def main():
import os, sys
from xbob.db.replay import Database, File
from .. import utils
basedir = os.path.dirname(os.path.dirname(os.path.realpath(sys.argv[0])))
ANNOTATIONS = os.path.join(basedir, 'annotations')
FEATURES = os.path.join(basedir, 'framediff')
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('annotations', metavar='DIR', type=str,
default=ANNOTATIONS, nargs='?', help='Base directory containing the (flandmark) annotations to be treated by this procedure (defaults to "%(default)s")')
parser.add_argument('path', metavar='PATH', type=str,
help='Base path to the movie file you need plotting')
parser.add_argument('output', metavar='FILE', type=str,
help='Name of the output file to save the video')
parser.add_argument('-M', '--maximum-displacement', metavar='FLOAT',
type=float, dest="max_displacement", default=0.2, help="Maximum displacement (w.r.t. to the eye width) between eye-centers to consider the eye for calculating eye-differences")
parser.add_argument('-S', '--skip-frames', metavar='INT', type=int,
default=10, dest='skip', help="Number of frames to skip once an eye-blink has been detected (defaults to %(default)s)")
parser.add_argument('-T', '--threshold-ratio', metavar='FLOAT', type=float,
default=3.0, dest='thres_ratio', help="How many standard deviations to use for counting positive blink picks to %(default)s)")
args = parser.parse_args()
db = Database()
# Gets the information concerning the input path or id
db.assert_validity()
splitted = args.path.split(os.sep)
k = [splitted.index(k) for k in splitted if k in \
('train', 'test', 'devel', 'enroll')][0]
splitted[-1] = os.path.splitext(splitted[-1])[0]
path_query = os.sep.join(splitted[k:])
obj = db.session.query(File).filter(File.path == path_query).one()
video = bob.io.VideoReader(args.path)
print "Opened movie file %s (%d frames), id = %d" % \
(args.path, len(video), obj.id)
# Choose the printed frames here.
start = 0
end = 225
# Loads the input video
frames = [bob.ip.rgb_to_gray(k) for k in video[start:end]]
# Recalculates the features
annotations = utils.flandmark_load_annotations(obj, args.annotations,
verbose=True)
# Light-normalizes detected faces
#utils.light_normalize_tantriggs(frames, annotations, start, end)
utils.light_normalize_histogram(frames, annotations, start, end)
features = numpy.zeros((len(frames), 2), dtype='float64')
sys.stdout.write("Computing features ")
sys.stdout.flush()
for k in range(start+1, end):
curr_annot = annotations[k] if annotations.has_key(k) else None
prev_annot = annotations[k-1] if annotations.has_key(k-1) else None
use_annotation = (prev_annot, curr_annot)
use_frames = (frames[k-1], frames[k])
# maximum of 5 pixel displacement acceptable
eye_diff, eye_pixels = utils.eval_eyes_difference(use_frames,
use_annotation, max_center_displacement=args.max_displacement)
facerem_diff, facerem_pixels = utils.eval_face_remainder_difference(
use_frames, use_annotation, eye_diff, eye_pixels)
if eye_pixels != 0: features[k][0] = eye_diff/float(eye_pixels)
else: features[k][0] = 0.
if facerem_pixels != 0: features[k][1] = facerem_diff/float(facerem_pixels)
else: features[k][1] = 1.
if eye_diff == 0: sys.stdout.write('x')
else: sys.stdout.write('.')
sys.stdout.flush()
scores = utils.score(features)
sys.stdout.write('\n')
sys.stdout.flush()
# plot N sequential images containing the video on the top and the advancing
# graph of the features of choice on the bottom
fig = mpl.figure()
sys.stdout.write("Writing %d frames " % (end-start))
sys.stdout.flush()
outv = None #output video place holder
orows, ocolumns = None, None #the size of every frame in outv
old_blinks = 0
for k in range(start,end):
mpl.subplot(211)
mpl.title("Frame %05d" % k)
use_annotation = annotations[k] if annotations.has_key(k) else None
if use_annotation:
x, y, width, height = use_annotation['bbox']
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['eyes'][0]
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['eyes'][1]
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['face_remainder']
bob.ip.draw_box(frames[k], x, y, width, height, 255)
mpl.imshow(frames[k], cmap=GrayColorMap) #top plot
mpl.subplot(212)
score_set = scores[start:k+1]
blinks = utils.count_blinks(score_set, args.thres_ratio, args.skip)
rmean = utils.rmean(score_set)[-1]
rstd = utils.rstd(score_set)[-1]
threshold = (args.thres_ratio * rstd) + rmean
mpl.plot(numpy.arange(start, k+1), score_set, linewidth=2, label='score')
mpl.hlines(rmean, start, end, color='red',
linestyles='dashed', alpha=0.8, label='mean')
mpl.hlines(threshold, start, end, color='red',
linestyles='solid', alpha=0.8, label='threshold')
yrange = scores.max() - scores.min()
mpl.axis((start, end, scores.min(), (0.2*yrange) + scores.max()))
mpl.grid(True)
mpl.xlabel("Frames | Blinks = %d" % blinks[-1])
mpl.ylabel("Magnitude")
figure = fig2array(fig)
if outv is None:
orows = 2*(figure.shape[1]/2)
ocolumns = 2*(figure.shape[2]/2)
outv = bob.io.VideoWriter(args.output, orows, ocolumns, video.frame_rate)
outv.append(figure[:,0:orows,0:ocolumns])
mpl.clf()
if blinks[-1] != old_blinks:
old_blinks = blinks[-1]
sys.stdout.write('%d' % old_blinks)
else:
sys.stdout.write('.')
sys.stdout.flush()
sys.stdout.write('\n')
def main():
import os, sys
from xbob.db.replay import Database, File
from .. import utils
basedir = os.path.dirname(os.path.dirname(os.path.realpath(sys.argv[0])))
ANNOTATIONS = os.path.join(basedir, 'annotations')
FEATURES = os.path.join(basedir, 'framediff')
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('annotations', metavar='DIR', type=str,
default=ANNOTATIONS, nargs='?', help='Base directory containing the (flandmark) annotations to be treated by this procedure (defaults to "%(default)s")')
parser.add_argument('path', metavar='PATH', type=str,
help='Base path to the movie file you need plotting')
parser.add_argument('output', metavar='FILE', type=str,
help='Name of the output file to save the video')
parser.add_argument('-M', '--maximum-displacement', metavar='FLOAT',
type=float, dest="max_displacement", default=0.2, help="Maximum displacement (w.r.t. to the eye width) between eye-centers to consider the eye for calculating eye-differences")
parser.add_argument('-S', '--skip-frames', metavar='INT', type=int,
default=10, dest='skip', help="Number of frames to skip once an eye-blink has been detected (defaults to %(default)s)")
parser.add_argument('-T', '--threshold-ratio', metavar='FLOAT', type=float,
default=3.0, dest='thres_ratio', help="How many standard deviations to use for counting positive blink picks to %(default)s)")
args = parser.parse_args()
db = Database()
# Gets the information concerning the input path or id
db.assert_validity()
splitted = args.path.split(os.sep)
k = [splitted.index(k) for k in splitted if k in \
('train', 'test', 'devel')][0]
splitted[-1] = os.path.splitext(splitted[-1])[0]
path_query = os.sep.join(splitted[k:])
obj = db.session.query(File).filter(File.path == path_query).one()
video = bob.io.VideoReader(args.path)
print "Opened movie file %s (%d frames), id = %d" % \
(args.path, len(video), obj.id)
# Choose the printed frames here.
start = 0
end = 225
# Loads the input video
frames = [bob.ip.rgb_to_gray(k) for k in video[start:end]]
# Recalculates the features
annotations = utils.flandmark_load_annotations(obj, args.annotations,
verbose=True)
# Light-normalizes detected faces
#utils.light_normalize_tantriggs(frames, annotations, start, end)
utils.light_normalize_histogram(frames, annotations, start, end)
features = numpy.zeros((len(frames), 2), dtype='float64')
sys.stdout.write("Computing features ")
sys.stdout.flush()
for k in range(start+1, end):
curr_annot = annotations[k] if annotations.has_key(k) else None
prev_annot = annotations[k-1] if annotations.has_key(k-1) else None
use_annotation = (prev_annot, curr_annot)
use_frames = (frames[k-1], frames[k])
# maximum of 5 pixel displacement acceptable
eye_diff, eye_pixels = utils.eval_eyes_difference(use_frames,
use_annotation, max_center_displacement=args.max_displacement)
facerem_diff, facerem_pixels = utils.eval_face_remainder_difference(
use_frames, use_annotation, eye_diff, eye_pixels)
if eye_pixels != 0: features[k][0] = eye_diff/float(eye_pixels)
else: features[k][0] = 0.
if facerem_pixels != 0: features[k][1] = facerem_diff/float(facerem_pixels)
else: features[k][1] = 1.
if eye_diff == 0: sys.stdout.write('x')
else: sys.stdout.write('.')
sys.stdout.flush()
scores = utils.score(features)
sys.stdout.write('\n')
sys.stdout.flush()
# plot N sequential images containing the video on the top and the advancing
# graph of the features of choice on the bottom
fig = mpl.figure()
sys.stdout.write("Writing %d frames " % (end-start))
sys.stdout.flush()
outv = None #output video place holder
orows, ocolumns = None, None #the size of every frame in outv
old_blinks = 0
for k in range(start,end):
mpl.subplot(211)
mpl.title("Frame %05d" % k)
use_annotation = annotations[k] if annotations.has_key(k) else None
if use_annotation:
x, y, width, height = use_annotation['bbox']
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['eyes'][0]
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['eyes'][1]
bob.ip.draw_box(frames[k], x, y, width, height, 255)
x, y, width, height = use_annotation['face_remainder']
bob.ip.draw_box(frames[k], x, y, width, height, 255)
mpl.imshow(frames[k], cmap=GrayColorMap) #top plot
mpl.subplot(212)
score_set = scores[start:k+1]
blinks = utils.count_blinks(score_set, args.thres_ratio, args.skip)
rmean = utils.rmean(score_set)[-1]
rstd = utils.rstd(score_set)[-1]
threshold = (args.thres_ratio * rstd) + rmean
mpl.plot(numpy.arange(start, k+1), score_set, linewidth=2, label='score')
mpl.hlines(rmean, start, end, color='red',
linestyles='dashed', alpha=0.8, label='mean')
mpl.hlines(threshold, start, end, color='red',
linestyles='solid', alpha=0.8, label='threshold')
yrange = scores.max() - scores.min()
mpl.axis((start, end, scores.min(), (0.2*yrange) + scores.max()))
mpl.grid(True)
mpl.xlabel("Frames | Blinks = %d" % blinks[-1])
mpl.ylabel("Magnitude")
figure = fig2array(fig)
if outv is None:
orows = 2*(figure.shape[1]/2)
ocolumns = 2*(figure.shape[2]/2)
outv = bob.io.VideoWriter(args.output, orows, ocolumns, video.frame_rate)
outv.append(figure[:,0:orows,0:ocolumns])
mpl.clf()
if blinks[-1] != old_blinks:
old_blinks = blinks[-1]
sys.stdout.write('%d' % old_blinks)
else:
sys.stdout.write('.')
sys.stdout.flush()
sys.stdout.write('\n')
