def augment_data(image, face, landmarks):
'''produce 2 type of augumentation: rotation and noising,
returns an array of 3 tuple (image, landmarks)'''
assert (type(face) is Region)
h, w = image.shape[:2]
angle = 30
pivot = face.center()
# 30 degree rotation
A = utils.rotate_image(image, angle, pivot)
B = utils.rotate_landmarks(landmarks, pivot, angle)
R1 = utils.points_region(B)
# -30 degree rotatation
C = utils.rotate_image(image, -angle, pivot)
D = utils.rotate_landmarks(landmarks, pivot, -angle)
R2 = utils.points_region(D)
# mirroring
E = utils.flip_image(image)
F = utils.naive_flip_landmarks(landmarks, w)
R3 = face.flip(width=w)
return [(A, B, R1), (C, D, R2), (E, F, R3)]
def test_augment():
utils.init_face_detector(True, 321)
utils.load_shape_predictor("dlib/shape_predictor_68_face_landmarks.dat")
for img in utils.images_inside("trainset"):
points = utils.detect_landmarks(img, region(img))
angle = 30
h, w = img.shape[:2]
center = (w / 2, h / 2)
# 30 degree rotation
rot1 = utils.rotate_image(img, angle)
rot_pts1 = utils.rotate_landmarks(points, center, angle)
# -30 degree rotatation
rot2 = utils.rotate_image(img, -angle)
rot_pts2 = utils.rotate_landmarks(points, center, -angle)
# mirroring
mir = utils.flip_image(img)
mir_pts = utils.detect_landmarks(mir, region(mir))
utils.draw_points(img, points)
utils.draw_points(rot1, rot_pts1, color=Colors.cyan)
utils.draw_points(rot2, rot_pts2, color=Colors.purple)
utils.draw_points(mir, mir_pts, color=Colors.green)
while True:
cv2.imshow("image", img)
cv2.imshow("mirrored", mir)
cv2.imshow("rotated30", rot1)
cv2.imshow("rotated-30", rot2)
key = cv2.waitKey(20) & 0Xff
if key == Keys.ESC:
break
elif key == Keys.Q:
return cv2.destroyAllWindows()
def generate(samples, batch_size):
num_samples = len(samples)
shuffle(samples)
while 1:
for offset in range(0, num_samples, batch_size):
batch_samples = samples[offset:offset + batch_size][:]
images = []
angles = []
for batch_sample in batch_samples:
#TODO decompose this stuff into functions
#load center
img_src = batch_sample[0].strip()
center_image = mpimg.imread(img_src)
center_image = utils.resize_image(center_image)
images.append(center_image)
angles.append(str(batch_sample[3]))
#add random brightness
center_bright = utils.bright_image(center_image)
images.append(center_bright)
angles.append(str(batch_sample[3]))
# flip center
if float(batch_sample[3]) != 0:
flipped_image = utils.flip_image(center_image)
flipped_angle = utils.flip_angle(float(batch_sample[3]))
images.append(flipped_image)
angles.append(flipped_angle)
####################################################
# left
img_src = batch_sample[1].strip()
left_image = mpimg.imread(img_src)
left_image = utils.resize_image(left_image)
images.append(left_image)
left_angle = float(batch_sample[3]) + .28
angles.append(str(left_angle))
left_bright = utils.bright_image(left_image)
images.append(left_bright)
angles.append(left_angle)
#flip left
if float(batch_sample[3]) != 0:
left_flipped_image = utils.flip_image(center_image)
left_flipped_angle = utils.flip_angle(
float(batch_sample[3])) + .28
images.append(left_flipped_image)
angles.append(left_flipped_angle)
################################################
#right
img_src = batch_sample[2].strip()
right_image = mpimg.imread(img_src)
right_image = utils.resize_image(right_image)
images.append(right_image)
right_angle = float(batch_sample[3]) - .28
angles.append(str(right_angle))
right_bright = utils.bright_image(right_image)
images.append(right_bright)
angles.append(right_angle)
#flip right
if float(batch_sample[3]) != 0:
right_flipped_image = utils.flip_image(center_image)
right_flipped_angle = utils.flip_angle(
float(batch_sample[3])) - .28
images.append(right_flipped_image)
angles.append(right_flipped_angle)
x_train = np.array(images)
y_train = np.array(angles)
yield shuffle(x_train, y_train)
units='dimensionless')
# loading the data flips the images vertically!
#block = io.read_block()
block = neo.Block()
seg = neo.Segment(name='segment 0', index=0)
block.segments.append(seg)
print('vlock', block)
print('seg', block.segments[0])
block.segments[0].imagesequences.append(imageSequences)
# change data orientation to be top=ventral, right=lateral
imgseq = block.segments[0].imagesequences[0]
imgseq = flip_image(imgseq, axis=-2)
imgseq = rotate_image(imgseq, rotation=-90)
block.segments[0].imagesequences[0] = imgseq
# Transform into analogsignals
block.segments[0].analogsignals = []
block = ImageSequence2AnalogSignal(block)
block.segments[0].analogsignals[0] = time_slice(
block.segments[0].analogsignals[0], args.t_start, args.t_stop)
if args.annotations is not None:
block.segments[0].analogsignals[0].annotations.\
update(parse_string2dict(args.annotations))
block.segments[0].analogsignals[0].annotations.update(
print(count + 40 * (x - 1))
count += 1
for i in range(1, 7 + 1):#7
new_image = utils.contras_image(image, random.uniform(80, 200) / 100, random.uniform(0, 50) / 10)
new_file = new_dir + "/" + str(count + 40 * (x - 1)) + ".jpg"
cv2.imwrite(new_file, new_image)
print(count + 40 * (x - 1))
count += 1
for i in range(1, 7 + 1):#7
new_image = utils.random_shadow(image)
new_file = new_dir + "/" + str(count + 40 * (x - 1)) + ".jpg"
cv2.imwrite(new_file, new_image)
print(count + 40 * (x - 1))
count += 1
for i in range(-1, 2):#3
new_image = utils.flip_image(image, i)
new_file = new_dir + "/" + str(count + 40 * (x - 1)) + ".jpg"
cv2.imwrite(new_file, new_image)
print(count + 40 * (x - 1))
count += 1
for i in range(1, 8 + 1):#8
_image = utils.foo(utils.rotate_image(image, random.uniform(0, 360)))
new_image = utils.contras_image(_image, random.uniform(80, 200) / 100, random.uniform(0, 50) / 10)
new_file = new_dir + "/" + str(count + 40 * (x - 1)) + ".jpg"
cv2.imwrite(new_file, new_image)
print(count + 40 * (x - 1))
count += 1
for i in range(1, 7 + 1): #7
_image = utils.foo(utils.rotate_image(image, random.uniform(0, 360)))
new_image = utils.random_shadow(_image)
new_file = new_dir + "/" + str(count + 40 * (x - 1)) + ".jpg"
