def comp_pixel_ternary(image): # 3x3 and 2x2 pixel cross-correlation within image
# orthogonal comp
orthdy__ = image[1:] - image[:-1] # vertical
orthdx__ = image[:, 1:] - image[:, :-1] # horizontal
# compute gdert__
p__ = (image[:-2, :-2] + image[:-2, 1:-1] + image[1:-1, :-2] + image[1:-1, 1:-1]) * 0.25
dy__ = (orthdy__[:-1, 1:-1] + orthdy__[:-1, :-2]) * 0.5
dx__ = (orthdx__[1:-1, :-1] + orthdx__[:-2, :-1]) * 0.5
g__ = ma.hypot(dy__, dx__)
gdert__ = ma.stack((p__, g__, dy__, dx__))
# diagonal comp
diag1__ = image[2:, 2:] - image[:-2, :-2]
diag2__ = image[2:, :-2] - image[:-2, 2:]
# compute rdert__
p3__ = image[1:-1, 1:-1]
dy3__ = (orthdy__[1:, 1:-1] + orthdy__[:-1, 1:-1]) * 0.25 +\
(diag1__ + diag2__) * 0.125
dx3__ = (orthdx__[1:-1, 1:] + orthdx__[1:-1, :-1]) * 0.25 + \
(diag1__ - diag2__) * 0.125
g3__ = ma.hypot(dy3__, dx3__)
rdert__ = ma.stack((p3__, g3__, dy3__, dx3__))
# gdert__ = comp_2x2(image) # cross-compare four adjacent pixels diagonally
# rdert__ = comp_3x3(image) # compare each pixel to 8 rim pixels
return gdert__, rdert__
def comp_a(dert__, rng):
"""
Compute and compare a over predetermined range.
"""
# Unpack dert__:
if len(dert__) in (5, 12): # idert or full dert with m.
i__, g__, m__, dy__, dx__ = dert__[:5]
else: # idert or full dert without m.
i__, g__, dy__, dx__ = dert__[:4]
if len(dert__) > 10: # if ra+:
a__ = dert__[-7:-5] # Computed angle (use reverse indexing to avoid m check).
day__ = dert__[-4:-2] # Accumulated day__.
dax__ = dert__[-2:] # Accumulated day__.
else: # if fa:
# Compute angles:
a__ = ma.stack((dy__, dx__), axis=0) / g__
a__.mask = g__.mask
# Initialize dax, day:
day__, dax__ = [ma.zeros((2,) + i__.shape) for _ in range(2)]
# Compute angle differences:
da__ = translated_operation(a__, rng, angle_diff)
comp_field = central_slice(rng)
# Decompose and add to corresponding day and dax:
day__[comp_field] = (da__ * Y_COEFFS[rng]).mean(axis=-1)
dax__[comp_field] = (da__ * X_COEFFS[rng]).mean(axis=-1)
# Apply mask:
msq = np.ones(a__.shape, dtype=int) # Rim mask.
msq[comp_field] = a__.mask[comp_field] + da__.mask.sum(axis=-1) # Summed d mask.
imsq = msq.nonzero()
day__[imsq] = dax__[imsq] = ma.masked # Apply mask.
# Compute ga:
ga__ = ma.hypot(
ma.arctan2(*day__),
ma.arctan2(*dax__)
)[np.newaxis, ...] * SCALER_ga
try: # dert with m is more common:
return ma.concatenate( # Concatenate on the first dimension.
(
ma.stack((i__, g__, m__, dy__, dx__), axis=0),
a__, ga__, day__, dax__,
),
axis=0,
)
except NameError: # m doesn't exist:
return ma.concatenate( # Concatenate on the first dimension.
(
ma.stack((i__, g__, dy__, dx__), axis=0),
a__, ga__, day__, dax__,
),
axis=0,
)
def comp_a(gdert__, rng):
"""
Compute and compare a over predetermined range.
"""
# Unpack dert__:
try:
g, gg, m, dy, dx = gdert__
except ValueError: # Initial dert doesn't contain m.
g, gg, dy, dx = gdert__
# Initialize dax, day:
day, dax = [ma.zeros((2, ) + g.shape) for _ in range(2)]
# Compute angles:
a = ma.stack((dy, dx), axis=0) / gg
a.mask = gg.mask
# Compute angle differences:
da = translated_operation(a, rng, angle_diff)
comp_field = central_slice(rng)
# Decompose and add to corresponding day and dax:
day[comp_field] = (da * Y_COEFFS[rng]).mean(axis=-1)
dax[comp_field] = (da * X_COEFFS[rng]).mean(axis=-1)
# Apply mask:
msq = np.ones(a.shape, dtype=int) # Rim mask.
msq[comp_field] = a.mask[comp_field] + da.mask.sum(
axis=-1) # Summed d mask.
imsq = msq.nonzero()
day[imsq] = dax[imsq] = ma.masked # Apply mask.
# Compute ga:
ga = ma.hypot(ma.arctan2(*day), ma.arctan2(*dax))[np.newaxis,
...] * SCALER_ga
try:
return ma.concatenate( # Concatenate on the first dimension.
(
ma.stack((g, gg, m), axis=0),
ga,
day,
dax,
),
axis=0,
)
except NameError: # m doesn't exist.
return ma.concatenate( # Concatenate on the first dimension.
(
ma.stack((g, gg), axis=0),
a,
ga,
day,
dax,
),
axis=0,
)
def _calc_cddcwd(self, values, threshold, calc_type):
""" Calculates maximum number of consecutive days with daily values (precipitation) < 1mm or >= 1mm (CDD or CWD).
"""
if calc_type == 'cdd':
cmp_func = operator.lt
elif calc_type == 'cwd':
cmp_func = operator.ge
else:
self.logger.error('Unknown calculation type: %s. Aborting!',
calc_type)
raise ValueError
data_shape = values.shape[1:]
cnt = ma.zeros(data_shape)
max_cnt = ma.zeros(data_shape)
for arr in values:
mask = cmp_func(arr, threshold)
cnt += mask # Count consecutive days.
cnt *= mask # Reset counter where condition does not meet.
max_cnt = ma.max(ma.stack((max_cnt, cnt)), axis=0)
max_cnt.mask = values[
0].mask # Restore mask from the original data array.
return max_cnt
def comp_g(dert__, rng):
"""
Compare g over predetermined range.
"""
# Unpack dert__:
g, m, dy, dx = dert__
# Compare gs:
d = translated_operation(g, rng, op.sub)
comp_field = central_slice(rng)
# Decompose and add to corresponding dy and dx:
dy[comp_field] += (d * Y_COEFFS[rng]).sum(axis=-1)
dx[comp_field] += (d * X_COEFFS[rng]).sum(axis=-1)
# Compute ms:
m[comp_field] += translated_operation(g, rng, ma.minimum).sum(axis=-1)
# Apply mask:
msq = np.ones(g.shape, dtype=int) # Rim mask.
msq[comp_field] = g.mask[comp_field] + d.mask.sum(
axis=-1) # Summed d mask.
imsq = msq.nonzero()
m[imsq] = dy[imsq] = dx[imsq] = ma.masked # Apply mask.
# Compute gg:
gg = ma.hypot(dy, dx) * SCALER_g[rng]
return ma.stack((g, gg, m, dy, dx),
axis=0) # ma.stack() for extra array dimension.
def calc_rxndays(self, values, time_grid, threshold):
""" Calculates Rxnday
Arguments:
values -- array of total precipitation
time_grid -- time grid
threshold -- number of days (n)
Returns: Rxnday values
"""
# First of all, we suppose values are for a month with some time step.
# Initially, let's sum first 'threshold' days.
# Then we slide a window along the time grid
# subtracting one day on the left and adding one day on the right.
queue = deque()
nday_sum = None
max_sum = None
it_all_data = groupby(zip(values, time_grid),
key=lambda x: (x[1].day, x[1].month))
for _, one_day_group in it_all_data: # Iterate over daily groups.
daily_sum = self._calc_daily_sum(one_day_group)
queue.append(daily_sum) # Store daily sums in a queue.
if nday_sum is None:
nday_sum = ma.zeros(daily_sum.shape)
nday_sum += daily_sum # Additionally sum daily sums.
if len(queue) > threshold: # When 'threshold' days are summed...
nday_sum -= queue.popleft(
) # ...subtruct one 'left-most' daily sum from the n-day sum.
if len(queue) == threshold:
if max_sum is None:
max_sum = deepcopy(nday_sum)
else:
max_sum = ma.max(ma.stack((max_sum, nday_sum)),
axis=0) # Search for maximum value.
return max_sum
def comp_angle(a__):
"""Compare angles."""
# handle mask
if isinstance(a__, ma.masked_array):
a__.data[a__.mask] = np.nan
a__.mask = ma.nomask
# comparison
da__, a__ = translated_operation(a__, rng=rng, operator=angle_diff)
# sum within kernels
day__ = (da__ * Y_COEFFS[rng]).sum(axis=-1)
dax__ = (da__ * X_COEFFS[rng]).sum(axis=-1)
# compute gradient magnitudes (how fast angles are changing)
ga__ = np.hypot(np.arctan2(*day__), np.arctan2(*dax__))
# pack into dert
dert__ = ma.stack((*a__, ga__, *day__, *dax__), axis=0)
# handle mask
dert__.mask = np.isnan(dert__.data)
return dert__
def bbox(self, header: PoseHeader):
data = ma.transpose(self.data, axes=POINTS_DIMS)
# Split data by components, `ma` doesn't support ".split"
components = []
idx = 0
for component in header.components:
components.append(data[list(range(idx,
idx + len(component.points)))])
idx += len(component.points)
boxes = [
ma.stack([ma.min(c, axis=0), ma.max(c, axis=0)])
for c in components
]
boxes_cat = ma.concatenate(boxes)
if type(boxes_cat.mask
) == np.bool_: # Sometimes, it doesn't concatenate the mask...
boxes_mask = ma.concatenate([b.mask for b in boxes])
boxes_cat = ma.array(boxes_cat, mask=boxes_mask)
new_data = ma.transpose(boxes_cat, axes=POINTS_DIMS)
confidence_mask = np.split(new_data.mask, [-1], axis=3)[0]
confidence_mask = np.squeeze(confidence_mask, axis=-1)
confidence = np.where(confidence_mask == True, 0, 1)
return NumPyPoseBody(self.fps, new_data, confidence)
def comp_g_old(
dert__
): # cross-comp of g in 2x2 kernels, between derts in ma.stack dert__
g__, dy__, dx__ = dert__[[3, 4, 5]] # g, dy, dx -> local i, idy, idx
g__[ma.where(
g__ == 0
)] = 1 # replace 0 values with 1 to avoid error, not needed in high-g blobs?
g0__, dy0__, dx0__ = g__[:-1, :-1], dy__[:-1, :-1], dx__[:-1, :
-1] # top left
g1__, dy1__, dx1__ = g__[:-1, 1:], dy__[:-1, 1:], dx__[:-1,
1:] # top right
g2__, dy2__, dx2__ = g__[1:, 1:], dy__[1:, 1:], dx__[1:,
1:] # bottom right
g3__, dy3__, dx3__ = g__[1:, :-1], dy__[1:, :-1], dx__[1:, :
-1] # bottom left
sin0__ = dy0__ / g0__
cos0__ = dx0__ / g0__
sin1__ = dy1__ / g1__
cos1__ = dx1__ / g1__
sin2__ = dy2__ / g2__
cos2__ = dx2__ / g2__
sin3__ = dy3__ / g3__
cos3__ = dx3__ / g3__
'''
cosine of difference between diagonally opposite angles, in vector representation
print(cos_da1__.shape, type(cos_da1__))
'''
cos_da0__ = (cos2__ * cos0__) + (sin2__ * sin0__
) # top left to bottom right
cos_da1__ = (cos3__ * cos1__) + (sin3__ * sin1__
) # top right to bottom left
dgy__ = ((g3__ + g2__) - (g0__ * cos_da0__ + g1__ * cos_da1__))
# y-decomposed cosine difference between gs
dgx__ = ((g1__ + g2__) - (g0__ * cos_da0__ + g3__ * cos_da1__))
# x-decomposed cosine difference between gs
gg__ = np.hypot(dgy__, dgx__) # gradient of gradient
mg0__ = np.minimum(g0__, g2__) * cos_da0__ # g match = min(g, _g) *cos(da)
mg1__ = np.minimum(g1__, g3__) * cos_da1__
mg__ = mg0__ + mg1__
g__ = g__[:-1, :
-1] # remove last row and column to align with derived params
dy__ = dy__[:-1, :-1]
dx__ = dx__[:-1, :-1] # -> idy, idx to compute cos for comp rg
# no longer needed: g__.mask = dy__.mask = dx__.mask = gg__.mask?
'''
next comp_rg will use g, dy, dx
next comp_gg will use gg, dgy, dgx
'''
return ma.stack((g__, dy__, dx__, gg__, dgy__, dgx__, mg__))
def read_v0_0(cls, header: PoseHeader, reader: BufferReader):
fps, _frames = reader.unpack(ConstStructs.double_ushort)
_dims = max([len(c.format) for c in header.components]) - 1
_points = sum([len(c.points) for c in header.components])
frames_d = []
frames_c = []
for _ in range(_frames):
_people = reader.unpack(ConstStructs.ushort)
people_d = []
people_c = []
for pid in range(_people):
reader.advance(ConstStructs.short) # Skip Person ID
person_d = []
person_c = []
for component in header.components:
points = np.array(
reader.unpack_numpy(
ConstStructs.float,
(len(component.points), len(component.format))))
dimensions, confidence = np.split(points, [-1], axis=1)
boolean_confidence = np.where(confidence > 0, 0,
1) # To create the mask
mask = np.column_stack(
tuple([boolean_confidence] *
(len(component.format) - 1)))
person_d.append(ma.masked_array(dimensions, mask=mask))
person_c.append(np.squeeze(confidence, axis=-1))
if pid == 0:
people_d.append(ma.concatenate(person_d))
people_c.append(np.concatenate(person_c))
# In case no person, should all be zeros
if len(people_d) == 0:
people_d.append(np.zeros((_points, _dims)))
people_c.append(np.zeros(_points))
frames_d.append(ma.stack(people_d))
frames_c.append(np.stack(people_c))
return cls(fps, ma.stack(frames_d), ma.stack(frames_c))
def comp_a(dert__, fga): # cross-comp of a or aga in 2x2 kernels
'''
if fga: dert = (g, gg, dgy, dgx, gm, ?(iga, iday, idax)
else: dert = (i, g, dy, dx, ?m)
'''
dert__ = shape_check(dert__) # remove derts of incomplete kernels
i__, g__, dy__, dx__, = dert__[0:4]
if fga: # input is adert
ga__, day__, dax__ = dert__[4], dert__[5:7], dert__[7:9]
a__ = [day__[0], day__[1], dax__[0], dax__[1]] / ga__
else:
a__ = [dy__, dx__] / g__ # similar to calc_a
# each shifted a in 2x2 kernel
a__topleft = a__[:, :-1, :-1]
a__topright = a__[:, :-1, 1:]
a__botright = a__[:, 1:, 1:]
a__botleft = a__[:, 1:, :-1]
# diagonal angle differences:
sin_da0__, cos_da0__ = angle_diff(a__topleft, a__botright, fga)
sin_da1__, cos_da1__ = angle_diff(a__topright, a__botleft, fga)
ma__ = np.hypot(sin_da0__ + 1, cos_da0__ + 1) + np.hypot(sin_da1__ + 1, cos_da1__ + 1)
# ma = inverse angle match = SAD: covert sin and cos da to 0->2 range
day__ = (-sin_da0__ - sin_da1__), (cos_da0__ + cos_da1__)
# angle change in y, sines are sign-reversed because da0 and da1 are top-down, no reversal in cosines
dax__ = (-sin_da0__ + sin_da1__), (cos_da0__ + cos_da1__)
# angle change in x, positive sign is right-to-left, so only sin_da0__ is sign-reversed
'''
sin(-θ) = -sin(θ), cos(-θ) = cos(θ):
sin(da) = -sin(-da), cos(da) = cos(-da) => (sin(-da), cos(-da)) = (-sin(da), cos(da))
'''
ga__ = np.hypot(np.arctan2(*day__), np.arctan2(*dax__))
# angle gradient, a scalar, to evaluate for comp_aga
adert__ = ma.stack((i__[:-1, :-1], # for summation in Dert
g__[:-1, :-1], # for summation in Dert
dy__[:-1, :-1], # passed on as idy
dx__[:-1, :-1], # passed on as idx # no use for m__[:-1, :-1]?
ga__,
*day__,
*dax__,
ma__,
cos_da0__,
cos_da1__
))
'''
next comp_g will use g, cos_da0__, cos_da1__, dy, dx (passed to comp_rg as idy, idx)
next comp_a will use ga, day, dax # comp_aga
'''
return adert__
def comp_2x2(image):
"""Deprecated."""
dy__ = (image[1:-1, 1:-1] + image[1:-1, :-2]
- image[:-2, 1:-1] - image[:-2, :-2]) * 0.5
dx__ = (image[1:-1, 1:-1] + image[:-2, 1:-1]
- image[1:-1, :-2] - image[:-2, :-2]) * 0.5
# sum pixel values and reconstruct central pixel as their average:
p__ = (image[:-2, :-2] + image[:-2, 1:-1]
+ image[1:-1, :-2] + image[1:-1, 1:-1]) * 0.25
g__ = np.hypot(dy__, dx__) # compute gradients per kernel, converted to 0-255 range
return ma.stack((p__, g__, dy__, dx__))
def comp_3x3(image):
# initialization
dy__ = np.zeros((image.shape[0] - 2, image.shape[1] - 2))
dx__ = np.zeros((image.shape[0] - 2, image.shape[1] - 2))
# start in 2nd pixels and end at 2nd last pixels
# 1st and last pixels doesn't have sufficient 8 surrounding pixels
for y in range(image.shape[0] - 2):
for x in range(image.shape[1] - 2):
# 3x3 kernel, given 9 pixels below:
# p1 p2 p3
# p8 c1 p4
# p7 p6 p5
# directional differences, d = p - c
# dx = sum(8 directional differences * x coefficients)
# dy = sum(8 directional differences * y coefficients)
# p__ = central pixel , which is c1
# get difference in 8 directions = directional pixels - center pixel
d_top_left = (image[y][x] - image[y + 1][x + 1])
d_top = (image[y][x + 1] - image[y + 1][x + 1])
d_top_right = (image[y][x + 2] - image[y + 1][x + 1])
d_right = (image[y + 1][x + 2] - image[y + 1][x + 1])
d_bottom_right = (image[y + 2][x + 2] - image[y + 1][x + 1])
d_bottom = (image[y + 2][x + 1] - image[y + 1][x + 1])
d_bottom_left = (image[y + 2][x] - image[y + 1][x + 1])
d_left = (image[y + 1][x] - image[y + 1][x + 1])
dy__[y,x] = (d_top_left * YCOEF[0]) +\
(d_top * YCOEF[1]) +\
(d_top_right * YCOEF[2]) +\
(d_right * YCOEF[3]) +\
(d_bottom_right * YCOEF[4]) +\
(d_bottom * YCOEF[5]) +\
(d_bottom_left * YCOEF[6]) +\
(d_left * YCOEF[7])
dx__[y,x] = (d_top_left * XCOEF[0]) +\
(d_top * XCOEF[1]) +\
(d_top_right * XCOEF[2]) +\
(d_right * XCOEF[3]) +\
(d_bottom_right * XCOEF[4]) +\
(d_bottom * XCOEF[5]) +\
(d_bottom_left * XCOEF[6]) +\
(d_left * XCOEF[7])
p__ = image[1:-1, 1:-1] # central pixels
g__ = np.hypot(
dy__, dx__) # compute gradients per kernel, converted to 0-255 range
return ma.stack((p__, g__, dy__, dx__))
def translated_array(a, rng):
"""
Like translated_operation, but without applying operation to slices
"""
out = ma.stack([a[ts] for ts in TRANSLATING_SLICES_[rng]])
# Rearrange axes:
for dim1, dim2 in pairwise(range(out.ndim)):
out = out.swapaxes(dim1, dim2)
return out
def comp_pixel_old(image): # 2x2 pixel cross-correlation within image
dy__ = image[1:] - image[:-1] # orthogonal vertical com
dx__ = image[:, 1:] - image[:, :-1] # orthogonal horizontal comp
p__ = image[:-1, :-1] # top-left pixel
mean_dy__ = (dy__[:, 1:] + dy__[:, :-1]) * 0.5 # mean dy per kernel
mean_dx__ = (dx__[1:, :] + dx__[:-1, :]) * 0.5 # mean dx per kernel
g__ = ma.hypot(mean_dy__, mean_dx__) # central gradient of four rim pixels
dert__ = ma.stack((p__, g__, mean_dy__, mean_dx__))
return dert__
def comp_g(dert__): # add fga if processing in comp_ga is different?
"""
Cross-comp of g or ga in 2x2 kernels, between derts in ma.stack dert__:
input dert = (i, g, dy, dx, ga, dyy, dxy, dyx, dxx, ma, cos_da0, cos_da1)
output dert = (g, gg, dgy, dgx, gm, ga, day, dax, dy, dx)
"""
dert__ = shape_check(dert__) # remove derts of incomplete kernels
g__, cos_da0__, cos_da1__ = dert__[[1, -2, -1]] # top dimension of numpy stack must be a list
cos_da0__ = cos_da0__[:-1, :-1]
cos_da1__ = cos_da1__[:-1, :-1]
g_topleft__ = g__[:-1, :-1]
g_topright__ = g__[:-1, 1:]
g_bottomleft__ = g__[1:, :-1]
g_bottomright__ = g__[1:, 1:]
dgy__ = ((g_bottomleft__ + g_bottomright__) -
(g_topleft__ * cos_da0__ + g_topright__ * cos_da1__))
# y-decomposed cosine difference between gs
dgx__ = ((g_topright__ + g_bottomright__) -
(g_topleft__ * cos_da0__ + g_bottomleft__ * cos_da1__))
# x-decomposed cosine difference between gs
gg__ = np.hypot(dgy__, dgx__) # gradient of gradient
mg0__ = np.minimum(g_topleft__, (g_bottomright__ * cos_da0__)) # g match = min(g, _g*cos(da))
mg1__ = np.minimum(g_topright__, (g_bottomleft__ * cos_da1__))
mg__ = mg0__ + mg1__
gdert = ma.stack((g__[:-1, :-1], # remove last row and column to align with derived params
gg__,
dgy__,
dgx__,
mg__,
dert__[4][:-1, :-1], # ga__
dert__[5][:-1, :-1], # dayy
dert__[6][:-1, :-1], # daxy
dert__[7][:-1, :-1], # dayx
dert__[8][:-1, :-1], # daxx
dert__[9][:-1, :-1], # ma__
dert__[2][:-1, :-1], # idy__
dert__[3][:-1, :-1] # idx__
))
'''
next comp_r will use g, idy, idx # comp_rg
next comp_a will use ga, day, dax # comp_agg, also dgy__, dgx__ as idy, idx?
'''
return gdert
def comp_pixel(
image
): # comparison between pixel and its neighbours within kernel, for the whole image
# Initialize variables:
if kwidth == 2:
# Compare:
dy__ = (image[1:, 1:] -
image[:-1, 1:]) + (image[1:, :-1] - image[:-1, :-1]) * 0.5
dx__ = (image[1:, 1:] -
image[1:, :-1]) + (image[:-1, 1:] - image[:-1, :-1]) * 0.5
# Sum pixel values:
p__ = (image[:-1, :-1] + image[:-1, 1:] + image[1:, :-1] +
image[1:, 1:]) * 0.25
else:
ycoef = np.array([-0.5, -1, -0.5, 0, 0.5, 1, 0.5, 0])
xcoef = np.array([-0.5, 0, 0.5, 1, 0.5, 0, -0.5, -1])
# Compare by subtracting centered image from translated image:
d___ = np.array(
list(
map(
lambda trans_slices: image[trans_slices] - image[
1:-1, 1:-1], [
(slice(None, -2), slice(None, -2)),
(slice(None, -2), slice(1, -1)),
(slice(None, -2), slice(2, None)),
(slice(1, -1), slice(2, None)),
(slice(2, None), slice(2, None)),
(slice(2, None), slice(1, -1)),
(slice(2, None), slice(None, -2)),
(slice(1, -1), slice(None, -2)),
]))).swapaxes(0, 2).swapaxes(0, 1)
# Decompose differences:
dy__ = (d___ * ycoef).sum(axis=2)
dx__ = (d___ * xcoef).sum(axis=2)
# Sum pixel values:
p__ = image[1:-1, 1:-1]
# Compute gradient magnitudes per kernel:
g__ = np.hypot(dy__, dx__) * 0.354801226089485
# no m__ = MAX_G - g__, immediate vm = -vg - Ave
return ma.around(ma.stack((p__, g__, dy__, dx__), axis=0))
def comp_pixel_m(image): # current version of 2x2 pixel cross-correlation within image
# following four slices provide inputs to a sliding 2x2 kernel:
topleft__ = image[:-1, :-1]
topright__ = image[:-1, 1:]
botleft__ = image[1:, :-1]
botright__ = image[1:, 1:]
dy__ = ((botleft__ + botright__) - (topleft__ + topright__)) # same as diagonal from left
dx__ = ((topright__ + botright__) - (topleft__ + botleft__)) # same as diagonal from right
g__ = np.hypot(dy__, dx__) # gradient per kernel
# inverse match = SAD: measure of variation within kernel
m__ = ( abs(topleft__ - botright__) + abs(topright__ - botleft__))
return ma.stack((topleft__, g__, dy__, dx__, m__))
def comp_r(dert__, fig):
i__, g__, dy__, dx__ = dert__[:]
dy__ = dy__[:, rng:-rng, rng:-rng]
dx__ = dx__[:, rng:-rng, rng:-rng]
# comparison
d__ = translated_operation(i__, rng=rng, operator=op.sub)
# sum within kernels
dy__ += (d__ * Y_COEFFS[rng]).sum(axis=-1)
dx__ += (d__ * X_COEFFS[rng]).sum(axis=-1)
# compute gradient magnitudes
g__ = ma.hypot(dy__, dx__)
return ma.stack((i__, g__, dy__, dx__))
def comp_3x3(image):
"""Deprecated."""
d___ = np.array( # subtract centered image from translated image:
[image[ts2] - image[ts1] for ts1, ts2 in TRANSLATING_SLICES_PAIRS_3x3]
).swapaxes(0, 2).swapaxes(0, 1)
# 3rd dimension: sequence of differences between pairs of
# diametrically opposed pixels corresponding to:
# |--(clockwise)--+ |--(clockwise)--+
# YCOEF: -0.5 -1 -0.5 ¦ XCOEF: -0.5 0 0.5 ¦
# 0 0 ¦ -1 1 ¦
# 0.5 1 0.5 ¦ -0.5 0 0.5 ¦
# <<----+ <<----+
# Decompose differences into dy and dx, same as Gy and Gx in conventional edge detection operators:
dy__ = (d___ * YCOEF).sum(axis=2)
dx__ = (d___ * XCOEF).sum(axis=2)
p__ = image[1:-1, 1:-1]
g__ = np.hypot(dy__, dx__) # compute gradients per kernel, converted to 0-255 range
return ma.stack((p__, g__, dy__, dx__))
def comp_pixel(image): # 3x3 or 2x2 pixel cross-correlation within image
if kwidth == 2: # cross-compare four adjacent pixels diagonally:
dy__ = (image[1:, 1:] - image[:-1, 1:]) + (image[1:, :-1] - image[:-1, :-1]) * 0.5
dx__ = (image[1:, 1:] - image[1:, :-1]) + (image[:-1, 1:] - image[:-1, :-1]) * 0.5
# or no coef: distance 1.41 * angle .705 -> 1? and conversion only for extended kernel, if centered?
# sum pixel values and reconstruct central pixel as their average:
p__ = (image[:-1, :-1] + image[:-1, 1:] + image[1:, :-1] + image[1:, 1:]) * 0.25
else: # kwidth == 3, compare central pixel to 8 rim pixels, current default option
ycoef = np.array([-0.5, -1, -0.5, 0, 0.5, 1, 0.5, 0])
xcoef = np.array([-0.5, 0, 0.5, 1, 0.5, 0, -0.5, -1])
d___ = np.array(list( # subtract centered image from translated image:
map(lambda trans_slices: image[trans_slices] - image[1:-1, 1:-1],
[
(slice(None, -2), slice(None, -2)),
(slice(None, -2), slice(1, -1)),
(slice(None, -2), slice(2, None)),
(slice(1, -1), slice(2, None)),
(slice(2, None), slice(2, None)),
(slice(2, None), slice(1, -1)),
(slice(2, None), slice(None, -2)),
(slice(1, -1), slice(None, -2)),
]
)
)).swapaxes(0, 2).swapaxes(0, 1)
# Decompose differences into dy and dx, same as Gy and Gx in conventional edge detection operators:
dy__ = (d___ * ycoef).sum(axis=2)
dx__ = (d___ * xcoef).sum(axis=2)
p__ = image[1:-1, 1:-1]
g__ = np.hypot(dy__, dx__) * 0.354801226089485 # compute gradients per kernel, converted to 0-255 range
return ma.around(ma.stack((p__, g__, dy__, dx__), axis=0))
def comp_pixel(image): # comparison of central pixel to rim pixels in 2x2 or 3x3 kernel, for the whole image
if kwidth == 2:
# Cross-compare four adjacent pixels:
dy__ = (image[1:, 1:] - image[:-1, 1:]) + (image[1:, :-1] - image[:-1, :-1]) * 0.5
dx__ = (image[1:, 1:] - image[1:, :-1]) + (image[:-1, 1:] - image[:-1, :-1]) * 0.5
# Sum pixel values:
p__ = (image[:-1, :-1]
+ image[:-1, 1:]
+ image[1:, :-1]
+ image[1:, 1:]) * 0.25
else:
ycoef = np.array([-0.5, -1, -0.5, 0, 0.5, 1, 0.5, 0])
xcoef = np.array([-0.5, 0, 0.5, 1, 0.5, 0, -0.5, -1])
# Compare by subtracting centered image from translated image:
d___ = np.array(list(
map(lambda trans_slices: image[trans_slices] - image[1:-1, 1:-1],
[
(slice(None, -2), slice(None, -2)),
(slice(None, -2), slice(1, -1)),
(slice(None, -2), slice(2, None)),
(slice(1, -1), slice(2, None)),
(slice(2, None), slice(2, None)),
(slice(2, None), slice(1, -1)),
(slice(2, None), slice(None, -2)),
(slice(1, -1), slice(None, -2)),
]
)
)).swapaxes(0, 2).swapaxes(0, 1)
# Decompose differences:
dy__ = (d___ * ycoef).sum(axis=2)
dx__ = (d___ * xcoef).sum(axis=2)
p__ = image[1:-1, 1:-1]
# Compute gradients per kernel:
g__ = np.hypot(dy__, dx__) * 0.354801226089485 # no m__ = MAX_G - g__
return ma.around(ma.stack((p__, g__, dy__, dx__), axis=0))
def comp_3x3_loop(image):
buff_ = deque() # buffer for derts
Y, X = image.shape
for p_ in image: # loop through rows
for p in p_: # loop through each pixel
dx = dy = 0 # uni-lateral differences
# loop through buffers:
for k, xcoeff, ycoeff in zip(
[0, X - 1, X, X + 1], # indices of _dert
[0.25, -0.125, 0, 0.125], # x axis coefficient
[0, 0.125, 0.25, 0.125]): # y axis coefficient
try:
_p, _dy, _dx = buff_[k] # unpack buff_[k]
d = p - _p # compute difference
dx_buff = d * xcoeff # decompose difference
dy_buff = d * ycoeff
dx += dx_buff # accumulate fuzzy difference over the kernel
_dx += dx_buff
dy += dy_buff
_dy += dy_buff
buff_[k] = _p, _dy, _dx # repack buff_[k]
except TypeError: # buff_[k] is None
pass
except IndexError: # k >= len(buff_)
break
buff_.appendleft(
(p, dy, dx)) # initialize dert with uni-lateral differences
buff_.appendleft(None) # add empty dert at the end of each row
# reshape data and compute g (temporary, to perform tests)
p__, dy__, dx__ = np.array([buff for buff in reversed(buff_)
if buff is not None])\
.reshape(Y, X, 3).swapaxes(1, 2).swapaxes(0, 1)[:, 1:-1, 1:-1]
g__ = ma.hypot(dy__, dx__)
return ma.stack((p__, g__, dy__, dx__))
def comp_2x2(image):
# initialize empty array
dx__ = np.zeros((
image.shape[0] - 1,
image.shape[1] - 1,
))
dy__ = np.zeros((
image.shape[0] - 1,
image.shape[1] - 1,
))
p__ = np.zeros((
image.shape[0] - 1,
image.shape[1] - 1,
))
# slide across each pixel, compute dy,dx and reconstruct central pixel
for y in range(image.shape[0] - 1):
for x in range(image.shape[1] - 1):
# 2x2 kernel, given 4 pixels below:
# p1 p2
# p3 p4
# dx = ((p2-p3) + (p4-p1)) /2 (get mean of diagonal difference)
# dy = ((p4-p1) + (p3-p2)) /2 (get mean of diagonal difference)
# p__ = (p1+p2+p3+p4) /4
dx__[y,x] = ((image[y,x+1] - image[y+1,x]) + \
(image[y+1,x+1] - image[y,x]))/2
dy__[y,x] = ((image[y+1,x+1] - image[y,x]) + \
(image[y+1,x] - image[y,x+1]))/2
p__[y,x] = ((image[y,x] + image[y+1,x]) + \
(image[y,x+1] + image[y+1,x+1]))/4
g__ = np.hypot(
dy__, dx__) # compute gradients per kernel, converted to 0-255 range
return ma.stack((p__, g__, dy__, dx__))
def comp_pixel(
image
): # 2x2 pixel cross-correlation within image, as in edge detection operators
# input slices into sliding 2x2 kernel, each slice is a shifted 2D frame of grey-scale pixels:
topleft__ = image[:-1, :-1]
topright__ = image[:-1, 1:]
bottomleft__ = image[1:, :-1]
bottomright__ = image[1:, 1:]
Gy__ = ((bottomleft__ + bottomright__) - (topleft__ + topright__)
) # same as decomposition of two diagonal differences into Gy
Gx__ = ((topright__ + bottomright__) - (topleft__ + bottomleft__)
) # same as decomposition of two diagonal differences into Gx
G__ = np.hypot(
Gy__,
Gx__) # central gradient per kernel, between its four vertex pixels
# why ma?
return ma.stack(
(topleft__, G__, Gy__,
Gx__)) # tuple of 2D arrays per param of dert (derivatives' tuple)
def comp_g(dert__, odd):
g__ = dert__[0]
a__ = dert__[1:]
# loop through each pair of comparands in a kernel
dgy__ = ma.zeros(np.subtract(g.shape, rng))
dgx__ = ma.zeros(np.subtract(g.shape, rng))
for x_coeff, y_coeff, (ts, _ts) in zip(X_COEFFS[rng], Y_COEFFS[rng],
TRANSLATING_SLICES_PAIRS_[rng]):
# find angle differences
da__ = angle_diff(a__[ts], a__[_ts])
# compute dg: dg = g - _g * cos(da) at each position
dg__ = g__[ts] - g__[_ts] * da__[1]
# accumulate dgy, dgx
dgx__ += dg__ * x_coeff
dgy__ += dg__ * y_coeff
gg__ = ma.hypot(dgy__, dgx__)
return ma.stack((g__, gg__, dgy__, dgx__))
def write_trace_species_raster(data, profile, outname):
"""
This function takes a list of numpy arrays (each representing a trace species)
along with a profile dictionary and an output filename.
It calculates the sum of the values of each layer of the stacked array - ie of the pixels "lying on top of each other"
/ or along axis=0. It then writes this aggregate 2D array as a new raster.
Effectively this represents the cumulative contributions of all species masked as "trace" in each pixel.
"""
# Stack the input data
all_layers = ma.stack(data)
# Add up all the pixels lying "on top of" each other in the stack
layers_sum = all_layers.sum(axis=0)
print(
f'\n Writing output raster with summed trace species contributions: {outname}\n'
)
# Write the data
with rio.open(outname, 'w', **profile) as dest:
dest.write(layers_sum.filled(-9999))
return
def _run(self, calc_mode):
""" Main method of the class. Reads data arrays, process them and returns results. """
self.logger.info('Started!')
self.logger.info('Calculation mode: %s', calc_mode)
# Get inputs
input_uids = self._data_helper.input_uids()
assert input_uids, '(CalcBasicStat::run) No input arguments!'
# Get parameters
parameters = None
if len(input_uids) == MAX_N_INPUT_ARGUMENTS:
parameters = self._data_helper.get(
input_uids[INPUT_PARAMETERS_INDEX])
# Check parameters
if not calc_mode in parameters:
self.logger.error(
'Error! No parameter \'%s\' in module parameters! Check task-file!',
calc_mode)
raise ValueError
# Get outputs
output_uids = self._data_helper.output_uids()
assert output_uids, '(CalcBasicStat::run) No output arguments!'
# Get time segments and levels
time_segments = self._data_helper.get_segments(input_uids[0])
vertical_levels = self._data_helper.get_levels(input_uids[0])
# Get input data info and pass through units to the result description.
data_info = self._data_helper.get_data_info(input_uids[0])
input_description = data_info['description']
result_description = {
'@name': input_description['@name'],
'@units': input_description['@units']
}
# Make desired statistical function shortcut for segment and final processing .
if calc_mode == 'timeMean':
seg_stat_func = ma.mean
final_stat_func = ma.mean
final_title = 'Average of ' + input_description['@title']
elif calc_mode == 'timeMin':
seg_stat_func = ma.min
final_stat_func = ma.min
final_title = 'Minimum of ' + input_description['@title']
elif calc_mode == 'timeMax':
seg_stat_func = ma.max
final_stat_func = ma.max
final_title = 'Maximum of ' + input_description['@title']
elif calc_mode == 'timeMeanPrec':
seg_stat_func = ma.sum
final_stat_func = ma.mean
final_title = 'Average sum of ' + input_description['@title']
for level in vertical_levels:
all_segments_data = []
for segment in time_segments:
one_segment_time_grid = []
# Get data
result = self._data_helper.get(input_uids[0],
segments=segment,
levels=level)
# Daily statistics.
if parameters[calc_mode] == 'day':
one_segment_data = []
data_time_iter = itertools.zip_longest(
result['data'][level][segment['@name']]['@values'],
result['data'][level][segment['@name']]['@time_grid'])
for date_key, group in itertools.groupby(
data_time_iter, key=self._data_time_key):
group_data = []
for data, _ in group:
group_data.append(data)
group_data = ma.stack(group_data)
one_segment_time_grid.append(date_key)
# Calulate time statistics for a current time group (day)
one_segment_data.append(
seg_stat_func(group_data, axis=0))
one_segment_data = ma.stack(one_segment_data)
# Calculate time statistics for a current time segment
if (parameters[calc_mode] == 'data') or (parameters[calc_mode]
== 'segment'):
if len(result['data'][level][segment['@name']]
['@time_grid']) > 1:
one_segment_data = seg_stat_func(
result['data'][level][segment['@name']]['@values'],
axis=0)
else:
one_segment_data = result['data'][level][
segment['@name']]['@values']
mid_time = result['data'][level][segment['@name']]['@time_grid'][0] + \
(result['data'][level][segment['@name']]['@time_grid'][-1] - \
result['data'][level][segment['@name']]['@time_grid'][0]) / 2
one_segment_time_grid.append(mid_time)
# For segment-wise averaging send to the output current time segment results
# or store them otherwise.
if (parameters[calc_mode] == 'day') or (parameters[calc_mode]
== 'segment'):
self._data_helper.put(output_uids[0],
values=one_segment_data,
level=level,
segment=segment,
longitudes=result['@longitude_grid'],
latitudes=result['@latitude_grid'],
times=one_segment_time_grid,
fill_value=result['@fill_value'],
meta=result['meta'],
description=result_description)
elif parameters[calc_mode] == 'data':
all_segments_data.append(one_segment_data)
# For data-wise analysis analyse segments analyses :)
if parameters[calc_mode] == 'data':
data_out = final_stat_func(ma.stack(all_segments_data), axis=0)
# Make a global segment covering all input time segments
full_range_segment = copy(
time_segments[0]
) # Take the beginning of the first segment...
full_range_segment['@ending'] = time_segments[-1][
'@ending'] # and the end of the last one.
full_range_segment[
'@name'] = 'GlobalSeg' # Give it a new name.
# Correct title and name
if final_title is not None:
result_description['@title'] = final_title
self._data_helper.put(output_uids[0],
values=data_out,
level=level,
segment=full_range_segment,
longitudes=result['@longitude_grid'],
latitudes=result['@latitude_grid'],
fill_value=result['@fill_value'],
meta=result['meta'],
description=result_description)
self.logger.info('Finished!')
def run(self):
""" Main method of the class. Reads data arrays, process them and returns results. """
self.logger.info('Started!')
# Get inputs
input_uids = self._data_helper.input_uids()
assert input_uids, 'Error! No input arguments!'
# Get parameters
parameters = None
if len(input_uids
) == MAX_N_INPUT_ARGUMENTS: # If parameters are given.
parameters = self._data_helper.get(
input_uids[INPUT_PARAMETERS_INDEX])
threshold = float(
self._get_parameter('Threshold', parameters, DEFAULT_VALUES))
calc_mode = self._get_parameter('Mode', parameters, DEFAULT_VALUES)
self.logger.info('Threshold: %s', threshold)
self.logger.info('Calculation mode: %s', calc_mode)
# Get outputs
output_uids = self._data_helper.output_uids()
assert output_uids, 'Error! No output arguments!'
# Get time segments and levels
time_segments = self._data_helper.get_segments(input_uids[DATA_UID])
vertical_levels = self._data_helper.get_levels(input_uids[DATA_UID])
data_func = ma.max # For calc_mode == 'data' we calculate max over all segments.
# Set result units.
result_description = deepcopy(
self._data_helper.get_data_info(input_uids[0])['description'])
result_description[
'@title'] = 'Count of days with precipitation >= {} mm'.format(
threshold)
result_description['@units'] = 'days'
# Main loop
for level in vertical_levels:
all_segments_data = []
for segment in time_segments:
# Read data
data = self._data_helper.get(input_uids[DATA_UID],
segments=segment,
levels=level)
values = data['data'][level][segment['@name']]['@values']
time_grid = data['data'][level][segment['@name']]['@time_grid']
one_segment_data = self.calc_rnnmm(values, time_grid,
threshold)
# For segment-wise averaging send to the output current time segment results
# or store them otherwise.
if calc_mode == 'segment':
self._data_helper.put(output_uids[0],
values=one_segment_data,
level=level,
segment=segment,
longitudes=data['@longitude_grid'],
latitudes=data['@latitude_grid'],
fill_value=data['@fill_value'],
description=result_description,
meta=data['meta'])
elif calc_mode == 'data':
all_segments_data.append(one_segment_data)
else:
self.logger.error(
'Error! Unknown calculation mode: \'%s\'', calc_mode)
raise ValueError
# For data-wise analysis analyse segments analyses :)
if calc_mode == 'data':
data_out = data_func(ma.stack(all_segments_data), axis=0)
# Make a global segment covering all input time segments
full_range_segment = deepcopy(
time_segments[0]
) # Take the beginning of the first segment...
full_range_segment['@ending'] = time_segments[-1][
'@ending'] # and the end of the last one.
full_range_segment[
'@name'] = 'GlobalSeg' # Give it a new name.
self._data_helper.put(output_uids[0],
values=data_out,
level=level,
segment=full_range_segment,
longitudes=data['@longitude_grid'],
latitudes=data['@latitude_grid'],
fill_value=data['@fill_value'],
meta=data['meta'],
description=result_description)
self.logger.info('Finished!')
swe_per_level.append(masked_data)
if parameter == 191 and level_type == 'sfc' and level in [
721, 722, 723
]:
print('Found SR, layer {}'.format(level))
raw_data = grib_message['values']
masked_data = ma.masked_where(
raw_data == grib_message['missingValue'], raw_data)
sr_per_level.append(masked_data)
print('Snow depth calculation...')
swe_all_levels = ma.stack(swe_per_level)
sr_all_levels = ma.stack(sr_per_level)
snow_depth = ma.sum(swe_all_levels / sr_all_levels, axis=0)
print('Generating output GRIB')
if out_message is not None:
missing_value = out_message['missingValue']
out_message['values'] = snow_depth.filled(fill_value=missing_value)
out_message['level'] = 721 # FIXME: correct level?
out_message['indicatorOfParameter'] = 141 # FIXME: correct parameter id?
with open(outfile_snow_depth, 'wb') as out_file:
out_message.write(out_file)
