def checkCrossLastTwo(self, ROI, ROI_W, ROI_H):
if len(self.locationHistory) >= 2 and self.timeInvisible == 0:
pt0 = self.locationHistory[-2]
pt1 = self.locationHistory[-1]
return tools.cross(ROI, ROI_W, ROI_H, pt0, pt1)
else:
return 0
def checkCross(self, ROI, ROI_W, ROI_H):
if len(self.locationHistory) >= 2:
pt0 = self.locationHistory[0]
pt1 = self.locationHistory[-1]
return tools.cross(ROI, ROI_W, ROI_H, pt0, pt1)
else:
return 0
def checkCross(self, ROI, ROI_W, ROI_H):
if len(self.locationHistory) >= 2:
pt0 = self.locationHistory[0]
pt1 = self.locationHistory[-1]
return tools.cross(ROI, ROI_W, ROI_H, pt0, pt1)
else:
return 0
def checkCrossLastTwo(self, ROI, ROI_W, ROI_H):
if len(self.locationHistory) >= 2 and self.timeInvisible == 0:
pt0 = self.locationHistory[-2]
pt1 = self.locationHistory[-1]
return tools.cross(ROI, ROI_W, ROI_H, pt0, pt1)
else:
return 0
get_mse_v(gamma_par(params, lamb), v, params, c),
get_mse_v(gamma_cross(params, lamb), v, params, c),
get_mse_v(gamma_ols(params), v, params, c)])
# Loop over simulations
for n in [250, 2500]:
for _ in range(5000):
# Simulate training data
data = simulate(n, d, pars, noise_W=np.sqrt(s2), noise_Z=np.sqrt(s2))
X, Y, A, W, Z = data['X'], data['Y'], data['A'], data['W'], data['Z']
# Compute estimators from training data
gammas = {
'ar': ar(X, Y, A, lamb=lamb),
"par": ar(X, Y, W, lamb=lamb),
"cross": cross(X, Y, W, Z, lamb=lamb),
"ols": ols(X, Y)}
# Simulate test data from intervention do(A:=v)
test_data = simulate(n, d, pars, noise_W=np.sqrt(s2), noise_Z=np.sqrt(s2), v=v)
# Append results
results.append([n, x] + [get_mse(test_data, gamma)
for gamma in gammas.values()])
# Store data in dataframe
df = pd.DataFrame(np.array(results), columns=["n", "x"] + list(gammas.keys()))
df = df.melt(["x", "n"], var_name="method")
# Encode population values as "theo" instead of 10e9 (Panda handles, but numpy didnt)
df['n'] = df['n'].replace(theo, "theo")
def run(self, as_script=True):
if self.invisible:
cv2.namedWindow("Control")
prev_gray = None
prev_points = None
while True:
ret, frame = self.cam.read()
if not ret:
break
fg_mask = self.operator.apply(frame)
fg_mask = ((fg_mask == 255) * 255).astype(np.uint8)
fg_mask = morph_openclose(fg_mask)
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
if prev_gray is not None and prev_points is not None:
p0 = np.float32([point for point in prev_points]).reshape(-1, 1, 2)
if drawing.draw_prev_points(frame, prev_points):
# p1, st, err = cv2.calcOpticalFlowPyrLK(prev_gray, frame_gray, p0, None, **lk_params)
frame_gray[fg_mask == 0] = 255
p1, st, err = cv2.calcOpticalFlowPyrLK(prev_gray, frame_gray, p0, None, **lk_params)
for p_i, p_f in zip(p0.reshape(-1, 2), p1.reshape(-1, 2)):
result = cross(ROI, ROI_W, ROI_H, p_i, p_f)
if result is 1:
self.arrivals += 1
if not self.quiet:
print("Arrival")
elif result is -1:
self.departures += 1
if not self.quiet:
print("Departure")
if self.drawTracks:
drawing.draw_line(frame, tuple(p_i), tuple(p_f))
prev_gray = frame_gray
contours, hier = drawing.draw_contours(frame, fg_mask)
areas, prev_points = drawing.draw_min_ellipse(contours, frame, MIN_AREA, MAX_AREA)
self.areas += areas
self.frame_idx += 1
if not self.invisible:
self.draw_overlays(frame, fg_mask)
cv2.imshow("Fas", frame_gray)
cv2.imshow('Tracking', frame)
cv2.imshow("Mask", fg_mask)
delay = 33
else:
delay = 1
if handle_keys(delay) == 1:
break
# Should we continue running or yield some information about the current frame
if as_script: continue
else: pass
return self.areas
def run(self, as_script=True):
if self.invisible:
cv2.namedWindow("Control")
prev_gray = None
prev_points = None
while True:
ret, frame = self.cam.read()
if not ret:
break
fg_mask = self.operator.apply(frame)
fg_mask = ((fg_mask == 255) * 255).astype(np.uint8)
fg_mask = morph_openclose(fg_mask)
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
if prev_gray is not None and prev_points is not None:
p0 = np.float32([point
for point in prev_points]).reshape(-1, 1, 2)
if drawing.draw_prev_points(frame, prev_points):
# p1, st, err = cv2.calcOpticalFlowPyrLK(prev_gray, frame_gray, p0, None, **lk_params)
frame_gray[fg_mask == 0] = 255
p1, st, err = cv2.calcOpticalFlowPyrLK(
prev_gray, frame_gray, p0, None, **lk_params)
for p_i, p_f in zip(p0.reshape(-1, 2), p1.reshape(-1, 2)):
result = cross(ROI, ROI_W, ROI_H, p_i, p_f)
if result is 1:
self.arrivals += 1
if not self.quiet:
print("Arrival")
elif result is -1:
self.departures += 1
if not self.quiet:
print("Departure")
if self.drawTracks:
drawing.draw_line(frame, tuple(p_i), tuple(p_f))
prev_gray = frame_gray
contours, hier = drawing.draw_contours(frame, fg_mask)
areas, prev_points = drawing.draw_min_ellipse(
contours, frame, MIN_AREA, MAX_AREA)
self.areas += areas
self.frame_idx += 1
if not self.invisible:
self.draw_overlays(frame, fg_mask)
cv2.imshow("Fas", frame_gray)
cv2.imshow('Tracking', frame)
cv2.imshow("Mask", fg_mask)
delay = 33
else:
delay = 1
if handle_keys(delay) == 1:
break
# Should we continue running or yield some information about the current frame
if as_script: continue
else: pass
return self.areas
def is_refl(a, b, c):
return is_near(poly[b]['pt'], poly[c]['pt']) \
or (not ld(poly[a]['pt'], poly[b]['pt'], poly[c]['pt']) < 1e-2 \
and cross(poly[a]['pt'], poly[b]['pt'], poly[c]['pt']) < 0)
# The experiment regards considers a larger variance in proxy of A1 than in A2.
noise = np.diag([1 for i in cA1] + [3 for i in cA2])
# 1) Simulate
n = 10000
out = None
# Loop repeats experiment 1000 times
for _ in tqdm(range(1000)):
# Simulate data
data = simulate(n, d, pars, noise_W=noise)
X, Y, A, W, Z = data['X'], data['Y'], data['A'], data['W'], data['Z']
# Fit estimators
par5 = ar(X, Y, W, lamb=5)
cross5 = cross(X, Y, W, Z, lamb=5)
ar5 = ar(X, Y, A, lamb=5)
# Cast to dataframe
df = pd.DataFrame(cb(par5, cross5, ar5), columns=["par5", "cross5", "ar5"])
# 'Causal' encodes for whether predictor is causal.
df['Causal'] = 3 * [1] + 3 * [0]
# 'X.coord' encodes variable number (e.g. X_1, X_2, X_3, ...)
df['X.coord'] = np.arange(1, 7)
# Melt dataframe
df = df.melt(id_vars=["Causal", "X.coord"],
var_name="Method",
value_name="Weight")
# Add results from this simulation to overall results
out = df if out is None else pd.concat((out, df))
