def shift(self, shift_cols, shift_rows):
"""Shift the image foreground
This function shifts the center of mass (CoM) of the image by the
specified number of rows and columns.
Parameters
----------
shift_cols : float
Number of columns by which to shift the CoM.
Positive: to the right, negative: to the left
shift_rows : float
Number of rows by which to shift the CoM.
Positive: downward, negative: upward
Returns
-------
stim : `ImageStimulus`
A copy of the stimulus object containing the shifted image
"""
img = self.data.reshape(self.img_shape)
tf = SimilarityTransform(translation=[shift_cols, shift_rows])
img = img_warp(img, tf.inverse)
return ImageStimulus(img,
electrodes=self.electrodes,
metadata=self.metadata)
def center(self, loc=None):
"""Center the image foreground
This function shifts the center of mass (CoM) to the image center.
Parameters
----------
loc : (col, row), optional
The pixel location at which to center the CoM. By default, shifts
the CoM to the image center.
Returns
-------
stim : `ImageStimulus`
A copy of the stimulus object containing the centered image
"""
# Calculate center of mass:
img = self.data.reshape(self.img_shape)
m = img_moments(img, order=1)
# No area found:
if np.isclose(m[0, 0], 0):
return img
# Center location:
if loc is None:
loc = np.array(self.img_shape[::-1]) / 2.0 - 0.5
# Shift the image by -centroid, +image center:
transl = (loc[0] - m[0, 1] / m[0, 0], loc[1] - m[1, 0] / m[0, 0])
tf_shift = SimilarityTransform(translation=transl)
img = img_warp(img, tf_shift.inverse)
return ImageStimulus(img,
electrodes=self.electrodes,
metadata=self.metadata)
def scale(self, scaling_factor):
"""Scale the image foreground
This function scales the image foreground (excluding black pixels)
by a factor.
Parameters
----------
scaling_factor : float
Factory by which to scale the image
Returns
-------
stim : `ImageStimulus`
A copy of the stimulus object containing the scaled image
"""
if scaling_factor <= 0:
raise ValueError("Scaling factor must be greater than zero")
# Calculate center of mass:
img = self.data.reshape(self.img_shape)
m = img_moments(img, order=1)
# No area found:
if np.isclose(m[0, 0], 0):
return img
# Shift the phosphene to (0, 0):
center_mass = np.array([m[0, 1] / m[0, 0], m[1, 0] / m[0, 0]])
tf_shift = SimilarityTransform(translation=-center_mass)
# Scale the phosphene:
tf_scale = SimilarityTransform(scale=scaling_factor)
# Shift the phosphene back to where it was:
tf_shift_inv = SimilarityTransform(translation=center_mass)
# Combine all three transforms:
tf = tf_shift + tf_scale + tf_shift_inv
img = img_warp(img, tf.inverse)
return ImageStimulus(img,
electrodes=self.electrodes,
metadata=self.metadata)
def plot_argus_phosphenes(data,
argus=None,
scale=1.0,
axon_map=None,
show_fovea=True,
ax=None):
"""Plots phosphenes centered over the corresponding electrodes
.. versionadded:: 0.7
Parameters
----------
data : pd.DataFrame
The Beyeler2019 dataset, a subset thereof, or a DataFrame with
identical organization (i.e., must contain columns 'subject', 'image',
'xrange', and 'yrange').
argus : :py:class:`~pulse2percept.implants.ArgusI` or :py:class:`~pulse2percept.implants.ArgusII`
Either an Argus I or Argus II implant. If None, the data is expected to
contain additional columns: "implant_type_str", "implant_x",
"implant_y", and "implant_rot", which together define the type and
position of the implant. "implant_type_str" must be either "ArgusI" or
"ArgusII".
scale : float
Scaling factor to apply to the phosphenes
axon_map : :py:class:`~pulse2percept.models.AxonMapModel`
An instance of the axon map model to use for visualization.
show_fovea : bool
Whether to indicate the location of the fovea with a square
ax : axis
Matplotlib axis
"""
if not isinstance(data, pd.DataFrame):
raise TypeError(f'"data" must be a Pandas DataFrame, not '
f'{type(data)}.')
req_cols = ['subject', 'electrode', 'image', 'xrange', 'yrange']
if not all(col in data.columns for col in req_cols):
raise ValueError(f'"data" must have columns {req_cols}.')
if len(data) == 0:
raise ValueError('"data" is empty.')
if len(data.subject.unique()) > 1:
raise ValueError('"data" cannot contain data from more than one '
'subject.')
if axon_map is not None and not isinstance(axon_map, AxonMapModel):
raise TypeError(f'"axon_map" must be an AxonMapModel instance, not '
f'{type(axon_map)}.')
if argus is None:
# Implant not given, must first be created from data columns:
try:
specs = data.iloc[0]
implant_type = ArgusI if specs.implant_type_str == 'ArgusI' else ArgusII
argus = implant_type(x=specs['implant_x'],
y=specs['implant_y'],
rot=specs['implant_rot'])
except (KeyError, AttributeError):
raise ValueError('If "argus" is not given, "data" must contain '
'columns "implant_type_str", "implant_x", '
'"implant_y", and "implant_rot".')
if not isinstance(argus, (ArgusI, ArgusII)):
raise TypeError(f'"argus" must be an Argus I or Argus II implant, '
f'not {type(argus)}.')
if isinstance(argus, ArgusI):
px_argus = PX_ARGUS1
img_argus = imread(PATH_ARGUS1)
else:
px_argus = PX_ARGUS2
img_argus = imread(PATH_ARGUS2)
if ax is None:
ax = plt.gca()
alpha_bg = 0.5 # alpha value for the array in the background
thresh_fg = 0.95 # Grayscale value above which to mask the drawings
# To simulate an implant in a left eye, flip the image left-right (along
# with the electrode x-coordinates):
if argus.eye == 'LE':
img_argus = np.fliplr(img_argus)
px_argus[:, 0] = img_argus.shape[1] - px_argus[:, 0] - 1
# Add some padding to the output image so the array is not cut off:
pad = 2000 # microns
x_el = [e.x for e in argus.electrode_objects]
y_el = [e.y for e in argus.electrode_objects]
x_min, y_min = Watson2014Map.ret2dva(
np.min(x_el) - pad,
np.min(y_el) - pad)
x_max, y_max = Watson2014Map.ret2dva(
np.max(x_el) + pad,
np.max(y_el) + pad)
# Coordinate transform from degrees of visual angle to output, and from
# image coordinates to output image:
pts_in = []
pts_dva = []
pts_out = []
try:
out_shape = data.img_shape.values[0]
except AttributeError:
# Dataset does not have 'img_shape' column:
try:
out_shape = data.image.values[0].shape
except IndexError:
out_shape = (768, 1024)
for xy, e in zip(px_argus, argus.electrode_objects):
x_dva, y_dva = Watson2014Map.ret2dva(e.x, e.y)
x_out = (x_dva - x_min) / (x_max - x_min) * (out_shape[1] - 1)
y_out = (y_dva - y_min) / (y_max - y_min) * (out_shape[0] - 1)
pts_in.append(xy)
pts_dva.append([x_dva, y_dva])
pts_out.append([x_out, y_out])
pts_in = np.array(pts_in)
pts_dva = np.array(pts_dva)
pts_out = np.array(pts_out)
dva2out = img_transform('similarity', pts_dva, pts_out)
argus2out = img_transform('similarity', pts_in, pts_out)
# Top left, top right, bottom left, bottom right corners:
x_range = data.xrange.values[0]
y_range = data.yrange.values[0]
pts_dva = [[x_range[0], y_range[0]], [x_range[0], y_range[1]],
[x_range[1], y_range[0]], [x_range[1], y_range[1]]]
# Calculate average drawings, but don't binarize:
all_imgs = np.zeros(out_shape)
num_imgs = data.groupby('electrode')['image'].count()
for _, row in data.iterrows():
e_pos = Watson2014Map.ret2dva(argus[row['electrode']].x,
argus[row['electrode']].y)
align_center = dva2out(e_pos)[0]
img_drawing = scale_image(row['image'], scale)
img_drawing = center_image(img_drawing, loc=align_center)
# We normalize by the number of phosphenes per electrode, so that if
# all phosphenes are the same, their brightness would add up to 1:
all_imgs += 1.0 / num_imgs[row['electrode']] * img_drawing
all_imgs = 1 - all_imgs
# Draw array schematic with specific alpha level:
img_arr = img_warp(img_argus,
argus2out.inverse,
cval=1.0,
output_shape=out_shape)
img_arr[:, :, 3] = alpha_bg
# Replace pixels where drawings are dark enough, set alpha=1:
rr, cc = np.unravel_index(
np.where(all_imgs.ravel() < thresh_fg)[0], all_imgs.shape)
for channel in range(3):
img_arr[rr, cc, channel] = all_imgs[rr, cc]
img_arr[rr, cc, 3] = 1
ax.imshow(img_arr, cmap='gray', zorder=ZORDER['background'])
if show_fovea:
fovea = dva2out([0, 0])[0]
ax.scatter(*fovea,
s=100,
marker='s',
c='w',
edgecolors='k',
zorder=ZORDER['foreground'])
if axon_map is not None:
axon_bundles = axon_map.grow_axon_bundles(n_bundles=100, prune=False)
# Draw axon pathways:
for bundle in axon_bundles:
# Flip y upside down for dva:
bundle = Watson2014Map.ret2dva(bundle[:, 0], -bundle[:, 1])
bundle = np.array(bundle).T
# Set segments outside the drawing window to NaN:
x_idx = np.logical_or(bundle[:, 0] < x_min, bundle[:, 0] > x_max)
bundle[x_idx, 0] = np.nan
y_idx = np.logical_or(bundle[:, 1] < y_min, bundle[:, 1] > y_max)
bundle[y_idx, 1] = np.nan
bundle = dva2out(bundle)
ax.plot(bundle[:, 0],
bundle[:, 1],
c=(0.6, 0.6, 0.6),
linewidth=2,
zorder=ZORDER['background'])
return ax
def plot_argus_phosphenes(X,
argus,
scale=1.0,
axon_map=None,
show_fovea=True,
ax=None):
"""Plots phosphenes centered over the corresponding electrodes
.. versionadded:: 0.7
Parameters
----------
X : pd.DataFrame
argus : :py:class:`~pulse2percept.implants.ArgusI` or :py:class:`~pulse2percept.implants.ArgusII`
Either an Argus I or Argus II implant
scale : float
Scaling factor to apply to the phosphenes
show_fovea : bool
Whether to indicate the location of the fovea with a square
ax : axis
Matplotlib axis
"""
if not isinstance(X, pd.DataFrame):
raise TypeError('"X" must be a Pandas DataFrame, not %s.' % type(X))
req_cols = ['subject', 'electrode', 'image', 'img_x_dva', 'img_y_dva']
if not all(col in X.columns for col in req_cols):
raise ValueError('"X" must have columns %s.' % req_cols)
if len(X) == 0:
raise ValueError('"X" is empty.')
if len(X.subject.unique()) > 1:
raise ValueError('"X" cannot contain data from more than one subject.')
if not isinstance(argus, (ArgusI, ArgusII)):
raise TypeError('"argus" must be an Argus I or Argus II implant, not '
'%s.' % type(argus))
if ax is None:
ax = plt.gca()
alpha_bg = 0.5 # alpha value for the array in the background
thresh_fg = 0.95 # Grayscale value above which to mask the drawings
if isinstance(argus, ArgusI):
px_argus = PX_ARGUS1
img_argus = imread(PATH_ARGUS1)
else:
px_argus = PX_ARGUS2
img_argus = imread(PATH_ARGUS2)
# To simulate an implant in a left eye, flip the image left-right (along
# with the electrode x-coordinates):
if argus.eye == 'LE':
img_argus = np.fliplr(img_argus)
px_argus[:, 0] = img_argus.shape[1] - px_argus[:, 0] - 1
# Add some padding to the output image so the array is not cut off:
pad = 2000 # microns
x_el = [e.x for e in argus.values()]
y_el = [e.y for e in argus.values()]
x_min = Watson2014Transform.ret2dva(np.min(x_el) - pad)
x_max = Watson2014Transform.ret2dva(np.max(x_el) + pad)
y_min = Watson2014Transform.ret2dva(np.min(y_el) - pad)
y_max = Watson2014Transform.ret2dva(np.max(y_el) + pad)
# Coordinate transform from degrees of visual angle to output, and from
# image coordinates to output image:
pts_in = []
pts_dva = []
pts_out = []
try:
out_shape = X.img_shape.values[0]
except AttributeError:
# Dataset does not have 'img_shape' column:
try:
out_shape = X.image.values[0].shape
except IndexError:
out_shape = (768, 1024)
for xy, e in zip(px_argus, argus.values()):
x_dva, y_dva = Watson2014Transform.ret2dva([e.x, e.y])
x_out = (x_dva - x_min) / (x_max - x_min) * (out_shape[1] - 1)
y_out = (y_dva - y_min) / (y_max - y_min) * (out_shape[0] - 1)
pts_in.append(xy)
pts_dva.append([x_dva, y_dva])
pts_out.append([x_out, y_out])
pts_in = np.array(pts_in)
pts_dva = np.array(pts_dva)
pts_out = np.array(pts_out)
dva2out = img_transform('similarity', pts_dva, pts_out)
argus2out = img_transform('similarity', pts_in, pts_out)
# Top left, top right, bottom left, bottom right corners:
x_range = X.img_x_dva
y_range = X.img_y_dva
pts_dva = [[x_range[0], y_range[0]], [x_range[0], y_range[1]],
[x_range[1], y_range[0]], [x_range[1], y_range[1]]]
# Calculate average drawings, but don't binarize:
all_imgs = np.zeros(out_shape)
num_imgs = X.groupby('electrode')['image'].count()
for _, row in X.iterrows():
e_pos = Watson2014Transform.ret2dva(
(argus[row['electrode']].x, argus[row['electrode']].y))
align_center = dva2out(e_pos)[0]
img_drawing = scale_phosphene(row['image'], scale)
img_drawing = center_phosphene(img_drawing, loc=align_center)
# We normalize by the number of phosphenes per electrode, so that if
# all phosphenes are the same, their brightness would add up to 1:
all_imgs += 1.0 / num_imgs[row['electrode']] * img_drawing
all_imgs = 1 - all_imgs
# Draw array schematic with specific alpha level:
img_arr = img_warp(img_argus,
argus2out.inverse,
cval=1.0,
output_shape=out_shape)
img_arr[:, :, 3] = alpha_bg
# Replace pixels where drawings are dark enough, set alpha=1:
rr, cc = np.unravel_index(
np.where(all_imgs.ravel() < thresh_fg)[0], all_imgs.shape)
for channel in range(3):
img_arr[rr, cc, channel] = all_imgs[rr, cc]
img_arr[rr, cc, 3] = 1
ax.imshow(img_arr, cmap='gray', zorder=1)
if show_fovea:
fovea = dva2out([0, 0])[0]
ax.scatter(*fovea, s=100, marker='s', c='w', edgecolors='k', zorder=99)
if axon_map is not None:
axon_bundles = axon_map.grow_axon_bundles(n_bundles=100, prune=False)
# Draw axon pathways:
for bundle in axon_bundles:
# Flip y upside down for dva:
bundle = Watson2014Transform.ret2dva(bundle) * [1, -1]
# Trim segments outside the drawing window:
idx = np.logical_and(
np.logical_and(bundle[:, 0] >= x_min, bundle[:, 0] <= x_max),
np.logical_and(bundle[:, 1] >= y_min, bundle[:, 1] <= y_max))
bundle = dva2out(bundle)
ax.plot(bundle[idx, 0],
bundle[idx, 1],
c=(0.6, 0.6, 0.6),
linewidth=2,
zorder=1)
return ax
