range(x1 - x0): inside = False gx00, gy00 = grid_index(j, i) gx01, gy01 = grid_index(j, i + 1) gx10, gy10 = grid_index(j + 1, i) gx11, gy11 = grid_index(j + 1, i + 1) mingx = min([gx00, gx01, gx10, gx11]) maxgx = max([gx00, gx01, gx10, gx11]) mingy = min([gy00, gy01, gy10, gy11]) maxgy = max([gy00, gy01, gy10, gy11]) if (mingx >= 0 and maxgx < 2 * grid_size + 1 and mingy >= 0 and maxgy < 2 * grid_size + 1): inside = True for k in range(1, z1 - z0 - 1): mask[k, j, i] = inside # Transform the image to the grid transformed = transform.TransformForwardNoModel( spec, cs, bbox, 0, image.as_double(), mask) grid = transformed.profile() # Get the sums and ensure they're the same eps = 1e-7 sum_grid = flex.sum(grid) sum_image = flex.sum(flex.double(flex.select(image, flags=mask))) assert abs(sum_grid - sum_image) <= eps mask = flex.bool(flex.grid(image.all()), True) transformed = transform.TransformForwardNoModel( spec, cs, bbox, 0, image.as_double(), mask) grid = transformed.profile() # Boost the bbox to make sure all intensity is included x0, x1, y0, y1, z0, z1 = bbox bbox2 = (x0 - 10, x1 + 10, y0 - 10, y1 + 10, z0 - 10, z1 + 10) # Do the reverse transform transformed = transform.TransformReverseNoModel( spec, cs, bbox2, 0, grid) image2 = transformed.profile() # Check the sum of pixels are the same sum_grid = flex.sum(grid) sum_image = flex.sum(image2) assert abs(sum_grid - sum_image) <= eps # Do the reverse transform transformed = transform.TransformReverseNoModel( spec, cs, bbox, 0, grid) image2 = transformed.profile() from dials.algorithms.statistics import pearson_correlation_coefficient cc = pearson_correlation_coefficient(image.as_1d().as_double(), image2.as_1d()) assert cc >= 0.99