def expmap(self, sd_vectors):
if len(sd_vectors.shape) < 3:
sd_vectors = sd_vectors[None, ...]
gx = sd_vectors[..., 0]
gy = sd_vectors[..., 1]
gz = sd_vectors[..., 2]
sgz = sd_vectors[..., 3]
gzsgz = np.sqrt(gz**2 + sgz**2)
gxgy = np.sqrt(gx**2 + gy**2)
gx = gx / gxgy
gy = gy / gxgy
gz = gz / gzsgz
sgz = sgz / gzsgz
phi = np.arctan2(gy, gx)
theta = np.arctan2(sgz, gz)
cart = sph2cart(phi, theta, np.ones_like(phi), theta_origin='z')
cart[np.isnan(cart)] = 0.0
return cart
def expmap(self, sd_vectors):
if len(sd_vectors.shape) < 3:
sd_vectors = sd_vectors[None, ...]
gx = sd_vectors[..., 0]
gy = sd_vectors[..., 1]
gz = sd_vectors[..., 2]
sgz = sd_vectors[..., 3]
gzsgz = np.sqrt(gz ** 2 + sgz ** 2)
gxgy = np.sqrt(gx ** 2 + gy ** 2)
gx = gx / gxgy
gy = gy / gxgy
gz = gz / gzsgz
sgz = sgz / gzsgz
phi = np.arctan2(gy, gx)
theta = np.arctan2(sgz, gz)
cart = sph2cart(phi, theta, np.ones_like(phi), theta_origin='z')
cart[np.isnan(cart)] = 0.0
return cart
def expmap(self, tangent_vectors):
# If we've been passed a single vector to map, then add the extra axis
# Number of sample first
if len(tangent_vectors.shape) < 3:
tangent_vectors = tangent_vectors[None, ...]
eps = np.spacing(1)
rho = np.sqrt(tangent_vectors[..., 0]**2 + tangent_vectors[..., 1]**2)
Z = np.exp(
np.arctan2(tangent_vectors[..., 1], tangent_vectors[..., 0]) * 1j)
V1 = rho * np.real(Z)
V2 = rho * np.imag(Z)
# To prevent divide by 0 warnings when rho is 0
ir = rho == 0
rho[ir] = eps
c = rho
c[ir] = eps
Y = np.real(
np.arcsin(
np.cos(c) * np.sin(self.long0lat1[..., 1]) +
np.cos(self.long0lat1[..., 1]) * np.sin(c) * V2 / rho))
X = np.real(self.long0lat1[..., 0] + np.arctan2(
V1 * np.sin(c),
np.cos(self.long0lat1[..., 1]) * np.cos(c) * rho -
np.sin(self.long0lat1[..., 1]) * V2 * np.sin(c)))
ns = sph2cart(X - np.pi, Y, np.ones_like(Y))
# Swap x and y axes
ns[..., [0, 1]] = ns[..., [1, 0]]
ns[..., 1] = -ns[..., 1]
return ns
def expmap(self, tangent_vectors):
# If we've been passed a single vector to map, then add the extra axis
# Number of sample first
if len(tangent_vectors.shape) < 3:
tangent_vectors = tangent_vectors[None, ...]
eps = np.spacing(1)
rho = np.sqrt(tangent_vectors[..., 0] ** 2 +
tangent_vectors[..., 1] ** 2)
Z = np.exp(np.arctan2(tangent_vectors[..., 1],
tangent_vectors[..., 0]) * 1j)
V1 = rho * np.real(Z)
V2 = rho * np.imag(Z)
# To prevent divide by 0 warnings when rho is 0
ir = rho == 0
rho[ir] = eps
c = rho
c[ir] = eps
Y = np.real(np.arcsin(np.cos(c) * np.sin(self.long0lat1[..., 1]) +
np.cos(self.long0lat1[..., 1]) *
np.sin(c) * V2 / rho))
X = np.real(self.long0lat1[..., 0] +
np.arctan2(V1 * np.sin(c), np.cos(self.long0lat1[..., 1]) *
np.cos(c) * rho - np.sin(self.long0lat1[..., 1]) *
V2 * np.sin(c)))
ns = sph2cart(X - np.pi, Y, np.ones_like(Y))
# Swap x and y axes
ns[..., [0, 1]] = ns[..., [1, 0]]
ns[..., 1] = -ns[..., 1]
return ns
yaleb_path = '/mnt/atlas/databases/yaleb'
yaleb_subjects = ['yaleB01', 'yaleB02', 'yaleB03', 'yaleB04', 'yaleB05',
'yaleB06', 'yaleB07', 'yaleB08', 'yaleB09', 'yaleB10']
feature_spaces = ['aep', 'cosine', 'normal', 'pga', 'spherical']
# Create the tuples of images to use for photometric stereo and build the
# lights
image_light_paths = [('{0}_P00A+000E+00.pgm', 0.0, 0.0),
('{0}_P00A+000E-20.pgm', 0.0, -20.0),
('{0}_P00A+020E-40.pgm', 20.0, -40.0),
('{0}_P00A-020E-40.pgm', -20.0, -40.0),
('{0}_P00A+035E+40.pgm', 35.0, 40.0),
('{0}_P00A-035E+15.pgm', -35.0, 40.0)]
azimuths = np.asarray([np.radians(light[1]) for light in image_light_paths])
elevations = np.asarray([np.radians(light[2]) for light in image_light_paths])
lights = sph2cart(azimuths, elevations, np.ones(azimuths.shape[0]),
theta_origin='z')
lights[:, [0, 1]] = lights[:, [1, 0]]
sfs_light = lights[0, :]
# (Subject, Feature space) - Alphabetical order
mean_depth_error_results = np.zeros([len(yaleb_subjects),
len(feature_spaces)])
mean_angular_error_results = np.zeros([len(yaleb_subjects),
len(feature_spaces)])
# (Subject, Feature space) - Alphabetical order
std_depth_error_results = np.zeros([len(yaleb_subjects),
len(feature_spaces)])
std_angular_error_results = np.zeros([len(yaleb_subjects),
len(feature_spaces)])