try:

os.makedirs(path)

except OSError:

pass

st = time.time()

for x, y in enumerate(boxes):

cur_path = '%s/%f-%.5d.png' % (path, st, x)

patch = cv2.resize(frame[y[0]:y[2], y[1]:y[3]], (32, 32))

"""

Args:

sz: Size (width, height)

num_boxes: Number of boxes

size: Box size

Returns:

Numpy array of shape num_boxes x 4 where each is tl_y, tl_x, br_y, br_x

"""

out = []

for size in sizes:

sz = np.asarray(sz, dtype=np.int)

num_boxes = int(np.prod(sz / float(size) * density))

print('sz:%s num_boxes:%s size:%s' % (sz, num_boxes, size))

if num_boxes <= 0:

continue

try:

tls = np.dstack([np.random.randint(0, sz[0] - size, num_boxes),

np.random.randint(0, sz[1] - size, num_boxes)])[0]

except ValueError:

pass

else:

out.append(np.hstack([tls, tls + np.array([size, size])]))

try:

return np.vstack(out)

except ValueError:

if not out:

return np.array([], dtype=np.int).reshape(0, 4)

num_rows = 0

while np.all(frame[-(num_rows + 2), :, :] == frame[-1, :, :]):

num_rows += 1

print('NR:%d' % num_rows)

if num_rows >= 1:

print('Removing tiling')

try:

os.makedirs('remove_tiling/')

except OSError:

pass

p = 'remove_tiling/%f' % random.random()

cv2.imwrite(p + '.png', frame)

frame = frame[:-(num_rows + 1), :, :]

cv2.imwrite(p + '-clean.png', frame)

for label, frame in vidfeat.load_label_frames('/aladdin_data_cropped/person/'):

if label == 1: # Some of the people have a tiling artifact on the bottom

frame = remove_tiling(frame)

boxes = sample_boxes(frame.shape[:2])

save_boxes('/aladdin_data_cropped/boxes/%d' % label, frame, boxes)