p.add_argument('--frame-offset',
help="Offset all frame numbers by this number",
type=int,
default=0)
p.set_defaults(normalize=False)
return p.parse_args()
#
# Setup
#
args = parse_arguments()
paths = build_paths(args)
paths.input = Path(args.input)
resize = tuple(map(
int, args.resize.split('x'))) if args.resize is not None else None
crop = tuple(map(int, args.crop.split('x'))) if args.crop is not None else None
trim = tuple(map(float,
args.trim.split(':'))) if args.trim is not None else None
flip = {
'v': 0,
'h': 1,
'hv': -1,
'vh': -1
}[args.flip] if args.flip is not None else None
normalize = args.normalize
crop_center = CropCenter[args.crop_center]
h_far) and a_close > a_far and h_close > h_far:
success = True
zthreshold += zthreshold_inc
window += window_inc
if not success:
raise Exception(
"Could not find a window and z-score threshold that result in expected transformations"
)
return a_close, h_close, a_far, h_far, window, zthreshold
args = parse_arguments()
source_paths = build_paths(args, directory_prefix='test')
target_paths = build_paths(args,
directory_prefix='train',
dataroot=args.target_dataroot)
if source_paths.norm_calculations.exists():
print("Loading normalization calculations from {}".format(
source_paths.norm_calculations))
calculations = np.load(source_paths.norm_calculations)
else:
s_ankles, s_noses, s_heights, s_images = calculate_metrics(
source_paths, limit=args.limit)
t_ankles, t_noses, t_heights, t_images = calculate_metrics(
target_paths, limit=args.limit)
s_calc = calculate_close_and_far_transformations(
required=True,
type=int)
p.add_argument('--frame-offset',
help='The frame offset for the two datasets',
required=True,
type=int)
p.add_argument('--target-label-offset', required=True, type=int)
p = Labeller.add_arguments(p)
return p.parse_args()
print("Synthesizing face puppet")
args = parse_arguments()
paths_base = build_paths(args, directory_prefix='base')
paths_out = build_paths(args, directory_prefix='test')
create_directories(paths_out)
labeller = Labeller.build_from_arguments(args, paths_base)
labeller_t = Labeller.build_from_arguments(
args, paths_base, label_offset=args.target_label_offset)
del labeller.face_labeller.landmarks["right_eyebrow"]
del labeller.face_labeller.landmarks["left_eyebrow"]
del labeller.face_labeller.landmarks["right_eye"]
del labeller.face_labeller.landmarks["left_eye"]
del labeller_t.face_labeller.landmarks["mouth"]
del labeller_t.face_labeller.landmarks["inner_mouth"]
del labeller_t.face_labeller.landmarks["nose"]
del labeller_t.face_labeller.landmarks["jaw"]
TAU = 6.2831853071795864769252867665590057683943
def parse_arguments():
p = argparse.ArgumentParser(description="Generate labels based on programmatic movement", fromfile_prefix_chars='@')
p.add_argument('--dataroot', type=str)
p.add_argument('-i', '--input', help='Path to a frame image that provides the basis for the generation', required=True)
p.add_argument('-n', '--nframes', help='Number of frames to generate', required=True, type=int)
p = Labeller.add_arguments(p)
return p.parse_args()
args = parse_arguments()
paths = build_paths(args, directory_prefix='test')
create_directories(paths)
labeller = Labeller.build_from_arguments(args, paths)
image_path = Path(args.input)
finput = cv.imread(str(image_path))
pose = Pose(labeller.detect_pose(finput))
larm = np.atleast_2d(pose.larm.copy())
origin = np.atleast_2d(pose.lshoulder.copy())
for i in tqdm(range(0,args.nframes)):
t = float(i) / float(args.nframes)
theta = TAU * .25 * t - TAU * .125
c, s = np.cos(theta), np.sin(theta)
