def __get_point_distances(self, mytank, obstacle):
d1 = get_distance(Point(obstacle[0]), mytank)
d2 = get_distance(Point(obstacle[1]), mytank)
d3 = get_distance(Point(obstacle[2]), mytank)
d4 = get_distance(Point(obstacle[3]), mytank)
return [d1, d2, d3, d4]
def __get_point_distances(self, mytank, obstacle): d1 = get_distance(Point(obstacle[0]), mytank) d2 = get_distance(Point(obstacle[1]), mytank) d3 = get_distance(Point(obstacle[2]), mytank) d4 = get_distance(Point(obstacle[3]), mytank) return [d1, d2, d3, d4]
def arrange(lines):
"""
Delete overlapping lines, connect incontinuous lines.
Inputs:
- lines: List of lines to be rearranged.
Each line is represent as a numpy array [P.x, P.y, Q.x, Q.y],
where P and Q are the endpoints of the line.
"""
flag = False
while not flag:
flag = True
for line1 in lines:
if not flag: break
for line2 in lines:
if line1 is line2: continue
p = [[gm.Point(*line[:2]),
gm.Point(*line[2:])] for line in (line1, line2)]
l = [gm.determine_linear_equation(*p[i]) for i in range(2)]
if gm.get_distance(l[1],
p[0][0]) <= delta_e and gm.get_distance(
l[1], p[0][1]) <= delta_e:
ends = p[0]
pl = [
gm.get_perpendicular_line(l[0], p[1][i])
for i in range(2)
]
out = [
not ((gm.f(pl[i], p[0][0]) > 0) ^
(gm.f(pl[i], p[0][1]) > 0)) for i in range(2)
]
if out[0] and out[1] and not ((gm.f(pl[0], p[0][0]) > 0) ^
(gm.f(pl[1], p[0][0]) > 0)):
flag = False
for i in range(2):
if flag:
break
for j in range(2):
if ends[i] - p[1][j] <= delta_v:
ends[i] = p[1][1 - j]
flag = True
break
else:
for i in range(2):
if out[i]:
ends[i] = p[1][i]
remove(lines, line1)
remove(lines, line2)
lines.append(
np.array(ends[0].to_tuple() + ends[1].to_tuple()))
flag = False
break
def draw_nose(img, keypoints):
if keypoints is None:
return img
if len(keypoints) == 17:
keypoints = get_updated_keypoints(keypoints)
center = (keypoints["nose"][0], keypoints["nose"][1])
hcenter = (keypoints["head_center"][0], keypoints["head_center"][1])
radius = int(get_distance(center, hcenter) / 2)
color = (0, 0, 255)
img = cv2.circle(img=img,
center=center,
radius=radius,
color=color,
thickness=-1,
lineType=cv2.LINE_AA)
return img
adv_acc_pos = 0
adv_fool = 0
# Post-training
print('Loading best weights')
model.load_weights(logdir + 'model_best_val_weights.h5')
test_loss = model.evaluate(x_test, x_test, batch_size=batch_size, verbose=0)
print('Test loss: {:.2f}'.format(test_loss))
# KNN test
def sample_points(data_, labels_, n_):
target_idxs_ = []
for l_ in np.unique(labels_):
t_ = np.argwhere(labels_ == l_).reshape(-1)
np.random.shuffle(t_)
target_idxs_.append(t_[:n_])
target_idxs_ = np.hstack(target_idxs_)
return data_[target_idxs_], labels_[target_idxs_]
embeddings_train = clip(encoder.predict(x_train), radius)
embeddings_test = clip(encoder.predict(x_test), radius)
knn = KNeighborsClassifier(n_neighbors=5, metric=get_distance(radius))
for n_labels in [100, 600, 1000]:
knn.fit(*sample_points(embeddings_train, y_train, n_labels))
knn_score = knn.score(embeddings_test, y_test)
print('KNN score ({} labels): {}'.format(n_labels, knn_score))