def build_a(values, rdir, c, kernel_size):
"""
Build a kernel containing pixel intensities
:param values:
:type values: np.ndarray
:param rdir:
:type rdir: np.ndarray
:param c:
:type c: np.ndarray
:param kernel_size:
:type kernel_size: float
"""
window = kernel_size / 2.
vdir = sph2vec(rdir.T)
vc = sph2vec(c)
d = np.abs(np.arccos(vdir.T.dot(vc)))
home_i = np.argmin(d)
home = values[home_i]
j = d <= window
a = np.zeros((j.sum(), 2), dtype=float)
dj = angle_between(rdir[j], rdir[home_i])
di = np.array([home - values[j]] * 2).T
zeros = dj == 0
a[~zeros] = (di[~zeros] / dj[~zeros].T).T
return a[:, ::-1]
def build_b(n_val, o_val, rdir, c, kernel_size):
window = kernel_size / 2.
vdir = sph2vec(rdir.T)
vc = sph2vec(c)
d = np.arccos(vdir.T.dot(vc))
j = d <= window
return n_val[j] - o_val[j]
def angular_deviation_3d(y_predict, y_target, theta=True, phi=True):
if theta:
thy = y_predict[:, 1]
tht = y_target[:, 1]
else:
thy = np.zeros_like(y_predict[:, 1])
tht = np.zeros_like(y_target[:, 1])
if phi:
phy = y_predict[:, 0]
pht = y_target[:, 0]
else:
phy = np.zeros_like(y_predict[:, 0])
pht = np.zeros_like(y_target[:, 0])
v1 = sph2vec(thy, phy)
v2 = sph2vec(tht, pht)
return np.rad2deg(np.arccos((v1 * v2).sum(axis=0)).std())
def gaussian_weight(size, rdir, c, even=False):
window = size / 2.
vdir = sph2vec(rdir.T)
vc = sph2vec(c)
d = np.arccos(vdir.T.dot(vc))
j = d <= window
d = d[j]
if even:
return np.diag(np.ones(j.sum()))
sigma = 2 * np.pi / rdir.shape[
1] # the standard deviation of your normal curve
correlation = 0. # see wiki for multivariate normal distributions
z = (2 * np.pi * sigma * sigma) * np.sqrt(1 - correlation * correlation)
exp = -1 / (2 *
(1 - correlation * correlation)) * (d * d) / (sigma * sigma)
return np.diag(1. / z * np.exp(exp))
def visualise(cls,
sensor,
sL=None,
show_sun=False,
sides=True,
interactive=False,
colormap=None,
scale=[0, 1],
title=None):
"""
:param sensor:
:type sensor: CompassSensor
:return:
"""
import matplotlib.pyplot as plt
from matplotlib.patches import Ellipse, Rectangle
from matplotlib.cm import get_cmap
import matplotlib as mpl
if interactive:
plt.ion()
if colormap is not None:
get_colour = lambda x: get_cmap(colormap)(x)
else:
get_colour = lambda x: x * np.array([.5, .5, 1.])
xyz = sph2vec(np.pi / 2 - sensor.theta_local, np.pi + sensor.phi_local,
sensor.R_c)
xyz[0] *= -1
lat, lon = hp.Rotator(rot=(np.rad2deg(-sensor.yaw),
np.rad2deg(-sensor.pitch),
np.rad2deg(-sensor.roll)))(
sensor.sky.lat, np.pi - sensor.sky.lon)
xyz_sun = sph2vec(np.pi / 2 - lat, np.pi + lon, sensor.R_c)
xyz_sun[0] *= -1
if sides:
figsize = (10, 10)
else:
if colormap is None or scale is None:
figsize = (5, 5)
else:
figsize = (5.05, 5.05)
plt.figure("Sensor Design" if title is None else title,
figsize=figsize)
# top view
if sides:
ax_t = plt.subplot2grid((4, 4), (1, 1),
colspan=2,
rowspan=2,
aspect="equal",
adjustable='box-forced')
else:
if colormap is not None and scale is not None:
ax_t = plt.subplot2grid((10, 10), (0, 0),
colspan=9,
rowspan=9,
aspect="equal",
adjustable='box-forced')
else:
ax_t = plt.subplot(111)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
sensor_outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.alpha + 2 * sensor.R_l,
height=2 * sensor.alpha + 2 * sensor.R_l)
ax_t.add_artist(outline)
outline.set_clip_box(ax_t.bbox)
outline.set_alpha(.2)
outline.set_facecolor("grey")
ax_t.add_artist(sensor_outline)
sensor_outline.set_clip_box(ax_t.bbox)
sensor_outline.set_alpha(.5)
sensor_outline.set_facecolor("grey")
stheta, sphi = sensor.theta_local, np.pi + sensor.phi_local
sL = sL if sL is not None else sensor.L
if sensor.mode == "event":
sL = np.clip(1. * sL + .5, 0., 1.)
for (x, y, z), th, ph, L in zip(xyz.T, stheta, sphi, sL):
lens = Ellipse(xy=[x, y],
width=1.5 * sensor.r_l,
height=1.5 * np.cos(th) * sensor.r_l,
angle=np.rad2deg(ph))
ax_t.add_artist(lens)
lens.set_clip_box(ax_t.bbox)
lens.set_facecolor(get_colour(np.asscalar(L)))
if show_sun:
ax_t.plot(xyz_sun[0], xyz_sun[1], 'ro', markersize=22)
ax_t.set_xlim(-sensor.R_c - 2, sensor.R_c + 2)
ax_t.set_ylim(-sensor.R_c - 2, sensor.R_c + 2)
ax_t.set_xticklabels([])
ax_t.set_yticklabels([])
if sides:
# side view #1 (x, z)
ax = plt.subplot2grid((4, 4), (0, 1),
colspan=2,
aspect="equal",
adjustable='box-forced',
sharex=ax_t)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
fade_one = Rectangle(xy=[-sensor.R_c, -sensor.R_c],
width=2 * sensor.R_c,
height=2 * sensor.R_c - sensor.height -
sensor.R_l)
ax.add_artist(outline)
outline.set_clip_box(ax.bbox)
outline.set_alpha(.5)
outline.set_facecolor("grey")
ax.add_artist(fade_one)
fade_one.set_clip_box(ax.bbox)
fade_one.set_alpha(.6)
fade_one.set_facecolor("white")
for (x, y, z), th, ph, L in zip(xyz.T, stheta, sphi, sL):
if y > 0:
continue
lens = Ellipse(xy=[x, z],
width=1.5 * sensor.r_l,
height=np.sin(-y / sensor.R_c) * 1.5 *
sensor.r_l,
angle=np.rad2deg(np.arcsin(-x / sensor.R_c)))
ax.add_artist(lens)
lens.set_clip_box(ax.bbox)
lens.set_facecolor(get_colour(L))
ax.set_xlim(-sensor.R_c - 2, sensor.R_c + 2)
ax.set_ylim(0, sensor.R_c + 2)
ax.set_xticks([])
ax.set_yticks([])
# side view #2 (-x, z)
ax = plt.subplot2grid((4, 4), (3, 1),
colspan=2,
aspect="equal",
adjustable='box-forced',
sharex=ax_t)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
fade_one = Rectangle(
xy=[-sensor.R_c, -sensor.R_c + sensor.height + sensor.R_l],
width=2 * sensor.R_c,
height=2 * sensor.R_c - sensor.height - sensor.R_l)
ax.add_artist(outline)
outline.set_clip_box(ax.bbox)
outline.set_alpha(.5)
outline.set_facecolor("grey")
ax.add_artist(fade_one)
fade_one.set_clip_box(ax.bbox)
fade_one.set_alpha(.6)
fade_one.set_facecolor("white")
for (x, y, z), th, ph, L in zip(xyz.T, stheta, sphi, sL):
if y < 0:
continue
lens = Ellipse(xy=[x, -z],
width=1.5 * sensor.r_l,
height=np.sin(-y / sensor.R_c) * 1.5 *
sensor.r_l,
angle=np.rad2deg(np.arcsin(x / sensor.R_c)))
ax.add_artist(lens)
lens.set_clip_box(ax.bbox)
lens.set_facecolor(get_colour(L))
ax.set_xlim(-sensor.R_c - 2, sensor.R_c + 2)
ax.set_ylim(-sensor.R_c - 2, 0)
ax.set_yticks([])
# side view #3 (y, z)
ax = plt.subplot2grid((4, 4), (1, 3),
rowspan=2,
aspect="equal",
adjustable='box-forced',
sharey=ax_t)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
fade_one = Rectangle(xy=[-sensor.R_c, -sensor.R_c],
width=2 * sensor.R_c - sensor.height -
sensor.R_l,
height=2 * sensor.R_c)
ax.add_artist(outline)
outline.set_clip_box(ax.bbox)
outline.set_alpha(.5)
outline.set_facecolor("grey")
ax.add_artist(fade_one)
fade_one.set_clip_box(ax.bbox)
fade_one.set_alpha(.6)
fade_one.set_facecolor("white")
for (x, y, z), th, ph, L in zip(xyz.T, stheta, sphi, sL):
if x > 0:
continue
lens = Ellipse(
xy=[z, y],
width=1.5 * sensor.r_l,
height=np.sin(-x / sensor.R_c) * 1.5 * sensor.r_l,
angle=np.rad2deg(np.arcsin(y / sensor.R_c)) + 90)
ax.add_artist(lens)
lens.set_clip_box(ax.bbox)
lens.set_facecolor(get_colour(L))
ax.set_ylim(-sensor.R_c - 2, sensor.R_c + 2)
ax.set_xlim(0, sensor.R_c + 2)
ax.set_yticks([])
ax.set_xticks([])
# side view #4 (-y, z)
ax = plt.subplot2grid((4, 4), (1, 0),
rowspan=2,
aspect="equal",
adjustable='box-forced',
sharey=ax_t)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
fade_one = Rectangle(
xy=[-sensor.R_c + sensor.height + sensor.R_l, -sensor.R_c],
width=2 * sensor.R_c - sensor.height - sensor.R_l,
height=2 * sensor.R_c)
ax.add_artist(outline)
outline.set_clip_box(ax.bbox)
outline.set_alpha(.5)
outline.set_facecolor("grey")
ax.add_artist(fade_one)
fade_one.set_clip_box(ax.bbox)
fade_one.set_alpha(.6)
fade_one.set_facecolor("white")
for (x, y, z), th, ph, L in zip(xyz.T, stheta, sphi, sL):
if x < 0:
continue
lens = Ellipse(
xy=[-z, y],
width=1.5 * sensor.r_l,
height=np.sin(-x / sensor.R_c) * 1.5 * sensor.r_l,
angle=np.rad2deg(np.arcsin(-y / sensor.R_c)) - 90)
ax.add_artist(lens)
lens.set_clip_box(ax.bbox)
lens.set_facecolor(get_colour(L))
ax.set_ylim(-sensor.R_c - 2, sensor.R_c + 2)
ax.set_xlim(-sensor.R_c - 2, 0)
ax.set_xticks([])
else:
plt.axis('off')
if colormap is not None and not sides and scale is not None:
ax = plt.subplot2grid((10, 10), (9, 9),
aspect="equal",
adjustable='box-forced',
projection='polar')
ax._direction = 2 * np.pi
ax.set_theta_zero_location("N")
# quant_steps = 360
cb = mpl.colorbar.ColorbarBase(ax,
cmap=get_cmap(colormap),
norm=mpl.colors.Normalize(
0.0, 2 * np.pi),
orientation='horizontal',
ticks=np.linspace(0,
2 * np.pi,
4,
endpoint=False))
cb.outline.set_visible(False)
# cb.ax.set_xticks([0, np.pi/6, np.pi/3, np.pi/2, 2*np.pi/3, 5*np.pi/6, np.pi])
cb.ax.set_xticklabels(
np.linspace(scale[0], scale[1], 4, endpoint=False))
cb.ax.tick_params(pad=1, labelsize=8)
plt.tight_layout(pad=0.)
if interactive:
plt.draw()
plt.pause(.1)
else:
plt.show()
def visualise_compnet(sensor, sL=None, interactive=True, cmap="coolwarm", vmin=0, vmax=1, title=None):
"""
:param sensor:
:type sensor: CompassSensor
:return:
"""
if interactive and not plt.isinteractive():
plt.ion()
if isinstance(cmap, basestring):
cmap = get_cmap(cmap)
xyz = sph2vec(np.pi/2 - sensor.theta_local, np.pi + sensor.phi_local, sensor.R_c)
xyz[0] *= -1
lat, lon = hp.Rotator(rot=(
np.rad2deg(-sensor.yaw), np.rad2deg(-sensor.pitch), np.rad2deg(-sensor.roll)
))(sensor.sky.lat, np.pi-sensor.sky.lon)
xyz_sun = sph2vec(np.pi/2 - lat, np.pi + lon, sensor.R_c)
xyz_sun[0] *= -1
fig = plt.figure("Compass-CompModel" if title is None or interactive else title, figsize=(12, 7))
fig.clear()
ax_t = plt.subplot2grid((1, 12), (0, 0), colspan=10)
outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.R_c,
height=2 * sensor.R_c)
sensor_outline = Ellipse(xy=np.zeros(2),
width=2 * sensor.alpha + 2 * sensor.R_l,
height=2 * sensor.alpha + 2 * sensor.R_l)
ax_t.add_patch(outline)
outline.set_clip_box(ax_t.bbox)
outline.set_alpha(.2)
outline.set_facecolor("grey")
stheta, sphi = sensor.theta_local, np.pi + sensor.phi_local
sL = np.clip(sL if sL is not None else sensor.L, 0., 1.)
for k, ((x, y, z), th, ph, L) in enumerate(zip(xyz.T, stheta, sphi, sL)):
for j, tl2 in enumerate(sensor.tl2):
line = ConnectionPatch([x, y], [sensor.R_c + 5, j * sensor.R_c / 8. - sensor.R_c + 1],
"data", "data", lw=.5, color=cmap(np.asscalar(L) * sensor.w_tl2[k, j] + .5))
ax_t.add_patch(line)
ax_t.add_patch(sensor_outline)
sensor_outline.set_clip_box(ax_t.bbox)
sensor_outline.set_alpha(.5)
sensor_outline.set_facecolor("grey")
for k, ((x, y, z), th, ph, L) in enumerate(zip(xyz.T, stheta, sphi, sL)):
lens = Ellipse(xy=[x, y], width=1.5 * sensor.r_l, height=1.5 * np.cos(th) * sensor.r_l,
angle=np.rad2deg(ph))
ax_t.add_patch(lens)
lens.set_clip_box(ax_t.bbox)
lens.set_facecolor(cmap(np.asscalar(L)))
scale = 5.
for j, tl2 in enumerate(sensor.tl2):
x = sensor.R_c + 5
y = j * sensor.R_c / 8. - sensor.R_c + 1
for k, cl1 in enumerate(sensor.cl1):
line = ConnectionPatch([x, y], [sensor.R_c + 10, k * sensor.R_c / 8. - sensor.R_c + 1],
"data", "data", lw=.5, color=cmap(scale * tl2 * sensor.w_cl1[j, k] + .5))
ax_t.add_patch(line)
neuron = Ellipse(xy=[x, y], width=.1 * sensor.R_c, height=.1 * sensor.R_c)
ax_t.add_artist(neuron)
neuron.set_clip_box(ax_t.bbox)
neuron.set_facecolor(cmap(scale * tl2 + .5))
for j, cl1 in enumerate(sensor.cl1):
x = sensor.R_c + 10
y = j * sensor.R_c / 8. - sensor.R_c + 1
for k, tb1 in enumerate(sensor.tb1):
line = ConnectionPatch([x, y], [sensor.R_c + 15, k * sensor.R_c / 8. - sensor.R_c / 2. + 1],
"data", "data", lw=.5, color=cmap(scale * cl1 * sensor.w_tb1[j, k] + .5))
ax_t.add_patch(line)
neuron = Ellipse(xy=[x, y], width=.1 * sensor.R_c, height=.1 * sensor.R_c)
ax_t.add_artist(neuron)
neuron.set_clip_box(ax_t.bbox)
neuron.set_facecolor(cmap(scale * cl1 + .5))
for j, tb1 in enumerate(sensor.tb1):
x = sensor.R_c + 15
y = j * sensor.R_c / 8. - sensor.R_c / 2 + 1
neuron = Ellipse(xy=[x, y], width=.1 * sensor.R_c, height=.1 * sensor.R_c)
ax_t.add_artist(neuron)
neuron.set_clip_box(ax_t.bbox)
neuron.set_facecolor(cmap(scale * tb1 + .5))
ax_t.text(0, sensor.R_c - sensor.r_l, "0", fontsize=15, verticalalignment='center', horizontalalignment='center')
ax_t.text(-sensor.R_c + sensor.r_l, 0, "90", fontsize=15, verticalalignment='center', horizontalalignment='center')
ax_t.text(0, -sensor.R_c + sensor.r_l, "180", fontsize=15, verticalalignment='center', horizontalalignment='center')
ax_t.text(sensor.R_c - sensor.r_l, 0, "-90", fontsize=15, verticalalignment='center', horizontalalignment='center')
norm = np.sqrt(np.square(xyz_sun[0]) + np.square(xyz_sun[1])) / (sensor.R_c - 3 * sensor.r_l)
ax_t.plot([-xyz_sun[0]/norm, xyz_sun[0]/norm], [-xyz_sun[1]/norm, xyz_sun[1]/norm], "k-")
ax_t.plot(xyz_sun[0], xyz_sun[1],
marker='o',
fillstyle='full',
markeredgecolor='black',
markerfacecolor='yellow',
markersize=15)
ax_t.text(0, 0, "z", fontsize=15, verticalalignment='center', horizontalalignment='center',
color='red', fontweight='bold')
ax_t.set_xlim(-sensor.R_c - 2, sensor.R_c + 17)
ax_t.set_ylim(-sensor.R_c - 2, sensor.R_c + 2)
ax_t.set_xticklabels([])
ax_t.set_yticklabels([])
plt.axis('off')
plt.subplot2grid((4, 12), (1, 10), colspan=2, rowspan=2)
theta, phi = decode_sph(sensor.tb1)
alpha = np.linspace(0, 2*np.pi, 100)
plt.plot(theta * np.sin(phi + alpha + np.pi/2), alpha)
plt.ylabel("phase (degrees)")
plt.ylim([0, 2*np.pi])
plt.xlim([theta+np.pi/360, -theta-np.pi/360])
plt.xticks([-theta, 0, theta], ["%d" % np.rad2deg(-theta), "0", "%d" % np.rad2deg(theta)])
plt.yticks([0, np.pi/2, np.pi, 3*np.pi/2, 2*np.pi], ["0", "90", "180", "270", "360"])
plt.tight_layout(pad=0.)
if interactive:
plt.draw()
plt.pause(.1)
else:
plt.show()
