def run_demo(args):
ie = IECore()
detector_person = Detector(ie,
path_to_model_xml=args.model_od,
device=args.device,
label_class=args.person_label)
single_human_pose_estimator = HumanPoseEstimator(
ie, path_to_model_xml=args.model_hpe, device=args.device)
if args.input != '':
img = cv2.imread(args.input[0], cv2.IMREAD_COLOR)
frames_reader, delay = (VideoReader(
args.input), 1) if img is None else (ImageReader(args.input), 0)
else:
raise ValueError('--input has to be set')
for frame in frames_reader:
bboxes = detector_person.detect(frame)
human_poses = [
single_human_pose_estimator.estimate(frame, bbox)
for bbox in bboxes
]
colors = [(0, 0, 255), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0)]
for pose, bbox in zip(human_poses, bboxes):
cv2.rectangle(frame, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]), (255, 0, 0),
2)
print("{} keypoints".format(len(pose)))
for id_kpt, kpt in enumerate(pose):
print("Position: {} id:{}".format((int(kpt[0]), int(kpt[1])),
id_kpt))
cv2.circle(frame, (int(kpt[0]), int(kpt[1])), 6,
colors[id_kpt], -1)
cv2.putText(
frame,
'summary: {:.1f} FPS (estimation: {:.1f} FPS / detection: {:.1f} FPS)'
.format(
float(1 / (detector_person.infer_time +
single_human_pose_estimator.infer_time *
len(human_poses))),
float(1 / single_human_pose_estimator.infer_time),
float(1 / detector_person.infer_time)), (5, 15),
cv2.FONT_HERSHEY_COMPLEX, 0.5, (0, 0, 200))
if args.no_show:
continue
cv2.imshow('Human Pose Estimation Demo', frame)
key = cv2.waitKey(delay)
if key == 27:
return
def run_demo(args):
ie = IECore()
detector_person = Detector(ie,
path_to_model_xml=args.model_od,
device=args.device,
label_class=args.person_label)
single_human_pose_estimator = HumanPoseEstimator(
ie, path_to_model_xml=args.model_hpe, device=args.device)
cap = open_images_capture(args.input, args.loop)
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
delay = int(cap.get_type() in ('VIDEO', 'CAMERA'))
video_writer = cv2.VideoWriter()
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(frame.shape[1], frame.shape[0])):
raise RuntimeError("Can't open video writer")
frames_processed = 0
presenter = monitors.Presenter(args.utilization_monitors, 25)
while frame is not None:
bboxes = detector_person.detect(frame)
human_poses = [
single_human_pose_estimator.estimate(frame, bbox)
for bbox in bboxes
]
presenter.drawGraphs(frame)
colors = [(0, 0, 255), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0), (255, 0, 0), (0, 255, 0), (255, 0, 0),
(0, 255, 0)]
for pose, bbox in zip(human_poses, bboxes):
cv2.rectangle(frame, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]), (255, 0, 0),
2)
for id_kpt, kpt in enumerate(pose):
cv2.circle(frame, (int(kpt[0]), int(kpt[1])), 3,
colors[id_kpt], -1)
cv2.putText(
frame,
'summary: {:.1f} FPS (estimation: {:.1f} FPS / detection: {:.1f} FPS)'
.format(
float(1 / (detector_person.infer_time +
single_human_pose_estimator.infer_time *
len(human_poses))),
float(1 / single_human_pose_estimator.infer_time),
float(1 / detector_person.infer_time)), (5, 15),
cv2.FONT_HERSHEY_COMPLEX, 0.5, (0, 0, 200))
frames_processed += 1
if video_writer.isOpened() and (args.output_limit <= 0 or
frames_processed <= args.output_limit):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Human Pose Estimation Demo', frame)
key = cv2.waitKey(delay)
if key == 27:
break
presenter.handleKey(key)
frame = cap.read()
print(presenter.reportMeans())
def run_demo(args):
cap = open_images_capture(args.input, args.loop)
log.info('OpenVINO Runtime')
log.info('\tbuild: {}'.format(get_version()))
core = Core()
log.info('Reading Object Detection model {}'.format(args.model_od))
detector_person = Detector(core, args.model_od,
device=args.device,
label_class=args.person_label)
log.info('The Object Detection model {} is loaded to {}'.format(args.model_od, args.device))
log.info('Reading Human Pose Estimation model {}'.format(args.model_hpe))
single_human_pose_estimator = HumanPoseEstimator(core, args.model_hpe,
device=args.device)
log.info('The Human Pose Estimation model {} is loaded to {}'.format(args.model_hpe, args.device))
delay = int(cap.get_type() in ('VIDEO', 'CAMERA'))
video_writer = cv2.VideoWriter()
frames_processed = 0
presenter = monitors.Presenter(args.utilization_monitors, 25)
metrics = PerformanceMetrics()
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
if args.output and not video_writer.open(args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), (frame.shape[1], frame.shape[0])):
raise RuntimeError("Can't open video writer")
while frame is not None:
bboxes = detector_person.detect(frame)
human_poses = [single_human_pose_estimator.estimate(frame, bbox) for bbox in bboxes]
presenter.drawGraphs(frame)
colors = [(0, 0, 255),
(255, 0, 0), (0, 255, 0), (255, 0, 0), (0, 255, 0),
(255, 0, 0), (0, 255, 0), (255, 0, 0), (0, 255, 0),
(255, 0, 0), (0, 255, 0), (255, 0, 0), (0, 255, 0),
(255, 0, 0), (0, 255, 0), (255, 0, 0), (0, 255, 0)]
for pose, bbox in zip(human_poses, bboxes):
cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[0] + bbox[2], bbox[1] + bbox[3]), (255, 0, 0), 2)
for id_kpt, kpt in enumerate(pose):
cv2.circle(frame, (int(kpt[0]), int(kpt[1])), 3, colors[id_kpt], -1)
metrics.update(start_time, frame)
frames_processed += 1
if video_writer.isOpened() and (args.output_limit <= 0 or frames_processed <= args.output_limit):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Human Pose Estimation Demo', frame)
key = cv2.waitKey(delay)
if key == 27:
break
presenter.handleKey(key)
start_time = perf_counter()
frame = cap.read()
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)