separate HSF

This commit is contained in:
Prohurtz 2022-12-26 09:05:40 -08:00
parent a7e9f007a6
commit 7ad7ada8f5
2 changed files with 562 additions and 417 deletions

View File

@ -49,14 +49,9 @@ from one_euro_filter import OneEuroFilter
if sys.platform.startswith("win"):
from winsound import PlaySound, SND_FILENAME, SND_ASYNC
import functools
import math
import os
import timeit
from collections import namedtuple
from functools import lru_cache
from osc_calibrate_filter import *
from haar_surround_feature import *
class InformationOrigin(Enum):
RANSAC = 1
@ -93,416 +88,6 @@ async def delayed_setting_change(setting, value):
if sys.platform.startswith("win"):
PlaySound('Audio/compleated.wav', SND_FILENAME | SND_ASYNC)
#HSF \/
# cache param
lru_maxsize_vvs = 16
lru_maxsize_vs = 64
# CV param
default_radius = 20
auto_radius_range = (default_radius - 10, default_radius + 10) # (10,30)
blink_init_frames = 60 * 3 # 60fps*3sec,Number of blink statistical frames
# step==(x,y)
default_step = (5, 5) # bigger the steps,lower the processing time! ofc acc also takes an impact
response_list = []
"""
Attention.
If using cv2.filter2D in this code, be careful with the kernel
https://stackoverflow.com/questions/39457468/convolution-without-any-padding-opencv-python
"""
def TimeitWrapper(*args, **kwargs):
"""
This decorator @TimeitWrapper() prints the function name and execution time in seconds.
:param args:
:param kwargs:
:return:
"""
def decorator(function):
@functools.wraps(function)
def wrapper(*args, **kwargs):
start = timeit.default_timer()
results = function(*args, **kwargs)
end = timeit.default_timer()
print('{} execution time: {:.10f} s'.format(function.__name__, end - start))
return results
return wrapper
return decorator
class TimeitResult(object):
"""
from https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L55
Object returned by the timeit magic with info about the run.
Contains the following attributes :
loops: (int) number of loops done per measurement
repeat: (int) number of times the measurement has been repeated
best: (float) best execution time / number
all_runs: (list of float) execution time of each run (in s)
"""
def __init__(self, loops, repeat, best, worst, all_runs, precision):
self.loops = loops
self.repeat = repeat
self.best = best
self.worst = worst
self.all_runs = all_runs
self._precision = precision
self.timings = [dt / self.loops for dt in all_runs]
@property
def average(self):
return math.fsum(self.timings) / len(self.timings)
@property
def stdev(self):
mean = self.average
return (math.fsum([(x - mean) ** 2 for x in self.timings]) / len(self.timings)) ** 0.5
def __str__(self):
pm = '+-'
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb1'.encode(sys.stdout.encoding)
pm = u'\xb1'
except:
pass
return "min:{best} max:{worst} mean:{mean} {pm} {std} per loop (mean {pm} std. dev. of {runs} run{run_plural}, {loops:,} loop{loop_plural} each)".format(
pm=pm,
runs=self.repeat,
loops=self.loops,
loop_plural="" if self.loops == 1 else "s",
run_plural="" if self.repeat == 1 else "s",
mean=format_time(self.average, self._precision),
std=format_time(self.stdev, self._precision),
best=format_time(self.best, self._precision),
worst=format_time(self.worst, self._precision),
)
def _repr_pretty_(self, p, cycle):
unic = self.__str__()
p.text(u'<TimeitResult : ' + unic + u'>')
class FPSResult(object):
"""
base https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L55
"""
def __init__(self, loops, repeat, best, worst, all_runs, precision):
self.loops = loops
self.repeat = repeat
self.best = 1 / best
self.worst = 1 / worst
self.all_runs = all_runs
self._precision = precision
self.fps = [1 / dt for dt in all_runs]
self.unit = "fps"
@property
def average(self):
return math.fsum(self.fps) / len(self.fps)
@property
def stdev(self):
mean = self.average
return (math.fsum([(x - mean) ** 2 for x in self.fps]) / len(self.fps)) ** 0.5
def __str__(self):
pm = '+-'
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb1'.encode(sys.stdout.encoding)
pm = u'\xb1'
except:
pass
return "min:{best} max:{worst} mean:{mean} {pm} {std} per loop (mean {pm} std. dev. of {runs} run{run_plural}, {loops:,} loop{loop_plural} each)".format(
pm=pm,
runs=self.repeat,
loops=self.loops,
loop_plural="" if self.loops == 1 else "s",
run_plural="" if self.repeat == 1 else "s",
mean="%.*g%s" % (self._precision, self.average, self.unit),
std="%.*g%s" % (self._precision, self.stdev, self.unit),
best="%.*g%s" % (self._precision, self.best, self.unit),
worst="%.*g%s" % (self._precision, self.worst, self.unit),
)
def _repr_pretty_(self, p, cycle):
unic = self.__str__()
p.text(u'<FPSResult : ' + unic + u'>')
def format_time(timespan, precision=3):
"""
https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L1473
Formats the timespan in a human readable form
"""
if timespan >= 60.0:
# we have more than a minute, format that in a human readable form
# Idea from http://snipplr.com/view/5713/
parts = [("d", 60 * 60 * 24), ("h", 60 * 60), ("min", 60), ("s", 1)]
time = []
leftover = timespan
for suffix, length in parts:
value = int(leftover / length)
if value > 0:
leftover = leftover % length
time.append(u'%s%s' % (str(value), suffix))
if leftover < 1:
break
return " ".join(time)
# Unfortunately the unicode 'micro' symbol can cause problems in
# certain terminals.
# See bug: https://bugs.launchpad.net/ipython/+bug/348466
# Try to prevent crashes by being more secure than it needs to
# E.g. eclipse is able to print a µ, but has no sys.stdout.encoding set.
units = [u"s", u"ms", u'us', "ns"] # the save value
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb5'.encode(sys.stdout.encoding)
units = [u"s", u"ms", u'\xb5s', "ns"]
except:
pass
scaling = [1, 1e3, 1e6, 1e9]
if timespan > 0.0:
order = min(-int(math.floor(math.log10(timespan)) // 3), 3)
else:
order = 3
return u"%.*g %s" % (precision, timespan * scaling[order], units[order])
class CvParameters:
# It may be a little slower because a dict named "self" is read for each function call.
def __init__(self, radius, step):
# self.prev_radius=radius
self._radius = radius
self.pad = 2 * radius
# self.prev_step=step
self._step = step
self._hsf = HaarSurroundFeature(radius)
def get_rpsh(self):
return self._radius, self.pad, self._step, self._hsf
# Essentially, the following would be preferable, but it would take twice as long to call.
# return self.radius, self.pad, self.step, self.hsf
@property
def radius(self):
return self._radius
@radius.setter
def radius(self, now_radius):
# self.prev_radius=self._radius
self._radius = now_radius
self.pad = 2 * now_radius
self.hsf = now_radius
@property
def step(self):
return self._step
@step.setter
def step(self, now_step):
# self.prev_step=self.step
self._step = now_step
@property
def hsf(self):
return self._hsf
@hsf.setter
def hsf(self, now_radius):
self._hsf = HaarSurroundFeature(now_radius)
class HaarSurroundFeature:
def __init__(self, r_inner, r_outer=None, val=None):
if r_outer is None:
r_outer = r_inner * 3
r_inner2 = r_inner * r_inner
count_inner = r_inner2
count_outer = r_outer * r_outer - r_inner2
if val is None:
val_inner = 1.0 / r_inner2
val_outer = -val_inner * count_inner / count_outer
else:
val_inner = val[0]
val_outer = val[1]
self.val_in = np.array(val_inner, dtype=np.float64)
self.val_out = np.array(val_outer, dtype=np.float64)
self.r_in = r_inner
self.r_out = r_outer
def get_kernel(self):
# Defined here, but not yet used?
# Create a kernel filled with the value of self.val_out
kernel = np.ones(shape=(2 * self.r_out - 1, 2 * self.r_out - 1), dtype=np.float64) * self.val_out
# Set the values of the inner area of the kernel using array slicing
start = (self.r_out - self.r_in)
end = (self.r_out + self.r_in - 1)
kernel[start:end, start:end] = self.val_in
return kernel
def to_gray(frame):
# Faster by quitting checking if the input image is already grayscale
# Perhaps it would be faster with less overhead to call cv2.cvtColor directly instead of using this function
return cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
@lru_cache(maxsize=lru_maxsize_vs)
def frameint_get_xy_step(imageshape, xysteps, pad, start_offset=None, end_offset=None):
"""
:param imageshape: (height(row),width(col)). row==y,cal==x
:param xysteps: (x,y)
:param pad: int
:param start_offset: (x,y) or None
:param end_offset: (x,y) or None
:return: xy_np:tuple(x,y)
"""
row, col = imageshape
row -= 1
col -= 1
x_step, y_step = xysteps
# This is not beautiful.
start_pad_x = start_pad_y = end_pad_x = end_pad_y = pad
if start_offset is not None:
start_pad_x += start_offset[0]
start_pad_y += start_offset[1]
if end_offset is not None:
end_pad_x += end_offset[0]
end_pad_y += end_offset[1]
y_np = np.arange(start_pad_y, row - end_pad_y, y_step)
x_np = np.arange(start_pad_x, col - end_pad_x, x_step)
xy_np = (x_np, y_np)
return xy_np
@lru_cache(maxsize=lru_maxsize_vvs)
def get_hsf_empty_array(len_syx, frameint_x, frame_int_dtype, fcshape):
# Function to reduce array allocation by providing an empty array first and recycling it with lru
inner_sum = np.empty(len_syx, dtype=frame_int_dtype)
outer_sum = np.empty(len_syx, dtype=frame_int_dtype)
p_temp = np.empty((len_syx[0], frameint_x), dtype=frame_int_dtype)
p00 = np.empty(len_syx, dtype=frame_int_dtype)
p11 = np.empty(len_syx, dtype=frame_int_dtype)
p01 = np.empty(len_syx, dtype=frame_int_dtype)
p10 = np.empty(len_syx, dtype=frame_int_dtype)
response_list = np.empty(len_syx, dtype=np.float64)
frame_conv = np.zeros(shape=fcshape[0], dtype=np.uint8)
frame_conv_stride = frame_conv[::fcshape[1], ::fcshape[2]]
return (inner_sum, outer_sum), p_temp, (p00, p11, p01, p10), response_list, (frame_conv, frame_conv_stride)
# @profile
def conv_int(frame_int, kernel, xy_step, padding, xy_steps_list):
"""
:param frame_int:
:param kernel: hsf
:param step: (x,y)
:param padding: int
:return:
"""
row, col = frame_int.shape
row -= 1
col -= 1
x_step, y_step = xy_step
# padding2 = 2 * padding
f_shape = row - 2 * padding, col - 2 * padding
r_in = kernel.r_in
len_sx, len_sy = len(xy_steps_list[0]), len(xy_steps_list[1])
inout_sum, p_temp, p_list, response_list, frameconvlist = get_hsf_empty_array((len_sy, len_sx), col + 1,
frame_int.dtype, (f_shape, y_step, x_step))
inner_sum, outer_sum = inout_sum
p00, p11, p01, p10 = p_list
frame_conv, frame_conv_stride = frameconvlist
y_rin_m = xy_steps_list[1] - r_in
x_rin_m = xy_steps_list[0] - r_in
y_rin_p = xy_steps_list[1] + r_in
x_rin_p = xy_steps_list[0] + r_in
# xx==(y,x),m==MINUS,p==PLUS, ex: mm==(y-,x-)
inarr_mm = frame_int[y_rin_m[0]:y_rin_m[-1] + 1:y_step, x_rin_m[0]:x_rin_m[-1] + 1:x_step]
inarr_mp = frame_int[y_rin_m[0]:y_rin_m[-1] + 1:y_step, x_rin_p[0]:x_rin_p[-1] + 1:x_step]
inarr_pm = frame_int[y_rin_p[0]:y_rin_p[-1] + 1:y_step, x_rin_m[0]:x_rin_m[-1] + 1:x_step]
inarr_pp = frame_int[y_rin_p[0]:y_rin_p[-1] + 1:y_step, x_rin_p[0]:x_rin_p[-1] + 1:x_step]
# == inarr_mm + inarr_pp - inarr_mp - inarr_pm
inner_sum[:, :] = inarr_mm
inner_sum += inarr_pp
inner_sum -= inarr_mp
inner_sum -= inarr_pm
# Bottleneck here, I want to make it smarter. Someone do it.
# (y,x)
# p00=max(y_ro_m,0),max(x_ro_m,0)
# p11=min(y_ro_p,ylim),min(x_ro_p,xlim)
# p01=max(y_ro_m,0),min(x_ro_p,xlim)
# p10=min(y_ro_p,ylim),max(x_ro_m,0)
y_ro_m = xy_steps_list[1] - kernel.r_out
x_ro_m = xy_steps_list[0] - kernel.r_out
y_ro_p = xy_steps_list[1] + kernel.r_out
x_ro_p = xy_steps_list[0] + kernel.r_out
# p00 calc
np.take(frame_int, y_ro_m, axis=0, mode="clip", out=p_temp)
np.take(p_temp, x_ro_m, axis=1, mode="clip", out=p00)
# p01 calc
np.take(p_temp, x_ro_p, axis=1, mode="clip", out=p01)
# p11 calc
np.take(frame_int, y_ro_p, axis=0, mode="clip", out=p_temp)
np.take(p_temp, x_ro_p, axis=1, mode="clip", out=p11)
# p10 calc
np.take(p_temp, x_ro_m, axis=1, mode="clip", out=p10)
# the point is this
# p00=np.take(np.take(frame_int, y_ro_m, axis=0, mode="clip"), x_ro_m, axis=1, mode="clip")
# p11=np.take(np.take(frame_int, y_ro_p, axis=0, mode="clip"), x_ro_p, axis=1, mode="clip")
# p01=np.take(np.take(frame_int, y_ro_m, axis=0, mode="clip"), x_ro_p, axis=1, mode="clip")
# p10=np.take(np.take(frame_int, y_ro_p, axis=0, mode="clip"), x_ro_m, axis=1, mode="clip")
outer_sum[:, :] = p00 + p11 - p01 - p10 - inner_sum
np.multiply(kernel.val_in, inner_sum, dtype=np.float64, out=response_list)
response_list += kernel.val_out * outer_sum
# min_response, max_val, min_loc, max_loc = cv2.minMaxLoc(self.response_list)
min_response, _, min_loc, _ = cv2.minMaxLoc(response_list)
center = ((xy_steps_list[0][min_loc[0]] - padding), (xy_steps_list[1][min_loc[1]] - padding))
frame_conv_stride[:, :] = response_list
# or
# frame_conv_stride[:, :] = self.response_list.astype(np.uint8)
return frame_conv, min_response, center
def ellipse_model(data, y, f):
@ -1621,8 +1206,13 @@ class EyeProcessor:
self.current_image, cv2.COLOR_BGR2GRAY
)
# print(self.settings.gui_RANSAC3D)
center_x, center_y, frame = HSF(self) #run algo
out_x, out_y = cal_osc(self, center_x, center_y) #filter and calibrate
self.output_images_and_update(frame, EyeInformation(InformationOrigin.HSF, out_x, out_y, 0, False)) #update app
self.ALGOSELECT() #run our algos in priority order set in settings
# self.ALGOSELECT() #run our algos in priority order set in settings

View File

@ -0,0 +1,555 @@
import cv2
import numpy as np
import timeit
from functools import lru_cache
import os
import sys
import functools
import math
#HSF \/
# cache param
lru_maxsize_vvs = 16
lru_maxsize_vs = 64
# CV param
default_radius = 20
auto_radius_range = (default_radius - 10, default_radius + 10) # (10,30)
blink_init_frames = 60 * 3 # 60fps*3sec,Number of blink statistical frames
# step==(x,y)
default_step = (5, 5) # bigger the steps,lower the processing time! ofc acc also takes an impact
response_list = []
"""
Attention.
If using cv2.filter2D in this code, be careful with the kernel
https://stackoverflow.com/questions/39457468/convolution-without-any-padding-opencv-python
"""
def TimeitWrapper(*args, **kwargs):
"""
This decorator @TimeitWrapper() prints the function name and execution time in seconds.
:param args:
:param kwargs:
:return:
"""
def decorator(function):
@functools.wraps(function)
def wrapper(*args, **kwargs):
start = timeit.default_timer()
results = function(*args, **kwargs)
end = timeit.default_timer()
print('{} execution time: {:.10f} s'.format(function.__name__, end - start))
return results
return wrapper
return decorator
class TimeitResult(object):
"""
from https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L55
Object returned by the timeit magic with info about the run.
Contains the following attributes :
loops: (int) number of loops done per measurement
repeat: (int) number of times the measurement has been repeated
best: (float) best execution time / number
all_runs: (list of float) execution time of each run (in s)
"""
def __init__(self, loops, repeat, best, worst, all_runs, precision):
self.loops = loops
self.repeat = repeat
self.best = best
self.worst = worst
self.all_runs = all_runs
self._precision = precision
self.timings = [dt / self.loops for dt in all_runs]
@property
def average(self):
return math.fsum(self.timings) / len(self.timings)
@property
def stdev(self):
mean = self.average
return (math.fsum([(x - mean) ** 2 for x in self.timings]) / len(self.timings)) ** 0.5
def __str__(self):
pm = '+-'
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb1'.encode(sys.stdout.encoding)
pm = u'\xb1'
except:
pass
return "min:{best} max:{worst} mean:{mean} {pm} {std} per loop (mean {pm} std. dev. of {runs} run{run_plural}, {loops:,} loop{loop_plural} each)".format(
pm=pm,
runs=self.repeat,
loops=self.loops,
loop_plural="" if self.loops == 1 else "s",
run_plural="" if self.repeat == 1 else "s",
mean=format_time(self.average, self._precision),
std=format_time(self.stdev, self._precision),
best=format_time(self.best, self._precision),
worst=format_time(self.worst, self._precision),
)
def _repr_pretty_(self, p, cycle):
unic = self.__str__()
p.text(u'<TimeitResult : ' + unic + u'>')
class FPSResult(object):
"""
base https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L55
"""
def __init__(self, loops, repeat, best, worst, all_runs, precision):
self.loops = loops
self.repeat = repeat
self.best = 1 / best
self.worst = 1 / worst
self.all_runs = all_runs
self._precision = precision
self.fps = [1 / dt for dt in all_runs]
self.unit = "fps"
@property
def average(self):
return math.fsum(self.fps) / len(self.fps)
@property
def stdev(self):
mean = self.average
return (math.fsum([(x - mean) ** 2 for x in self.fps]) / len(self.fps)) ** 0.5
def __str__(self):
pm = '+-'
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb1'.encode(sys.stdout.encoding)
pm = u'\xb1'
except:
pass
return "min:{best} max:{worst} mean:{mean} {pm} {std} per loop (mean {pm} std. dev. of {runs} run{run_plural}, {loops:,} loop{loop_plural} each)".format(
pm=pm,
runs=self.repeat,
loops=self.loops,
loop_plural="" if self.loops == 1 else "s",
run_plural="" if self.repeat == 1 else "s",
mean="%.*g%s" % (self._precision, self.average, self.unit),
std="%.*g%s" % (self._precision, self.stdev, self.unit),
best="%.*g%s" % (self._precision, self.best, self.unit),
worst="%.*g%s" % (self._precision, self.worst, self.unit),
)
def _repr_pretty_(self, p, cycle):
unic = self.__str__()
p.text(u'<FPSResult : ' + unic + u'>')
def format_time(timespan, precision=3):
"""
https://github.com/ipython/ipython/blob/339c0d510a1f3cb2158dd8c6e7f4ac89aa4c89d8/IPython/core/magics/execution.py#L1473
Formats the timespan in a human readable form
"""
if timespan >= 60.0:
# we have more than a minute, format that in a human readable form
# Idea from http://snipplr.com/view/5713/
parts = [("d", 60 * 60 * 24), ("h", 60 * 60), ("min", 60), ("s", 1)]
time = []
leftover = timespan
for suffix, length in parts:
value = int(leftover / length)
if value > 0:
leftover = leftover % length
time.append(u'%s%s' % (str(value), suffix))
if leftover < 1:
break
return " ".join(time)
# Unfortunately the unicode 'micro' symbol can cause problems in
# certain terminals.
# See bug: https://bugs.launchpad.net/ipython/+bug/348466
# Try to prevent crashes by being more secure than it needs to
# E.g. eclipse is able to print a µ, but has no sys.stdout.encoding set.
units = [u"s", u"ms", u'us', "ns"] # the save value
if hasattr(sys.stdout, 'encoding') and sys.stdout.encoding:
try:
u'\xb5'.encode(sys.stdout.encoding)
units = [u"s", u"ms", u'\xb5s', "ns"]
except:
pass
scaling = [1, 1e3, 1e6, 1e9]
if timespan > 0.0:
order = min(-int(math.floor(math.log10(timespan)) // 3), 3)
else:
order = 3
return u"%.*g %s" % (precision, timespan * scaling[order], units[order])
class CvParameters:
# It may be a little slower because a dict named "self" is read for each function call.
def __init__(self, radius, step):
# self.prev_radius=radius
self._radius = radius
self.pad = 2 * radius
# self.prev_step=step
self._step = step
self._hsf = HaarSurroundFeature(radius)
def get_rpsh(self):
return self._radius, self.pad, self._step, self._hsf
# Essentially, the following would be preferable, but it would take twice as long to call.
# return self.radius, self.pad, self.step, self.hsf
@property
def radius(self):
return self._radius
@radius.setter
def radius(self, now_radius):
# self.prev_radius=self._radius
self._radius = now_radius
self.pad = 2 * now_radius
self.hsf = now_radius
@property
def step(self):
return self._step
@step.setter
def step(self, now_step):
# self.prev_step=self.step
self._step = now_step
@property
def hsf(self):
return self._hsf
@hsf.setter
def hsf(self, now_radius):
self._hsf = HaarSurroundFeature(now_radius)
class HaarSurroundFeature:
def __init__(self, r_inner, r_outer=None, val=None):
if r_outer is None:
r_outer = r_inner * 3
r_inner2 = r_inner * r_inner
count_inner = r_inner2
count_outer = r_outer * r_outer - r_inner2
if val is None:
val_inner = 1.0 / r_inner2
val_outer = -val_inner * count_inner / count_outer
else:
val_inner = val[0]
val_outer = val[1]
self.val_in = np.array(val_inner, dtype=np.float64)
self.val_out = np.array(val_outer, dtype=np.float64)
self.r_in = r_inner
self.r_out = r_outer
def get_kernel(self):
# Defined here, but not yet used?
# Create a kernel filled with the value of self.val_out
kernel = np.ones(shape=(2 * self.r_out - 1, 2 * self.r_out - 1), dtype=np.float64) * self.val_out
# Set the values of the inner area of the kernel using array slicing
start = (self.r_out - self.r_in)
end = (self.r_out + self.r_in - 1)
kernel[start:end, start:end] = self.val_in
return kernel
def to_gray(frame):
# Faster by quitting checking if the input image is already grayscale
# Perhaps it would be faster with less overhead to call cv2.cvtColor directly instead of using this function
return cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
@lru_cache(maxsize=lru_maxsize_vs)
def frameint_get_xy_step(imageshape, xysteps, pad, start_offset=None, end_offset=None):
"""
:param imageshape: (height(row),width(col)). row==y,cal==x
:param xysteps: (x,y)
:param pad: int
:param start_offset: (x,y) or None
:param end_offset: (x,y) or None
:return: xy_np:tuple(x,y)
"""
row, col = imageshape
row -= 1
col -= 1
x_step, y_step = xysteps
# This is not beautiful.
start_pad_x = start_pad_y = end_pad_x = end_pad_y = pad
if start_offset is not None:
start_pad_x += start_offset[0]
start_pad_y += start_offset[1]
if end_offset is not None:
end_pad_x += end_offset[0]
end_pad_y += end_offset[1]
y_np = np.arange(start_pad_y, row - end_pad_y, y_step)
x_np = np.arange(start_pad_x, col - end_pad_x, x_step)
xy_np = (x_np, y_np)
return xy_np
@lru_cache(maxsize=lru_maxsize_vvs)
def get_hsf_empty_array(len_syx, frameint_x, frame_int_dtype, fcshape):
# Function to reduce array allocation by providing an empty array first and recycling it with lru
inner_sum = np.empty(len_syx, dtype=frame_int_dtype)
outer_sum = np.empty(len_syx, dtype=frame_int_dtype)
p_temp = np.empty((len_syx[0], frameint_x), dtype=frame_int_dtype)
p00 = np.empty(len_syx, dtype=frame_int_dtype)
p11 = np.empty(len_syx, dtype=frame_int_dtype)
p01 = np.empty(len_syx, dtype=frame_int_dtype)
p10 = np.empty(len_syx, dtype=frame_int_dtype)
response_list = np.empty(len_syx, dtype=np.float64)
frame_conv = np.zeros(shape=fcshape[0], dtype=np.uint8)
frame_conv_stride = frame_conv[::fcshape[1], ::fcshape[2]]
return (inner_sum, outer_sum), p_temp, (p00, p11, p01, p10), response_list, (frame_conv, frame_conv_stride)
# @profile
def conv_int(frame_int, kernel, xy_step, padding, xy_steps_list):
"""
:param frame_int:
:param kernel: hsf
:param step: (x,y)
:param padding: int
:return:
"""
row, col = frame_int.shape
row -= 1
col -= 1
x_step, y_step = xy_step
# padding2 = 2 * padding
f_shape = row - 2 * padding, col - 2 * padding
r_in = kernel.r_in
len_sx, len_sy = len(xy_steps_list[0]), len(xy_steps_list[1])
inout_sum, p_temp, p_list, response_list, frameconvlist = get_hsf_empty_array((len_sy, len_sx), col + 1,
frame_int.dtype, (f_shape, y_step, x_step))
inner_sum, outer_sum = inout_sum
p00, p11, p01, p10 = p_list
frame_conv, frame_conv_stride = frameconvlist
y_rin_m = xy_steps_list[1] - r_in
x_rin_m = xy_steps_list[0] - r_in
y_rin_p = xy_steps_list[1] + r_in
x_rin_p = xy_steps_list[0] + r_in
# xx==(y,x),m==MINUS,p==PLUS, ex: mm==(y-,x-)
inarr_mm = frame_int[y_rin_m[0]:y_rin_m[-1] + 1:y_step, x_rin_m[0]:x_rin_m[-1] + 1:x_step]
inarr_mp = frame_int[y_rin_m[0]:y_rin_m[-1] + 1:y_step, x_rin_p[0]:x_rin_p[-1] + 1:x_step]
inarr_pm = frame_int[y_rin_p[0]:y_rin_p[-1] + 1:y_step, x_rin_m[0]:x_rin_m[-1] + 1:x_step]
inarr_pp = frame_int[y_rin_p[0]:y_rin_p[-1] + 1:y_step, x_rin_p[0]:x_rin_p[-1] + 1:x_step]
# == inarr_mm + inarr_pp - inarr_mp - inarr_pm
inner_sum[:, :] = inarr_mm
inner_sum += inarr_pp
inner_sum -= inarr_mp
inner_sum -= inarr_pm
# Bottleneck here, I want to make it smarter. Someone do it.
# (y,x)
# p00=max(y_ro_m,0),max(x_ro_m,0)
# p11=min(y_ro_p,ylim),min(x_ro_p,xlim)
# p01=max(y_ro_m,0),min(x_ro_p,xlim)
# p10=min(y_ro_p,ylim),max(x_ro_m,0)
y_ro_m = xy_steps_list[1] - kernel.r_out
x_ro_m = xy_steps_list[0] - kernel.r_out
y_ro_p = xy_steps_list[1] + kernel.r_out
x_ro_p = xy_steps_list[0] + kernel.r_out
# p00 calc
np.take(frame_int, y_ro_m, axis=0, mode="clip", out=p_temp)
np.take(p_temp, x_ro_m, axis=1, mode="clip", out=p00)
# p01 calc
np.take(p_temp, x_ro_p, axis=1, mode="clip", out=p01)
# p11 calc
np.take(frame_int, y_ro_p, axis=0, mode="clip", out=p_temp)
np.take(p_temp, x_ro_p, axis=1, mode="clip", out=p11)
# p10 calc
np.take(p_temp, x_ro_m, axis=1, mode="clip", out=p10)
# the point is this
# p00=np.take(np.take(frame_int, y_ro_m, axis=0, mode="clip"), x_ro_m, axis=1, mode="clip")
# p11=np.take(np.take(frame_int, y_ro_p, axis=0, mode="clip"), x_ro_p, axis=1, mode="clip")
# p01=np.take(np.take(frame_int, y_ro_m, axis=0, mode="clip"), x_ro_p, axis=1, mode="clip")
# p10=np.take(np.take(frame_int, y_ro_p, axis=0, mode="clip"), x_ro_m, axis=1, mode="clip")
outer_sum[:, :] = p00 + p11 - p01 - p10 - inner_sum
np.multiply(kernel.val_in, inner_sum, dtype=np.float64, out=response_list)
response_list += kernel.val_out * outer_sum
# min_response, max_val, min_loc, max_loc = cv2.minMaxLoc(self.response_list)
min_response, _, min_loc, _ = cv2.minMaxLoc(response_list)
center = ((xy_steps_list[0][min_loc[0]] - padding), (xy_steps_list[1][min_loc[1]] - padding))
frame_conv_stride[:, :] = response_list
# or
# frame_conv_stride[:, :] = self.response_list.astype(np.uint8)
return frame_conv, min_response, center
def HSF(self):
frame = self.current_image_gray
if self.now_mode == self.cv_mode[1]:
prev_res_len = len(self.response_list)
# adjustment of radius
if prev_res_len == 1:
# len==1==self.response_list==[self.settings.gui_HSF_radius]
self.cvparam.radius = self.auto_radius_range[0]
elif prev_res_len == 2:
# len==2==self.response_list==[self.settings.gui_HSF_radius, self.auto_radius_range[0]]
self.cvparam.radius = self.auto_radius_range[1]
elif prev_res_len == 3:
# len==3==self.response_list==[self.settings.gui_HSF_radius,self.auto_radius_range[0],self.auto_radius_range[1]]
sort_res = sorted(self.response_list, key=lambda x: x[1])[0]
# Extract the radius with the lowest response value
if sort_res[0] == self.settings.gui_HSF_radius:
# If the default value is best, change self.now_mode to init after setting radius to the default value.
self.cvparam.radius = self.settings.gui_HSF_radius
self.now_mode = self.cv_mode[2] if not self.skip_blink_detect else self.cv_mode[3]
self.response_list = []
elif sort_res[0] == self.auto_radius_range[0]:
self.radius_cand_list = [i for i in range(self.auto_radius_range[0], self.settings.gui_HSF_radius, self.default_step[0])][1:]
# self.default_step is defined separately for xy, but radius is shared by xy, so it may be buggy
# It should be no problem to set it to anything other than self.default_step
self.cvparam.radius = self.radius_cand_list.pop()
else:
self.radius_cand_list = [i for i in range(self.settings.gui_HSF_radius, self.auto_radius_range[1], self.default_step[0])][1:]
# self.default_step is defined separately for xy, but radius is shared by xy, so it may be buggy
# It should be no problem to set it to anything other than self.default_step
self.cvparam.radius = self.radius_cand_list.pop()
else:
# Try the contents of the self.radius_cand_list in order until the self.radius_cand_list runs out
# Better make it a binary search.
if len(self.radius_cand_list) == 0:
sort_res = sorted(self.response_list, key=lambda x: x[1])[0]
self.cvparam.radius = sort_res[0]
self.now_mode = self.cv_mode[2] if not self.skip_blink_detect else self.cv_mode[3]
self.response_list = []
else:
self.cvparam.radius = self.radius_cand_list.pop()
radius, pad, step, hsf = self.cvparam.get_rpsh()
# For measuring processing time of image processing
cv_start_time = timeit.default_timer()
gray_frame = frame
# Calculate the integral image of the frame
int_start_time = timeit.default_timer()
# BORDER_CONSTANT is faster than BORDER_REPLICATE There seems to be almost no negative impact when BORDER_CONSTANT is used.
frame_pad = cv2.copyMakeBorder(gray_frame, pad, pad, pad, pad, cv2.BORDER_CONSTANT)
frame_int = cv2.integral(frame_pad)
# Convolve the feature with the integral image
conv_int_start_time = timeit.default_timer()
xy_step = frameint_get_xy_step(frame_int.shape, step, pad, start_offset=None, end_offset=None)
frame_conv, response, center_xy = conv_int(frame_int, hsf, step, pad, xy_step)
crop_start_time = timeit.default_timer()
# Define the center point and radius
center_x, center_y = center_xy
upper_x = center_x + 25 #TODO make this a setting
lower_x = center_x - 25
upper_y = center_y + 25
lower_y = center_y - 25
# Crop the image using the calculated bounds
cropped_image = gray_frame[lower_y:upper_y, lower_x:upper_x] # y is 50px, x is 45? why?
if self.now_mode == self.cv_mode[0] or self.now_mode == self.cv_mode[1]:
# If mode is first_frame or radius_adjust, record current radius and response
self.response_list.append((radius, response))
elif self.now_mode == self.cv_mode[2]:
# Statistics for blink detection
if len(self.response_list) < self.blink_init_frames:
# Record the average value of cropped_image
self.response_list.append(cv2.mean(cropped_image)[0])
else:
# Calculate self.response_max by computing interquartile range, IQR
# Change self.cv_mode to normal
self.response_list = np.array(self.response_list)
# 25%,75%
# This value may need to be adjusted depending on the environment.
quartile_1, quartile_3 = np.percentile(self.response_list, [25, 75])
iqr = quartile_3 - quartile_1
# response_min = quartile_1 - (iqr * 1.5)
self.response_max = quartile_3 + (iqr * 1.5)
self.now_mode = self.cv_mode[3]
else:
if 0 in cropped_image.shape:
# If shape contains 0, it is not detected well.
print("[WARN] HSF: Something's wrong.")
else:
# If the average value of cropped_image is greater than self.response_max
# (i.e., if the cropimage is whitish
if self.response_max is not None and cv2.mean(cropped_image)[0] > self.response_max:
# blink
cv2.circle(frame, (center_x, center_y), 10, (0, 0, 255), -1)
# If you want to update self.response_max. it may be more cost-effective to rewrite self.response_list in the following way
# https://stackoverflow.com/questions/42771110/fastest-way-to-left-cycle-a-numpy-array-like-pop-push-for-a-queue
cv2.circle(frame, (center_x, center_y), 10, (0, 0, 255), -1)
# print(center_x, center_y)
if self.now_mode != self.cv_mode[0] and self.now_mode != self.cv_mode[1]:
if cropped_image.size < 400:
pass
if self.now_mode == self.cv_mode[0]:
self.now_mode = self.cv_mode[1]
try:
self.failed = 0
return center_x, center_y, frame
except:
self.failed = self.failed + 1
return 0, 0, frame
#self.output_images_and_update(thresh, EyeInformation(InformationOrigin.FAILURE, 0, 0, 0, False))
# return
#self.output_images_and_update(larger_threshold,EyeInformation(InformationOrigin.HSF, out_x, out_y, 0, False),)
# return
#self.output_images_and_update(larger_threshold, EyeInformation(InformationOrigin.HSF, 0, 0, 0, True))