def main():
args = build_argparser().parse_args()
log.info('Initializing Inference Engine...')
ie = IECore()
config_user_specified, config_min_latency = get_plugin_configs(
args.device, args.num_streams, args.num_threads)
labels_map = None
if args.labels:
with open(args.labels, 'r') as f:
labels_map = [x.strip() for x in f]
log.info('Loading network...')
completed_request_results = {}
modes = cycle(Modes)
prev_mode = mode = next(modes)
log.info('Using {} mode'.format(mode.name))
mode_info = {mode: ModeInfo()}
exceptions = []
detectors = {
Modes.USER_SPECIFIED:
Detector(ie,
args.model,
device=args.device,
plugin_config=config_user_specified,
results=completed_request_results,
max_num_requests=args.num_infer_requests,
labels_map=labels_map,
keep_aspect_ratio_resize=args.keep_aspect_ratio,
caught_exceptions=exceptions),
Modes.MIN_LATENCY:
Detector(ie,
args.model,
device=args.device.split(':')[-1].split(',')[0],
plugin_config=config_min_latency,
results=completed_request_results,
max_num_requests=1,
labels_map=labels_map,
keep_aspect_ratio_resize=args.keep_aspect_ratio,
caught_exceptions=exceptions)
}
try:
input_stream = int(args.input)
except ValueError:
input_stream = args.input
cap = cv2.VideoCapture(input_stream)
wait_key_time = 1
next_frame_id = 0
next_frame_id_to_show = 0
input_repeats = 0
log.info('Starting inference...')
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
print(
"To switch between min_latency/user_specified modes, press TAB key in the output window"
)
palette = ColorPalette(len(labels_map) if labels_map is not None else 100)
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 4),
round(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 8)))
while (cap.isOpened() \
or completed_request_results \
or len(detectors[mode].empty_requests) < len(detectors[mode].requests)) \
and not exceptions:
if next_frame_id_to_show in completed_request_results:
frame_meta, raw_outputs = completed_request_results.pop(
next_frame_id_to_show)
objects = detectors[mode].postprocess(raw_outputs, frame_meta)
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(objects) and args.raw_output_message:
log.info(' Class ID | Confidence | XMIN | YMIN | XMAX | YMAX ')
origin_im_size = frame.shape[:-1]
presenter.drawGraphs(frame)
for obj in objects:
if obj.score > args.prob_threshold:
xmin = max(int(obj.xmin), 0)
ymin = max(int(obj.ymin), 0)
xmax = min(int(obj.xmax), origin_im_size[1])
ymax = min(int(obj.ymax), origin_im_size[0])
class_id = int(obj.class_id)
color = palette[class_id]
det_label = labels_map[class_id] if labels_map and len(
labels_map) >= class_id else str(class_id)
if args.raw_output_message:
log.info('{:^9} | {:10f} | {:4} | {:4} | {:4} | {:4} '.
format(det_label, obj.score, xmin, ymin, xmax,
ymax))
cv2.rectangle(frame, (xmin, ymin), (xmax, ymax), color, 2)
cv2.putText(frame,
'#{} {:.1%}'.format(det_label, obj.score),
(xmin, ymin - 7), cv2.FONT_HERSHEY_COMPLEX,
0.6, color, 1)
mode_message = '{} mode'.format(mode.name)
put_highlighted_text(frame, mode_message,
(10, int(origin_im_size[0] - 20)),
cv2.FONT_HERSHEY_COMPLEX, 0.75, (10, 10, 200),
2)
next_frame_id_to_show += 1
if prev_mode == mode:
mode_info[mode].frames_count += 1
elif len(completed_request_results) == 0:
mode_info[prev_mode].last_end_time = perf_counter()
prev_mode = mode
# Frames count is always zero if mode has just been switched (i.e. prev_mode != mode).
if mode_info[mode].frames_count != 0:
fps_message = 'FPS: {:.1f}'.format(mode_info[mode].frames_count / \
(perf_counter() - mode_info[mode].last_start_time))
mode_info[mode].latency_sum += perf_counter() - start_time
latency_message = 'Latency: {:.1f} ms'.format((mode_info[mode].latency_sum / \
mode_info[mode].frames_count) * 1e3)
# Draw performance stats over frame.
put_highlighted_text(frame, fps_message, (15, 20),
cv2.FONT_HERSHEY_COMPLEX, 0.75,
(200, 10, 10), 2)
put_highlighted_text(frame, latency_message, (15, 50),
cv2.FONT_HERSHEY_COMPLEX, 0.75,
(200, 10, 10), 2)
if not args.no_show:
cv2.imshow('Detection Results', frame)
key = cv2.waitKey(wait_key_time)
ESC_KEY = 27
TAB_KEY = 9
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
# Switch mode.
# Disable mode switch if the previous switch has not been finished yet.
if key == TAB_KEY and mode_info[mode].frames_count > 0:
mode = next(modes)
detectors[prev_mode].await_all()
mode_info[prev_mode].last_end_time = perf_counter()
mode_info[mode] = ModeInfo()
log.info('Using {} mode'.format(mode.name))
else:
presenter.handleKey(key)
elif detectors[mode].empty_requests and cap.isOpened():
start_time = perf_counter()
ret, frame = cap.read()
if not ret:
if input_repeats < args.loop or args.loop < 0:
cap.open(input_stream)
input_repeats += 1
else:
cap.release()
continue
detectors[mode](frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
detectors[mode].await_any()
if exceptions:
raise exceptions[0]
for exec_net in detectors.values():
exec_net.await_all()
for mode_value, mode_stats in mode_info.items():
log.info('')
log.info('Mode: {}'.format(mode_value.name))
end_time = mode_stats.last_end_time if mode_stats.last_end_time is not None \
else perf_counter()
log.info('FPS: {:.1f}'.format(mode_stats.frames_count / \
(end_time - mode_stats.last_start_time)))
log.info('Latency: {:.1f} ms'.format((mode_stats.latency_sum / \
mode_stats.frames_count) * 1e3))
print(presenter.reportMeans())
def main():
args = build_argparser().parse_args()
if args.architecture_type != 'yolov4' and args.anchors:
log.warning(
'The "--anchors" option works only for "-at==yolov4". Option will be omitted'
)
if args.architecture_type != 'yolov4' and args.masks:
log.warning(
'The "--masks" option works only for "-at==yolov4". Option will be omitted'
)
if args.architecture_type not in ['nanodet', 'nanodet-plus'
] and args.num_classes:
log.warning(
'The "--num_classes" option works only for "-at==nanodet" and "-at==nanodet-plus". Option will be omitted'
)
cap = open_images_capture(args.input, args.loop)
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(
create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests,
model_parameters={'input_layouts': args.layout})
elif args.adapter == 'ovms':
model_adapter = OVMSAdapter(args.model)
configuration = {
'resize_type': args.resize_type,
'mean_values': args.mean_values,
'scale_values': args.scale_values,
'reverse_input_channels': args.reverse_input_channels,
'path_to_labels': args.labels,
'confidence_threshold': args.prob_threshold,
'input_size': args.input_size, # The CTPN specific
'num_classes':
args.num_classes, # The NanoDet and NanoDetPlus specific
}
model = DetectionModel.create_model(args.architecture_type, model_adapter,
configuration)
model.log_layers_info()
detector_pipeline = AsyncPipeline(model)
next_frame_id = 0
next_frame_id_to_show = 0
palette = ColorPalette(len(model.labels) if model.labels else 100)
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
presenter = None
output_transform = None
video_writer = cv2.VideoWriter()
while True:
if detector_pipeline.callback_exceptions:
raise detector_pipeline.callback_exceptions[0]
# Process all completed requests
results = detector_pipeline.get_result(next_frame_id_to_show)
if results:
objects, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(objects) and args.raw_output_message:
print_raw_results(objects, model.labels, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_detections(frame, objects, palette, model.labels,
output_transform)
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Detection Results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
continue
if detector_pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
output_transform = OutputTransform(frame.shape[:2],
args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(output_resolution[0] / 4),
round(output_resolution[1] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), output_resolution):
raise RuntimeError("Can't open video writer")
# Submit for inference
detector_pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
detector_pipeline.await_any()
detector_pipeline.await_all()
if detector_pipeline.callback_exceptions:
raise detector_pipeline.callback_exceptions[0]
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = detector_pipeline.get_result(next_frame_id_to_show)
objects, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(objects) and args.raw_output_message:
print_raw_results(objects, model.labels, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_detections(frame, objects, palette, model.labels,
output_transform)
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Detection Results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
detector_pipeline.preprocess_metrics.get_latency(),
detector_pipeline.inference_metrics.get_latency(),
detector_pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
target_bgr = open_images_capture(args.target_bgr, loop=True) if args.target_bgr else None
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams, args.num_threads)
model_adapter = OpenvinoAdapter(create_core(), args.model, device=args.device, plugin_config=plugin_config,
max_num_requests=args.num_infer_requests, model_parameters = {'input_layouts': args.layout})
elif args.adapter == 'ovms':
model_adapter = OVMSAdapter(args.model)
labels = ['__background__', 'person'] if args.labels is None else load_labels(args.labels)
assert len(labels), 'The file with class labels is empty'
configuration = {
'confidence_threshold': args.prob_threshold,
'resize_type': args.resize_type
}
model, need_bgr_input = get_model(model_adapter, configuration, args)
input_bgr = open_images_capture(args.background, False).read() if need_bgr_input else None
person_id = -1
for i, label in enumerate(labels):
if label == 'person':
person_id = i
break
assert person_id >= 0, 'Person class did not find in labels list.'
model.log_layers_info()
pipeline = AsyncPipeline(model)
next_frame_id = 0
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
presenter = None
output_transform = None
video_writer = cv2.VideoWriter()
while True:
if pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
bgr = target_bgr.read() if target_bgr is not None else None
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
output_transform = OutputTransform(frame.shape[:2], args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(args.utilization_monitors, 55,
(round(output_resolution[0] / 4), round(output_resolution[1] / 8)))
if args.output and not video_writer.open(args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), tuple(output_resolution)):
raise RuntimeError("Can't open video writer")
# Submit for inference
data = {'src': frame, 'bgr': input_bgr} if input_bgr is not None else frame
pipeline.submit_data(data, next_frame_id, {'frame': frame, 'start_time': start_time})
next_frame_id += 1
else:
# Wait for empty request
pipeline.await_any()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process all completed requests
results = pipeline.get_result(next_frame_id_to_show)
if results:
objects, frame_meta = results
if args.raw_output_message:
print_raw_results(objects, next_frame_id_to_show)
frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
frame = render_results(frame, objects, output_resolution, bgr, person_id,
args.blur_bgr, args.show_with_original_frame)
render_metrics.update(rendering_start_time)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (args.output_limit <= 0 or next_frame_id_to_show <= args.output_limit-1):
video_writer.write(frame)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Background subtraction results', frame)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
presenter.handleKey(key)
pipeline.await_all()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = pipeline.get_result(next_frame_id_to_show)
objects, frame_meta = results
if args.raw_output_message:
print_raw_results(objects, next_frame_id_to_show, model.labels)
frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
frame = render_results(frame, objects, output_resolution, bgr, person_id,
args.blur_bgr, args.show_with_original_frame)
render_metrics.update(rendering_start_time)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (args.output_limit <= 0 or next_frame_id_to_show <= args.output_limit-1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Background subtraction results', frame)
cv2.waitKey(1)
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
pipeline.preprocess_metrics.get_latency(),
pipeline.inference_metrics.get_latency(),
pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
next_frame_id = 1
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
video_writer = cv2.VideoWriter()
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(
create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests,
model_parameters={'input_layouts': args.layout})
elif args.adapter == 'ovms':
model_adapter = OVMSAdapter(args.model)
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
model = Deblurring(model_adapter, preload=False)
model.reshape(frame.shape)
model.log_layers_info()
pipeline = AsyncPipeline(model)
pipeline.submit_data(frame, 0, {'frame': frame, 'start_time': start_time})
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(frame.shape[1] / 4), round(frame.shape[0] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(2 * frame.shape[1], frame.shape[0])):
raise RuntimeError("Can't open video writer")
while True:
if pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
break
# Submit for inference
pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
pipeline.await_any()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process all completed requests
results = pipeline.get_result(next_frame_id_to_show)
if results:
result_frame, frame_meta = results
input_frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
if input_frame.shape != result_frame.shape:
input_frame = cv2.resize(
input_frame,
(result_frame.shape[1], result_frame.shape[0]))
final_image = cv2.hconcat([input_frame, result_frame])
render_metrics.update(rendering_start_time)
presenter.drawGraphs(final_image)
metrics.update(start_time, final_image)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(final_image)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Deblurring Results', final_image)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
presenter.handleKey(key)
pipeline.await_all()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = pipeline.get_result(next_frame_id_to_show)
result_frame, frame_meta = results
input_frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
if input_frame.shape != result_frame.shape:
input_frame = cv2.resize(
input_frame, (result_frame.shape[1], result_frame.shape[0]))
final_image = cv2.hconcat([input_frame, result_frame])
render_metrics.update(rendering_start_time)
presenter.drawGraphs(final_image)
metrics.update(start_time, final_image)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(final_image)
if not args.no_show:
cv2.imshow('Deblurring Results', final_image)
key = cv2.waitKey(1)
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
pipeline.preprocess_metrics.get_latency(),
pipeline.inference_metrics.get_latency(),
pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
parser = argparse.ArgumentParser(description='Whiteboard inpainting demo')
parser.add_argument('-i', '--input', required=True,
help='Required. Path to a video file or a device node of a web-camera.')
parser.add_argument('--loop', default=False, action='store_true',
help='Optional. Enable reading the input in a loop.')
parser.add_argument('-o', '--output', required=False,
help='Optional. Name of the output file(s) to save.')
parser.add_argument('-limit', '--output_limit', required=False, default=1000, type=int,
help='Optional. Number of frames to store in output. '
'If 0 is set, all frames are stored.')
parser.add_argument('-m_i', '--m_instance_segmentation', type=str, required=False,
help='Required. Path to the instance segmentation model.')
parser.add_argument('-m_s', '--m_semantic_segmentation', type=str, required=False,
help='Required. Path to the semantic segmentation model.')
parser.add_argument('-t', '--threshold', type=float, default=0.6,
help='Optional. Threshold for person instance segmentation model.')
parser.add_argument('--no_show', help="Optional. Don't show output.", action='store_true')
parser.add_argument('-d', '--device', type=str, default='CPU',
help='Optional. Specify a target device to infer on. CPU, GPU, HDDL or MYRIAD is '
'acceptable. The demo will look for a suitable plugin for the device specified.')
parser.add_argument('-l', '--cpu_extension', type=str, default=None,
help='MKLDNN (CPU)-targeted custom layers. Absolute \
path to a shared library with the kernels impl.')
parser.add_argument('-u', '--utilization_monitors', default='', type=str,
help='Optional. List of monitors to show initially.')
args = parser.parse_args()
cap = open_images_capture(args.input, args.loop)
if cap.get_type() not in ('VIDEO', 'CAMERA'):
raise RuntimeError("The input should be a video file or a numeric camera ID")
if bool(args.m_instance_segmentation) == bool(args.m_semantic_segmentation):
raise ValueError('Set up exactly one of segmentation models: '
'--m_instance_segmentation or --m_semantic_segmentation')
labels_dir = Path(__file__).resolve().parents[3] / 'data/dataset_classes'
mouse = MouseClick()
if not args.no_show:
cv2.namedWindow(WINNAME)
cv2.setMouseCallback(WINNAME, mouse.get_points)
log.info('OpenVINO Inference Engine')
log.info('\tbuild: {}'.format(get_version()))
core = Core()
model_path = args.m_instance_segmentation if args.m_instance_segmentation else args.m_semantic_segmentation
log.info('Reading model {}'.format(model_path))
if args.m_instance_segmentation:
labels_file = str(labels_dir / 'coco_80cl_bkgr.txt')
segmentation = MaskRCNN(core, args.m_instance_segmentation, labels_file,
args.threshold, args.device, args.cpu_extension)
elif args.m_semantic_segmentation:
labels_file = str(labels_dir / 'cityscapes_19cl_bkgr.txt')
segmentation = SemanticSegmentation(core, args.m_semantic_segmentation, labels_file,
args.threshold, args.device, args.cpu_extension)
log.info('The model {} is loaded to {}'.format(model_path, args.device))
metrics = PerformanceMetrics()
video_writer = cv2.VideoWriter()
black_board = False
frame_number = 0
key = -1
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
out_frame_size = (frame.shape[1], frame.shape[0] * 2)
output_frame = np.full((frame.shape[0], frame.shape[1], 3), 255, dtype='uint8')
presenter = monitors.Presenter(args.utilization_monitors, 20,
(out_frame_size[0] // 4, out_frame_size[1] // 16))
if args.output and not video_writer.open(args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), out_frame_size):
raise RuntimeError("Can't open video writer")
while frame is not None:
mask = None
detections = segmentation.get_detections([frame])
expand_mask(detections, frame.shape[1] // 27)
if len(detections[0]) > 0:
mask = detections[0][0][2]
for i in range(1, len(detections[0])):
mask = cv2.bitwise_or(mask, detections[0][i][2])
if mask is not None:
mask = np.stack([mask, mask, mask], axis=-1)
else:
mask = np.zeros(frame.shape, dtype='uint8')
clear_frame = remove_background(frame, invert_colors=not black_board)
output_frame = np.where(mask, output_frame, clear_frame)
merged_frame = np.vstack([frame, output_frame])
merged_frame = cv2.resize(merged_frame, out_frame_size)
metrics.update(start_time, merged_frame)
if video_writer.isOpened() and (args.output_limit <= 0 or frame_number <= args.output_limit-1):
video_writer.write(merged_frame)
presenter.drawGraphs(merged_frame)
if not args.no_show:
cv2.imshow(WINNAME, merged_frame)
key = check_pressed_keys(key)
if key == 27: # 'Esc'
break
if key == ord('i'): # catch pressing of key 'i'
black_board = not black_board
output_frame = 255 - output_frame
else:
presenter.handleKey(key)
if mouse.crop_available:
x0, x1 = min(mouse.points[0][0], mouse.points[1][0]), \
max(mouse.points[0][0], mouse.points[1][0])
y0, y1 = min(mouse.points[0][1], mouse.points[1][1]), \
max(mouse.points[0][1], mouse.points[1][1])
x1, y1 = min(x1, output_frame.shape[1] - 1), min(y1, output_frame.shape[0] - 1)
board = output_frame[y0: y1, x0: x1, :]
if board.shape[0] > 0 and board.shape[1] > 0:
cv2.namedWindow('Board', cv2.WINDOW_KEEPRATIO)
cv2.imshow('Board', board)
frame_number += 1
start_time = perf_counter()
frame = cap.read()
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
with open(args.labels, 'rt') as labels_file:
class_labels = labels_file.read().splitlines()
assert len(class_labels), 'The file with class labels is empty'
# Plugin initialization for specified device and load extensions library if specified.
log.info('OpenVINO Inference Engine')
log.info('\tbuild: {}'.format(get_version()))
core = Core()
if args.cpu_extension and 'CPU' in args.device:
core.add_extension(args.cpu_extension, 'CPU')
# Read IR
log.info('Reading model {}'.format(args.model))
model = core.read_model(args.model)
image_input, image_info_input, (
n, c, h,
w), model_type, output_names, postprocessor = check_model(model)
args.no_keep_aspect_ratio = model_type == 'yolact' or args.no_keep_aspect_ratio
compiled_model = core.compile_model(model, args.device)
infer_request = compiled_model.create_infer_request()
log.info('The model {} is loaded to {}'.format(args.model, args.device))
if args.no_track:
tracker = None
else:
tracker = StaticIOUTracker()
if args.delay:
delay = args.delay
else:
delay = int(cap.get_type() in ('VIDEO', 'CAMERA'))
frames_processed = 0
metrics = PerformanceMetrics()
visualizer = Visualizer(class_labels,
show_boxes=args.show_boxes,
show_scores=args.show_scores)
video_writer = cv2.VideoWriter()
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
out_frame_size = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 45,
(round(out_frame_size[0] / 4), round(out_frame_size[1] / 8)))
if args.output and not video_writer.open(args.output,
cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), out_frame_size):
raise RuntimeError("Can't open video writer")
while frame is not None:
if args.no_keep_aspect_ratio:
# Resize the image to a target size.
scale_x = w / frame.shape[1]
scale_y = h / frame.shape[0]
input_image = cv2.resize(frame, (w, h))
else:
# Resize the image to keep the same aspect ratio and to fit it to a window of a target size.
scale_x = scale_y = min(h / frame.shape[0], w / frame.shape[1])
input_image = cv2.resize(frame, None, fx=scale_x, fy=scale_y)
input_image_size = input_image.shape[:2]
input_image = np.pad(input_image,
((0, h - input_image_size[0]),
(0, w - input_image_size[1]), (0, 0)),
mode='constant',
constant_values=0)
# Change data layout from HWC to CHW.
input_image = input_image.transpose((2, 0, 1))
input_image = input_image.reshape((n, c, h, w)).astype(np.float32)
input_image_info = np.asarray(
[[input_image_size[0], input_image_size[1], 1]], dtype=np.float32)
# Run the model.
feed_dict = {image_input: input_image}
if image_info_input:
feed_dict[image_info_input] = input_image_info
infer_request.infer(feed_dict)
outputs = {
name: infer_request.get_tensor(name).data[:]
for name in output_names
}
# Parse detection results of the current request
scores, classes, boxes, masks = postprocessor(outputs, scale_x,
scale_y,
*frame.shape[:2], h, w,
args.prob_threshold)
if len(boxes) and args.raw_output_message:
log.debug(
' -------------------------- Frame # {} -------------------------- '
.format(frames_processed))
log.debug(
' Class ID | Confidence | XMIN | YMIN | XMAX | YMAX '
)
for box, cls, score in zip(boxes, classes, scores):
log.debug(
'{:>10} | {:>10f} | {:>8.2f} | {:>8.2f} | {:>8.2f} | {:>8.2f} '
.format(cls, score, *box))
# Get instance track IDs.
masks_tracks_ids = None
if tracker is not None:
masks_tracks_ids = tracker(masks, classes)
# Visualize masks.
frame = visualizer(frame, boxes, classes, scores, presenter, masks,
masks_tracks_ids)
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:
# Show resulting image.
cv2.imshow('Results', frame)
if not args.no_show:
key = cv2.waitKey(delay)
esc_code = 27
if key == esc_code:
break
presenter.handleKey(key)
start_time = perf_counter()
frame = cap.read()
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
if args.labels:
with open(args.labels) as f:
labels = [line.strip() for line in f]
else:
labels = None
ie = IECore()
if 'MYRIAD' in args.device:
myriad_config = {'VPU_HW_STAGES_OPTIMIZATION': 'YES'}
ie.set_config(myriad_config, 'MYRIAD')
if args.cpu_extension and 'CPU' in args.device:
ie.add_extension(args.cpu_extension, 'CPU')
decoder_target_device = 'CPU'
if args.device != 'CPU':
encoder_target_device = args.device
else:
encoder_target_device = decoder_target_device
encoder_xml = args.m_encoder
encoder_bin = args.m_encoder.replace('.xml', '.bin')
models = [
IEModel(encoder_xml,
encoder_bin,
ie,
encoder_target_device,
num_requests=(3 if args.device == 'MYRIAD' else 1))
]
if args.architecture_type == 'en-de':
if args.m_decoder is None:
raise RuntimeError(
'No decoder for encoder-decoder model type (-m_de) provided')
decoder_xml = args.m_decoder
decoder_bin = args.m_decoder.replace('.xml', '.bin')
models.append(
IEModel(decoder_xml,
decoder_bin,
ie,
decoder_target_device,
num_requests=2))
seq_size = models[1].input_size[1]
elif args.architecture_type == 'en-mean':
models.append(DummyDecoder(num_requests=2))
seq_size = args.decoder_seq_size
elif args.architecture_type == 'i3d-rgb':
seq_size = models[0].input_size[2]
presenter = monitors.Presenter(args.utilization_monitors, 70)
result_presenter = ResultRenderer(
no_show=args.no_show,
presenter=presenter,
output=args.output,
limit=args.output_limit,
labels=labels,
label_smoothing_window=args.label_smoothing)
cap = open_images_capture(args.input, args.loop)
run_pipeline(cap,
args.architecture_type,
models,
result_presenter.render_frame,
seq_size=seq_size,
fps=cap.fps())
print(presenter.reportMeans())
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
frame_processor = FrameProcessor(args)
frame_num = 0
metrics = PerformanceMetrics()
presenter = None
output_transform = None
input_crop = None
if args.crop_size[0] > 0 and args.crop_size[1] > 0:
input_crop = np.array(args.crop_size)
elif not (args.crop_size[0] == 0 and args.crop_size[1] == 0):
raise ValueError('Both crop height and width should be positive')
video_writer = cv2.VideoWriter()
while True:
start_time = perf_counter()
frame = cap.read()
if frame is None:
if frame_num == 0:
raise ValueError("Can't read an image from the input")
break
if input_crop:
frame = center_crop(frame, input_crop)
if frame_num == 0:
output_transform = OutputTransform(frame.shape[:2],
args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(args.utilization_monitors, 55,
(round(output_resolution[0] / 4),
round(output_resolution[1] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
output_resolution):
raise RuntimeError("Can't open video writer")
detections = frame_processor.process(frame)
presenter.drawGraphs(frame)
frame = draw_detections(frame, frame_processor, detections,
output_transform)
metrics.update(start_time, frame)
frame_num += 1
if video_writer.isOpened() and (args.output_limit <= 0
or frame_num <= args.output_limit):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Face recognition demo', frame)
key = cv2.waitKey(1)
# Quit
if key in {ord('q'), ord('Q'), 27}:
break
presenter.handleKey(key)
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)
def main():
""" Main function. """
log.basicConfig(format='[ %(levelname)s ] %(message)s',
level=log.INFO,
stream=sys.stdout)
args = build_argparser().parse_args()
class_map = load_class_map(args.class_map)
assert class_map is not None
ie_core = load_ie_core(args.device, args.cpu_extension)
person_detector = PersonDetector(args.detection_model,
args.device,
ie_core,
num_requests=2,
output_shape=DETECTOR_OUTPUT_SHAPE)
action_recognizer = ActionRecognizer(args.action_model,
args.device,
ie_core,
num_requests=2,
img_scale=ACTION_IMAGE_SCALE,
num_classes=len(class_map))
person_tracker = Tracker(person_detector, TRACKER_SCORE_THRESHOLD,
TRACKER_IOU_THRESHOLD)
video_stream = VideoStream(args.input, ACTION_NET_INPUT_FPS,
action_recognizer.input_length)
video_stream.start()
visualizer = Visualizer(VISUALIZER_TRG_FPS)
visualizer.register_window('Demo')
presenter = monitors.Presenter(args.utilization_monitors)
samples_library = None
if args.samples_dir is not None and os.path.exists(args.samples_dir):
visualizer.register_window('Gesture library')
visualizer.start()
library_queue = visualizer.get_queue('Gesture library')
samples_library = VideoLibrary(args.samples_dir,
SAMPLES_MAX_WINDOW_SIZE,
list(class_map.values()), library_queue,
SAMPLES_TRG_FPS)
samples_library.start()
else:
visualizer.start()
last_caption = None
active_object_id = -1
tracker_labels_map = dict()
tracker_labels = set()
frames_processed = 0
start_time = time.perf_counter()
while True:
frame = video_stream.get_live_frame()
batch = video_stream.get_batch()
if frame is None or batch is None:
break
if frames_processed == 0:
video_writer = cv2.VideoWriter()
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'),
video_stream.fps(), (frame.shape[1], frame.shape[0])):
raise RuntimeError("Can't open video writer")
detections, tracker_labels_map = person_tracker.add_frame(
frame, len(OBJECT_IDS), tracker_labels_map)
if detections is None:
active_object_id = -1
last_caption = None
if len(detections) == 1:
active_object_id = 0
if active_object_id >= 0:
cur_det = [det for det in detections if det.id == active_object_id]
if len(cur_det) != 1:
active_object_id = -1
last_caption = None
continue
recognizer_result = action_recognizer(batch,
cur_det[0].roi.reshape(-1))
if recognizer_result is not None:
action_class_id = np.argmax(recognizer_result)
action_class_label = \
class_map[action_class_id] if class_map is not None else action_class_id
action_class_score = np.max(recognizer_result)
if action_class_score > args.action_threshold:
last_caption = 'Last gesture: {} '.format(
action_class_label)
end_time = time.perf_counter()
elapsed_time = end_time - start_time
start_time = end_time
presenter.drawGraphs(frame)
if active_object_id >= 0:
current_fps = 1.0 / elapsed_time
cv2.putText(frame, 'FPS: {:.2f}'.format(current_fps), (10, 40),
cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 255, 0), 2)
if detections is not None:
tracker_labels = set(det.id for det in detections)
for det in detections:
roi_color = (0, 255,
0) if active_object_id == det.id else (128, 128,
128)
border_width = 2 if active_object_id == det.id else 1
person_roi = det.roi[0]
cv2.rectangle(frame, (person_roi[0], person_roi[1]),
(person_roi[2], person_roi[3]), roi_color,
border_width)
cv2.putText(frame, str(det.id),
(person_roi[0] + 10, person_roi[1] + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.8, roi_color, 2)
if last_caption is not None:
cv2.putText(frame, last_caption, (10, frame.shape[0] - 10),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
frames_processed += 1
if video_writer.isOpened() and (args.output_limit <= 0 or
frames_processed <= args.output_limit):
video_writer.write(frame)
if args.no_show:
continue
visualizer.put_queue(frame, 'Demo')
key = visualizer.get_key()
if key == 27: # esc
break
elif key == ord(' '): # space
active_object_id = -1
last_caption = None
elif key == 13: # enter
last_caption = None
elif key in OBJECT_IDS: # 0-9
local_bbox_id = int(chr(key))
if local_bbox_id in tracker_labels:
active_object_id = local_bbox_id
else:
presenter.handleKey(key)
if samples_library is not None:
if key == ord('f'): # forward
samples_library.next()
elif key == ord('b'): # backward
samples_library.prev()
if samples_library is not None:
samples_library.release()
visualizer.release()
video_stream.release()
print(presenter.reportMeans())
def main():
log.basicConfig(format="[ %(levelname)s ] %(message)s", level=log.INFO, stream=sys.stdout)
args = build_argparser().parse_args()
log.info("Creating Inference Engine...")
ie = IECore()
if args.cpu_extension and 'CPU' in args.device:
ie.add_extension(args.cpu_extension, "CPU")
# Read IR
log.info("Loading network")
net = ie.read_network(args.model, os.path.splitext(args.model)[0] + ".bin")
if "CPU" in args.device:
supported_layers = ie.query_network(net, "CPU")
not_supported_layers = [l for l in net.layers.keys() if l not in supported_layers]
if len(not_supported_layers) != 0:
log.error("Following layers are not supported by the plugin for specified device {}:\n {}".
format(args.device, ', '.join(not_supported_layers)))
log.error("Please try to specify cpu extensions library path in sample's command line parameters using -l "
"or --cpu_extension command line argument")
sys.exit(1)
img_info_input_blob = None
feed_dict = {}
for blob_name in net.input_info:
if len(net.input_info[blob_name].input_data.shape) == 4:
input_blob = blob_name
elif len(net.input_info[blob_name].input_data.shape) == 2:
img_info_input_blob = blob_name
else:
raise RuntimeError("Unsupported {}D input layer '{}'. Only 2D and 4D input layers are supported"
.format(len(net.input_info[blob_name].input_data.shape), blob_name))
output_postprocessor = get_output_postprocessor(net)
log.info("Loading IR to the plugin...")
exec_net = ie.load_network(network=net, num_requests=2, device_name=args.device)
# Read and pre-process input image
n, c, h, w = net.input_info[input_blob].input_data.shape
if img_info_input_blob:
feed_dict[img_info_input_blob] = [h, w, 1]
if args.input == 'cam':
input_stream = 0
else:
input_stream = args.input
cap = cv2.VideoCapture(input_stream)
assert cap.isOpened(), "Can't open " + str(input_stream)
if args.labels:
with open(args.labels, 'r') as f:
labels_map = [x.strip() for x in f]
else:
labels_map = None
cur_request_id = 0
next_request_id = 1
log.info("Starting inference in async mode...")
is_async_mode = True
render_time = 0
if is_async_mode:
ret, frame = cap.read()
frame_h, frame_w = frame.shape[:2]
presenter = monitors.Presenter(args.utilization_monitors, 45,
(round(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 4), round(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 8)))
print("To close the application, press 'CTRL+C' here or switch to the output window and press ESC key")
print("To switch between sync/async modes, press TAB key in the output window")
while cap.isOpened():
if is_async_mode:
ret, next_frame = cap.read()
else:
ret, frame = cap.read()
if ret:
frame_h, frame_w = frame.shape[:2]
if not ret:
break # abandons the last frame in case of async_mode
# Main sync point:
# in the truly Async mode we start the NEXT infer request, while waiting for the CURRENT to complete
# in the regular mode we start the CURRENT request and immediately wait for it's completion
inf_start = time.time()
if is_async_mode:
in_frame = cv2.resize(next_frame, (w, h))
in_frame = in_frame.transpose((2, 0, 1)) # Change data layout from HWC to CHW
in_frame = in_frame.reshape((n, c, h, w))
feed_dict[input_blob] = in_frame
exec_net.start_async(request_id=next_request_id, inputs=feed_dict)
else:
in_frame = cv2.resize(frame, (w, h))
in_frame = in_frame.transpose((2, 0, 1)) # Change data layout from HWC to CHW
in_frame = in_frame.reshape((n, c, h, w))
feed_dict[input_blob] = in_frame
exec_net.start_async(request_id=cur_request_id, inputs=feed_dict)
if exec_net.requests[cur_request_id].wait(-1) == 0:
inf_end = time.time()
det_time = inf_end - inf_start
# Parse detection results of the current request
for obj in output_postprocessor(exec_net.requests[cur_request_id].output_blobs):
# Draw only objects when probability more than specified threshold
if obj[2] > args.prob_threshold:
xmin = int(obj[3] * frame_w)
ymin = int(obj[4] * frame_h)
xmax = int(obj[5] * frame_w)
ymax = int(obj[6] * frame_h)
class_id = int(obj[1])
# Draw box and label\class_id
color = (min(class_id * 12.5, 255), min(class_id * 7, 255), min(class_id * 5, 255))
cv2.rectangle(frame, (xmin, ymin), (xmax, ymax), color, 2)
det_label = labels_map[class_id] if labels_map else str(class_id)
cv2.putText(frame, det_label + ' ' + str(round(obj[2] * 100, 1)) + ' %', (xmin, ymin - 7),
cv2.FONT_HERSHEY_COMPLEX, 0.6, color, 1)
# Draw performance stats
inf_time_message = "Inference time: N\A for async mode" if is_async_mode else \
"Inference time: {:.3f} ms".format(det_time * 1000)
render_time_message = "OpenCV rendering time: {:.3f} ms".format(render_time * 1000)
async_mode_message = "Async mode is on. Processing request {}".format(cur_request_id) if is_async_mode else \
"Async mode is off. Processing request {}".format(cur_request_id)
cv2.putText(frame, inf_time_message, (15, 15), cv2.FONT_HERSHEY_COMPLEX, 0.5, (200, 10, 10), 1)
cv2.putText(frame, render_time_message, (15, 30), cv2.FONT_HERSHEY_COMPLEX, 0.5, (10, 10, 200), 1)
cv2.putText(frame, async_mode_message, (10, int(frame_h - 20)), cv2.FONT_HERSHEY_COMPLEX, 0.5,
(10, 10, 200), 1)
presenter.drawGraphs(frame)
render_start = time.time()
if not args.no_show:
cv2.imshow("Detection Results", frame)
render_end = time.time()
render_time = render_end - render_start
if is_async_mode:
cur_request_id, next_request_id = next_request_id, cur_request_id
frame = next_frame
frame_h, frame_w = frame.shape[:2]
if not args.no_show:
key = cv2.waitKey(1)
if key == 27:
break
if (9 == key):
is_async_mode = not is_async_mode
log.info("Switched to {} mode".format("async" if is_async_mode else "sync"))
else:
presenter.handleKey(key)
print(presenter.reportMeans())
def main():
""" Main function. """
log.basicConfig(format='[ %(levelname)s ] %(message)s',
level=log.INFO,
stream=sys.stdout)
args = build_argparser().parse_args()
place_recognition = PlaceRecognition(args.model, args.device,
args.gallery_folder,
args.cpu_extension, args.gallery_size)
cap = open_images_capture(args.input, args.loop)
compute_embeddings_times = []
search_in_gallery_times = []
frames_processed = 0
presenter = monitors.Presenter(args.utilization_monitors, 0)
video_writer = cv2.VideoWriter()
while True:
frame = cap.read()
if frame is None:
if frames_processed == 0:
raise ValueError("Can't read an image from the input")
break
elapsed, probe_embedding = time_elapsed(
place_recognition.compute_embedding, frame)
compute_embeddings_times.append(elapsed)
elapsed, (sorted_indexes, distances) = time_elapsed(
place_recognition.search_in_gallery, probe_embedding)
search_in_gallery_times.append(elapsed)
image, key = visualize(
frame, [str(place_recognition.impaths[i]) for i in sorted_indexes],
distances[sorted_indexes],
place_recognition.input_size,
np.mean(compute_embeddings_times),
np.mean(search_in_gallery_times),
imshow_delay=3,
presenter=presenter,
no_show=args.no_show)
if frames_processed == 0:
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(image.shape[1], image.shape[0])):
raise RuntimeError("Can't open video writer")
frames_processed += 1
if video_writer.isOpened() and (args.output_limit <= 0 or
frames_processed <= args.output_limit):
video_writer.write(image)
if key == 27:
break
print(presenter.reportMeans())
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
# Plugin initialization for specified device and load extensions library if specified
log.info('OpenVINO Runtime')
log.info('\tbuild: {}'.format(get_version()))
core = Core()
# Read IR
log.info('Reading Proposal model {}'.format(args.model_pnet))
p_net = core.read_model(args.model_pnet)
if len(p_net.inputs) != 1:
raise RuntimeError("Pnet supports only single input topologies")
if len(p_net.outputs) != 2:
raise RuntimeError("Pnet supports two output topologies")
log.info('Reading Refine model {}'.format(args.model_rnet))
r_net = core.read_model(args.model_rnet)
if len(r_net.inputs) != 1:
raise RuntimeError("Rnet supports only single input topologies")
if len(r_net.outputs) != 2:
raise RuntimeError("Rnet supports two output topologies")
log.info('Reading Output model {}'.format(args.model_onet))
o_net = core.read_model(args.model_onet)
if len(o_net.inputs) != 1:
raise RuntimeError("Onet supports only single input topologies")
if len(o_net.outputs) != 3:
raise RuntimeError("Onet supports three output topologies")
pnet_input_tensor_name = p_net.inputs[0].get_any_name()
rnet_input_tensor_name = r_net.inputs[0].get_any_name()
onet_input_tensor_name = o_net.inputs[0].get_any_name()
for node in p_net.outputs:
if node.shape[1] == 2:
pnet_cls_name = node.get_any_name()
elif node.shape[1] == 4:
pnet_roi_name = node.get_any_name()
else:
raise RuntimeError("Unsupported output layer for Pnet")
for node in r_net.outputs:
if node.shape[1] == 2:
rnet_cls_name = node.get_any_name()
elif node.shape[1] == 4:
rnet_roi_name = node.get_any_name()
else:
raise RuntimeError("Unsupported output layer for Rnet")
for node in o_net.outputs:
if node.shape[1] == 2:
onet_cls_name = node.get_any_name()
elif node.shape[1] == 4:
onet_roi_name = node.get_any_name()
elif node.shape[1] == 10:
onet_pts_name = node.get_any_name()
else:
raise RuntimeError("Unsupported output layer for Onet")
next_frame_id = 0
metrics = PerformanceMetrics()
presenter = None
video_writer = cv2.VideoWriter()
is_loaded_before = False
while True:
start_time = perf_counter()
origin_image = cap.read()
if origin_image is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
presenter = monitors.Presenter(args.utilization_monitors, 55,
(round(origin_image.shape[1] / 4),
round(origin_image.shape[0] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(origin_image.shape[1], origin_image.shape[0])):
raise RuntimeError("Can't open video writer")
next_frame_id += 1
rgb_image = cv2.cvtColor(origin_image, cv2.COLOR_BGR2RGB)
oh, ow, _ = rgb_image.shape
scales = utils.calculate_scales(rgb_image)
# *************************************
# Pnet stage
# *************************************
pnet_res = []
for i, scale in enumerate(scales):
hs = int(oh * scale)
ws = int(ow * scale)
image = preprocess_image(rgb_image, ws, hs)
p_net.reshape({
pnet_input_tensor_name: PartialShape([1, 3, ws, hs])
}) # Change weidth and height of input blob
compiled_pnet = core.compile_model(p_net, args.device)
infer_request_pnet = compiled_pnet.create_infer_request()
if i == 0 and not is_loaded_before:
log.info("The Proposal model {} is loaded to {}".format(
args.model_pnet, args.device))
infer_request_pnet.infer(inputs={pnet_input_tensor_name: image})
p_res = {
name: infer_request_pnet.get_tensor(name).data[:]
for name in {pnet_roi_name, pnet_cls_name}
}
pnet_res.append(p_res)
image_num = len(scales)
rectangles = []
for i in range(image_num):
roi = pnet_res[i][pnet_roi_name]
cls = pnet_res[i][pnet_cls_name]
_, _, out_h, out_w = cls.shape
out_side = max(out_h, out_w)
rectangle = utils.detect_face_12net(cls[0][1], roi[0], out_side,
1 / scales[i], ow, oh,
score_threshold[0],
iou_threshold[0])
rectangles.extend(rectangle)
rectangles = utils.NMS(rectangles, iou_threshold[1], 'iou')
# Rnet stage
if len(rectangles) > 0:
r_net.reshape({
rnet_input_tensor_name:
PartialShape([len(rectangles), 3, 24, 24])
}) # Change batch size of input blob
compiled_rnet = core.compile_model(r_net, args.device)
infer_request_rnet = compiled_rnet.create_infer_request()
if not is_loaded_before:
log.info("The Refine model {} is loaded to {}".format(
args.model_rnet, args.device))
rnet_input = []
for rectangle in rectangles:
crop_img = rgb_image[int(rectangle[1]):int(rectangle[3]),
int(rectangle[0]):int(rectangle[2])]
crop_img = preprocess_image(crop_img, 24, 24)
rnet_input.extend(crop_img)
infer_request_rnet.infer(
inputs={rnet_input_tensor_name: rnet_input})
rnet_res = {
name: infer_request_rnet.get_tensor(name).data[:]
for name in {rnet_roi_name, rnet_cls_name}
}
roi = rnet_res[rnet_roi_name]
cls = rnet_res[rnet_cls_name]
rectangles = utils.filter_face_24net(cls, roi, rectangles, ow, oh,
score_threshold[1],
iou_threshold[2])
# Onet stage
if len(rectangles) > 0:
o_net.reshape({
onet_input_tensor_name:
PartialShape([len(rectangles), 3, 48, 48])
}) # Change batch size of input blob
compiled_onet = core.compile_model(o_net, args.device)
infer_request_onet = compiled_onet.create_infer_request()
if not is_loaded_before:
log.info("The Output model {} is loaded to {}".format(
args.model_onet, args.device))
is_loaded_before = True
onet_input = []
for rectangle in rectangles:
crop_img = rgb_image[int(rectangle[1]):int(rectangle[3]),
int(rectangle[0]):int(rectangle[2])]
crop_img = preprocess_image(crop_img, 48, 48)
onet_input.extend(crop_img)
infer_request_onet.infer(
inputs={onet_input_tensor_name: onet_input})
onet_res = {
name: infer_request_onet.get_tensor(name).data[:]
for name in {onet_roi_name, onet_cls_name, onet_pts_name}
}
roi = onet_res[onet_roi_name]
cls = onet_res[onet_cls_name]
pts = onet_res[onet_pts_name]
rectangles = utils.filter_face_48net(cls, roi, pts, rectangles, ow,
oh, score_threshold[2],
iou_threshold[3])
# display results
for rectangle in rectangles:
# Draw detected boxes
cv2.putText(origin_image,
'confidence: {:.2f}'.format(rectangle[4]),
(int(rectangle[0]), int(rectangle[1])),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0))
cv2.rectangle(origin_image, (int(rectangle[0]), int(rectangle[1])),
(int(rectangle[2]), int(rectangle[3])), (255, 0, 0),
1)
# Draw landmarks
for i in range(5, 15, 2):
cv2.circle(origin_image,
(int(rectangle[i + 0]), int(rectangle[i + 1])), 2,
(0, 255, 0))
metrics.update(start_time, origin_image)
if video_writer.isOpened() and (args.output_limit <= 0
or next_frame_id <= args.output_limit):
video_writer.write(origin_image)
if not args.no_show:
cv2.imshow('MTCNN Results', origin_image)
key = cv2.waitKey(1)
if key in {ord('q'), ord('Q'), 27}:
break
presenter.handleKey(key)
metrics.log_total()
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(
create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests)
elif args.adapter == 'remote':
log.info('Reading model {}'.format(args.model))
serving_config = {"address": "localhost", "port": 9000}
model_adapter = RemoteAdapter(args.model, serving_config)
model = SegmentationModel.create_model(args.architecture_type,
model_adapter,
{'path_to_labels': args.labels})
if args.architecture_type == 'segmentation':
visualizer = SegmentationVisualizer(args.colors)
if args.architecture_type == 'salient_object_detection':
visualizer = SaliencyMapVisualizer()
model.log_layers_info()
pipeline = AsyncPipeline(model)
next_frame_id = 0
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
presenter = None
output_transform = None
video_writer = cv2.VideoWriter()
only_masks = args.only_masks
while True:
if pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
output_transform = OutputTransform(frame.shape[:2],
args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(output_resolution[0] / 4),
round(output_resolution[1] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), output_resolution):
raise RuntimeError("Can't open video writer")
# Submit for inference
pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
pipeline.await_any()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process all completed requests
results = pipeline.get_result(next_frame_id_to_show)
if results:
objects, frame_meta = results
if args.raw_output_message:
print_raw_results(objects, next_frame_id_to_show, model.labels)
frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
frame = render_segmentation(frame, objects, visualizer,
output_transform, only_masks)
render_metrics.update(rendering_start_time)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Segmentation Results', frame)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
if key == 9:
only_masks = not only_masks
presenter.handleKey(key)
pipeline.await_all()
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = pipeline.get_result(next_frame_id_to_show)
while results is None:
results = pipeline.get_result(next_frame_id_to_show)
objects, frame_meta = results
if args.raw_output_message:
print_raw_results(objects, next_frame_id_to_show, model.labels)
frame = frame_meta['frame']
start_time = frame_meta['start_time']
rendering_start_time = perf_counter()
frame = render_segmentation(frame, objects, visualizer,
output_transform, only_masks)
render_metrics.update(rendering_start_time)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Segmentation Results', frame)
key = cv2.waitKey(1)
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
pipeline.preprocess_metrics.get_latency(),
pipeline.inference_metrics.get_latency(),
pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(
create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests)
elif args.adapter == 'remote':
log.info('Reading model {}'.format(args.model))
serving_config = {"address": "localhost", "port": 9000}
model_adapter = RemoteAdapter(args.model, serving_config)
model = MonoDepthModel(model_adapter)
model.log_layers_info()
pipeline = AsyncPipeline(model)
next_frame_id = 0
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
presenter = None
output_transform = None
video_writer = cv2.VideoWriter()
while True:
if pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
output_transform = OutputTransform(frame.shape[:2],
args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(output_resolution[0] / 4),
round(output_resolution[1] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), output_resolution):
raise RuntimeError("Can't open video writer")
# Submit for inference
pipeline.submit_data(frame, next_frame_id,
{'start_time': start_time})
next_frame_id += 1
else:
# Wait for empty request
pipeline.await_any()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process all completed requests
results = pipeline.get_result(next_frame_id_to_show)
if results:
depth_map, frame_meta = results
depth_map = apply_color_map(depth_map, output_transform)
start_time = frame_meta['start_time']
presenter.drawGraphs(depth_map)
metrics.update(start_time, depth_map)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(depth_map)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow(DEMO_NAME, depth_map)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
presenter.handleKey(key)
pipeline.await_all()
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = pipeline.get_result(next_frame_id_to_show)
while results is None:
results = pipeline.get_result(next_frame_id_to_show)
depth_map, frame_meta = results
depth_map = apply_color_map(depth_map, output_transform)
start_time = frame_meta['start_time']
presenter.drawGraphs(depth_map)
metrics.update(start_time, depth_map)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(depth_map)
if not args.no_show:
cv2.imshow(DEMO_NAME, depth_map)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
presenter.handleKey(key)
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
delay = int(cap.get_type() in {'VIDEO', 'CAMERA'})
if args.adapter == 'openvino':
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(
create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests)
elif args.adapter == 'ovms':
model_adapter = OVMSAdapter(args.model)
config = {
'mean_values': args.mean_values,
'scale_values': args.scale_values,
'reverse_input_channels': args.reverse_input_channels,
'topk': args.topk,
'path_to_labels': args.labels
}
model = Classification(model_adapter, config)
model.log_layers_info()
async_pipeline = AsyncPipeline(model)
next_frame_id = 0
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
presenter = None
output_transform = None
video_writer = cv2.VideoWriter()
ESC_KEY = 27
key = -1
while True:
if async_pipeline.callback_exceptions:
raise async_pipeline.callback_exceptions[0]
# Process all completed requests
results = async_pipeline.get_result(next_frame_id_to_show)
if results:
classifications, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if args.raw_output_message:
print_raw_results(classifications, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_labels(frame, classifications, output_transform)
if delay or args.no_show:
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Classification Results', frame)
key = cv2.waitKey(delay)
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
continue
if async_pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
output_transform = OutputTransform(frame.shape[:2],
args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(output_resolution[0] / 4),
round(output_resolution[1] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), output_resolution):
raise RuntimeError("Can't open video writer")
# Submit for inference
async_pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
async_pipeline.await_any()
async_pipeline.await_all()
if key not in {ord('q'), ord('Q'), ESC_KEY}:
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show,
next_frame_id):
results = async_pipeline.get_result(next_frame_id_to_show)
while results is None:
results = async_pipeline.get_result(next_frame_id_to_show)
classifications, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if args.raw_output_message:
print_raw_results(classifications, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_labels(frame, classifications, output_transform)
if delay or args.no_show:
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Classification Results', frame)
key = cv2.waitKey(delay)
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
if delay or args.no_show:
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
async_pipeline.preprocess_metrics.get_latency(),
async_pipeline.inference_metrics.get_latency(),
async_pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
""" Main function. """
log.basicConfig(format='[ %(levelname)s ] %(message)s',
level=log.INFO,
stream=sys.stdout)
args = build_argparser().parse_args()
img_retrieval = ImageRetrieval(args.model, args.device, args.gallery,
INPUT_SIZE, args.cpu_extension)
frames = RoiDetectorOnVideo(args.i)
compute_embeddings_times = []
search_in_gallery_times = []
positions = []
presenter = monitors.Presenter(args.utilization_monitors, 0)
for image, view_frame in frames:
position = None
sorted_indexes = []
if image is not None:
image = central_crop(image, divide_by=5, shift=1)
elapsed, probe_embedding = time_elapsed(
img_retrieval.compute_embedding, image)
compute_embeddings_times.append(elapsed)
elapsed, (sorted_indexes, distances) = time_elapsed(
img_retrieval.search_in_gallery, probe_embedding)
search_in_gallery_times.append(elapsed)
sorted_classes = [
img_retrieval.gallery_classes[i] for i in sorted_indexes
]
if args.ground_truth is not None:
position = sorted_classes.index(
img_retrieval.text_label_to_class_id[args.ground_truth])
positions.append(position)
log.info("ROI detected, found: %d, position of target: %d",
sorted_classes[0], position)
else:
log.info("ROI detected, found: %s", sorted_classes[0])
key = visualize(
view_frame,
position, [img_retrieval.impaths[i] for i in sorted_indexes],
distances[sorted_indexes] if position is not None else None,
img_retrieval.input_size,
np.mean(compute_embeddings_times),
np.mean(search_in_gallery_times),
imshow_delay=3,
presenter=presenter,
no_show=args.no_show)
if key == 27:
break
print(presenter.reportMeans())
if positions:
compute_metrics(positions)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
next_frame_id = 1
next_frame_id_to_show = 0
metrics = PerformanceMetrics()
render_metrics = PerformanceMetrics()
video_writer = cv2.VideoWriter()
plugin_config = get_user_config(args.device, args.num_streams,
args.num_threads)
model_adapter = OpenvinoAdapter(create_core(),
args.model,
device=args.device,
plugin_config=plugin_config,
max_num_requests=args.num_infer_requests)
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
config = {
'target_size': args.tsize,
'aspect_ratio': frame.shape[1] / frame.shape[0],
'prob_threshold': args.prob_threshold,
'padding_mode': 'center' if args.architecture_type == 'higherhrnet'
else None, # the 'higherhrnet' and 'ae' specific
'delta':
0.5 if 'higherhrnet' else None, # the 'higherhrnet' and 'ae' specific
}
model = ImageModel.create_model(ARCHITECTURES[args.architecture_type],
model_adapter, config)
model.log_layers_info()
hpe_pipeline = AsyncPipeline(model)
hpe_pipeline.submit_data(frame, 0, {
'frame': frame,
'start_time': start_time
})
output_transform = OutputTransform(frame.shape[:2], args.output_resolution)
if args.output_resolution:
output_resolution = output_transform.new_resolution
else:
output_resolution = (frame.shape[1], frame.shape[0])
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(output_resolution[0] / 4), round(output_resolution[1] / 8)))
if args.output and not video_writer.open(args.output,
cv2.VideoWriter_fourcc(*'MJPG'),
cap.fps(), output_resolution):
raise RuntimeError("Can't open video writer")
while True:
if hpe_pipeline.callback_exceptions:
raise hpe_pipeline.callback_exceptions[0]
# Process all completed requests
results = hpe_pipeline.get_result(next_frame_id_to_show)
if results:
(poses, scores), frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(poses) and args.raw_output_message:
print_raw_results(poses, scores, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_poses(frame, poses, args.prob_threshold,
output_transform)
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
next_frame_id_to_show += 1
if not args.no_show:
cv2.imshow('Pose estimation results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
continue
if hpe_pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
break
# Submit for inference
hpe_pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
hpe_pipeline.await_any()
hpe_pipeline.await_all()
# Process completed requests
for next_frame_id_to_show in range(next_frame_id_to_show, next_frame_id):
results = hpe_pipeline.get_result(next_frame_id_to_show)
while results is None:
results = hpe_pipeline.get_result(next_frame_id_to_show)
(poses, scores), frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(poses) and args.raw_output_message:
print_raw_results(poses, scores, next_frame_id_to_show)
presenter.drawGraphs(frame)
rendering_start_time = perf_counter()
frame = draw_poses(frame, poses, args.prob_threshold, output_transform)
render_metrics.update(rendering_start_time)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Pose estimation results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
metrics.log_total()
log_latency_per_stage(cap.reader_metrics.get_latency(),
hpe_pipeline.preprocess_metrics.get_latency(),
hpe_pipeline.inference_metrics.get_latency(),
hpe_pipeline.postprocess_metrics.get_latency(),
render_metrics.get_latency())
for rep in presenter.reportMeans():
log.info(rep)
def main():
parser = argparse.ArgumentParser(description='Whiteboard inpainting demo')
parser.add_argument('-i', type=str, help='Input sources (index of camera \
or path to a video file)', required=True)
parser.add_argument('-m_i', '--m_instance_segmentation', type=str, required=False,
help='Path to the instance segmentation model')
parser.add_argument('-m_s', '--m_semantic_segmentation', type=str, required=False,
help='Path to the semantic segmentation model')
parser.add_argument('-t', '--threshold', type=float, default=0.6,
help='Threshold for person instance segmentation model')
parser.add_argument('--output_video', type=str, default='', required=False,
help='Optional. Path to output video')
parser.add_argument("--no_show", help="Optional. Don't show output", action='store_true')
parser.add_argument('-d', '--device', type=str, default='CPU',
help='Optional. Specify a target device to infer on. CPU, GPU, FPGA, HDDL or MYRIAD is '
'acceptable. The demo will look for a suitable plugin for the device specified')
parser.add_argument('-l', '--cpu_extension', type=str, default=None,
help='MKLDNN (CPU)-targeted custom layers.Absolute \
path to a shared library with the kernels impl.')
parser.add_argument('-u', '--utilization_monitors', default='', type=str,
help='Optional. List of monitors to show initially')
args = parser.parse_args()
capture = VideoCapture(args.i)
if bool(args.m_instance_segmentation) == bool(args.m_semantic_segmentation):
raise ValueError('Set up exactly one of segmentation models: '\
'--m_instance_segmentation or --m_semantic_segmentation')
frame_size, fps = capture.get_source_parameters()
out_frame_size = (int(frame_size[0]), int(frame_size[1] * 2))
presenter = monitors.Presenter(args.utilization_monitors, 20,
(out_frame_size[0] // 4, out_frame_size[1] // 16))
root_dir = osp.dirname(osp.abspath(__file__))
mouse = MouseClick()
if not args.no_show:
cv2.namedWindow(WINNAME)
cv2.setMouseCallback(WINNAME, mouse.get_points)
if args.output_video:
fourcc = cv2.VideoWriter_fourcc(*'XVID')
output_video = cv2.VideoWriter(args.output_video, fourcc, fps, out_frame_size)
else:
output_video = None
log.info("Initializing Inference Engine")
ie = IECore()
if args.m_instance_segmentation:
labels_file = osp.join(root_dir, 'coco_labels.txt')
segmentation = MaskRCNN(ie, args.m_instance_segmentation, labels_file,
args.threshold, args.device, args.cpu_extension)
elif args.m_semantic_segmentation:
labels_file = osp.join(root_dir, 'cityscapes_labels.txt')
segmentation = SemanticSegmentation(ie, args.m_semantic_segmentation, labels_file,
args.threshold, args.device, args.cpu_extension)
black_board = False
output_frame = np.full((frame_size[1], frame_size[0], 3), 255, dtype='uint8')
frame_number = 0
key = -1
while True:
start = time.time()
_, frame = capture.get_frame()
mask = None
if frame is not None:
detections = segmentation.get_detections([frame])
expand_mask(detections, frame_size[0] // 27)
if len(detections[0]) > 0:
mask = detections[0][0][2]
for i in range(1, len(detections[0])):
mask = cv2.bitwise_or(mask, detections[0][i][2])
else:
break
if mask is not None:
mask = np.stack([mask, mask, mask], axis=-1)
else:
mask = np.zeros(frame.shape, dtype='uint8')
clear_frame = remove_background(frame, invert_colors=not black_board)
output_frame = np.where(mask, output_frame, clear_frame)
merged_frame = np.vstack([frame, output_frame])
merged_frame = cv2.resize(merged_frame, out_frame_size)
if output_video is not None:
output_video.write(merged_frame)
presenter.drawGraphs(merged_frame)
if not args.no_show:
cv2.imshow(WINNAME, merged_frame)
key = check_pressed_keys(key)
if key == 27: # 'Esc'
break
if key == ord('i'): # catch pressing of key 'i'
black_board = not black_board
output_frame = 255 - output_frame
else:
presenter.handleKey(key)
if mouse.crop_available:
x0, x1 = min(mouse.points[0][0], mouse.points[1][0]), \
max(mouse.points[0][0], mouse.points[1][0])
y0, y1 = min(mouse.points[0][1], mouse.points[1][1]), \
max(mouse.points[0][1], mouse.points[1][1])
x1, y1 = min(x1, output_frame.shape[1] - 1), min(y1, output_frame.shape[0] - 1)
board = output_frame[y0: y1, x0: x1, :]
if board.shape[0] > 0 and board.shape[1] > 0:
cv2.namedWindow('Board', cv2.WINDOW_KEEPRATIO)
cv2.imshow('Board', board)
end = time.time()
print('\rProcessing frame: {}, fps = {:.3}' \
.format(frame_number, 1. / (end - start)), end="")
frame_number += 1
print('')
log.info(presenter.reportMeans())
if output_video is not None:
output_video.release()
def main():
log.basicConfig(format='[ %(levelname)s ] %(message)s', level=log.INFO, stream=sys.stdout)
args = build_argparser().parse_args()
# Plugin initialization for specified device and load extensions library if specified.
log.info('Creating Inference Engine...')
ie = IECore()
if args.cpu_extension and 'CPU' in args.device:
ie.add_extension(args.cpu_extension, 'CPU')
# Read IR
log.info('Loading network')
net = ie.read_network(args.model, os.path.splitext(args.model)[0] + '.bin')
image_input, image_info_input, (n, c, h, w), postprocessor = check_model(net)
log.info('Loading IR to the plugin...')
exec_net = ie.load_network(network=net, device_name=args.device, num_requests=2)
try:
input_source = int(args.input_source)
except ValueError:
input_source = args.input_source
cap = cv2.VideoCapture(input_source)
if not cap.isOpened():
log.error('Failed to open "{}"'.format(args.input_source))
cap.set(cv2.CAP_PROP_BUFFERSIZE, 1)
if args.no_track:
tracker = None
else:
tracker = StaticIOUTracker()
with open(args.labels, 'rt') as labels_file:
class_labels = labels_file.read().splitlines()
presenter = monitors.Presenter(args.utilization_monitors, 45,
(round(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 4), round(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 8)))
visualizer = Visualizer(class_labels, show_boxes=args.show_boxes, show_scores=args.show_scores)
render_time = 0
log.info('Starting inference...')
print("To close the application, press 'CTRL+C' here or switch to the output window and press ESC key")
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
if args.no_keep_aspect_ratio:
# Resize the image to a target size.
scale_x = w / frame.shape[1]
scale_y = h / frame.shape[0]
input_image = cv2.resize(frame, (w, h))
else:
# Resize the image to keep the same aspect ratio and to fit it to a window of a target size.
scale_x = scale_y = min(h / frame.shape[0], w / frame.shape[1])
input_image = cv2.resize(frame, None, fx=scale_x, fy=scale_y)
input_image_size = input_image.shape[:2]
input_image = np.pad(input_image, ((0, h - input_image_size[0]),
(0, w - input_image_size[1]),
(0, 0)),
mode='constant', constant_values=0)
# Change data layout from HWC to CHW.
input_image = input_image.transpose((2, 0, 1))
input_image = input_image.reshape((n, c, h, w)).astype(np.float32)
input_image_info = np.asarray([[input_image_size[0], input_image_size[1], 1]], dtype=np.float32)
# Run the net.
inf_start = time.time()
feed_dict = {image_input: input_image}
if image_info_input:
feed_dict[image_info_input] = input_image_info
outputs = exec_net.infer(feed_dict)
inf_end = time.time()
det_time = inf_end - inf_start
# Parse detection results of the current request
scores, classes, boxes, masks = postprocessor(
outputs, scale_x, scale_y, *frame.shape[:2], h, w, args.prob_threshold)
render_start = time.time()
if len(boxes) and args.raw_output_message:
log.info('Detected boxes:')
log.info(' Class ID | Confidence | XMIN | YMIN | XMAX | YMAX ')
for box, cls, score, mask in zip(boxes, classes, scores, masks):
log.info('{:>10} | {:>10f} | {:>8.2f} | {:>8.2f} | {:>8.2f} | {:>8.2f} '.format(cls, score, *box))
# Get instance track IDs.
masks_tracks_ids = None
if tracker is not None:
masks_tracks_ids = tracker(masks, classes)
# Visualize masks.
frame = visualizer(frame, boxes, classes, scores, presenter, masks, masks_tracks_ids)
# Draw performance stats.
inf_time_message = 'Inference time: {:.3f} ms'.format(det_time * 1000)
render_time_message = 'OpenCV rendering time: {:.3f} ms'.format(render_time * 1000)
cv2.putText(frame, inf_time_message, (15, 15), cv2.FONT_HERSHEY_COMPLEX, 0.5, (200, 10, 10), 1)
cv2.putText(frame, render_time_message, (15, 30), cv2.FONT_HERSHEY_COMPLEX, 0.5, (10, 10, 200), 1)
# Print performance counters.
if args.perf_counts:
perf_counts = exec_net.requests[0].get_perf_counts()
log.info('Performance counters:')
print('{:<70} {:<15} {:<15} {:<15} {:<10}'.format('name', 'layer_type', 'exet_type', 'status',
'real_time, us'))
for layer, stats in perf_counts.items():
print('{:<70} {:<15} {:<15} {:<15} {:<10}'.format(layer, stats['layer_type'], stats['exec_type'],
stats['status'], stats['real_time']))
if not args.no_show:
# Show resulting image.
cv2.imshow('Results', frame)
render_end = time.time()
render_time = render_end - render_start
if not args.no_show:
key = cv2.waitKey(args.delay)
esc_code = 27
if key == esc_code:
break
presenter.handleKey(key)
print(presenter.reportMeans())
cv2.destroyAllWindows()
cap.release()
def main():
args = build_argparser().parse_args()
# ------------- 1. Plugin initialization for specified device and load extensions library if specified -------------
log.info("Creating Inference Engine...")
ie = IECore()
config_user_specified = {}
config_min_latency = {}
devices_nstreams = {}
if args.num_streams:
devices_nstreams = {device: args.num_streams for device in ['CPU', 'GPU'] if device in args.device} \
if args.num_streams.isdigit() \
else dict([device.split(':') for device in args.num_streams.split(',')])
if 'CPU' in args.device:
if args.cpu_extension:
ie.add_extension(args.cpu_extension, 'CPU')
if args.number_threads is not None:
config_user_specified['CPU_THREADS_NUM'] = str(args.number_threads)
if 'CPU' in devices_nstreams:
config_user_specified['CPU_THROUGHPUT_STREAMS'] = devices_nstreams['CPU'] \
if int(devices_nstreams['CPU']) > 0 \
else 'CPU_THROUGHPUT_AUTO'
config_min_latency['CPU_THROUGHPUT_STREAMS'] = '1'
if 'GPU' in args.device:
if 'GPU' in devices_nstreams:
config_user_specified['GPU_THROUGHPUT_STREAMS'] = devices_nstreams['GPU'] \
if int(devices_nstreams['GPU']) > 0 \
else 'GPU_THROUGHPUT_AUTO'
config_min_latency['GPU_THROUGHPUT_STREAMS'] = '1'
# -------------------- 2. Reading the IR generated by the Model Optimizer (.xml and .bin files) --------------------
log.info("Loading network")
net = ie.read_network(args.model, os.path.splitext(args.model)[0] + ".bin")
# ---------------------------------- 3. Load CPU extension for support specific layer ------------------------------
if "CPU" in args.device:
supported_layers = ie.query_network(net, "CPU")
not_supported_layers = [
l for l in net.layers.keys() if l not in supported_layers
]
if len(not_supported_layers) != 0:
log.error(
"Following layers are not supported by the plugin for specified device {}:\n {}"
.format(args.device, ', '.join(not_supported_layers)))
log.error(
"Please try to specify cpu extensions library path in sample's command line parameters using -l "
"or --cpu_extension command line argument")
sys.exit(1)
assert len(
net.input_info
) == 1, "Sample supports only YOLO V3 based single input topologies"
# ---------------------------------------------- 4. Preparing inputs -----------------------------------------------
log.info("Preparing inputs")
input_blob = next(iter(net.input_info))
# Read and pre-process input images
if net.input_info[input_blob].input_data.shape[1] == 3:
input_height, input_width = net.input_info[
input_blob].input_data.shape[2:]
nchw_shape = True
else:
input_height, input_width = net.input_info[
input_blob].input_data.shape[1:3]
nchw_shape = False
if args.labels:
with open(args.labels, 'r') as f:
labels_map = [x.strip() for x in f]
else:
labels_map = None
input_stream = 0 if args.input == "cam" else args.input
mode = Mode(Modes.USER_SPECIFIED)
cap = cv2.VideoCapture(input_stream)
wait_key_time = 1
# ----------------------------------------- 5. Loading model to the plugin -----------------------------------------
log.info("Loading model to the plugin")
exec_nets = {}
exec_nets[Modes.USER_SPECIFIED] = ie.load_network(
network=net,
device_name=args.device,
config=config_user_specified,
num_requests=args.num_infer_requests)
exec_nets[Modes.MIN_LATENCY] = ie.load_network(
network=net,
device_name=args.device.split(":")[-1].split(",")[0],
config=config_min_latency,
num_requests=1)
empty_requests = deque(exec_nets[mode.current].requests)
completed_request_results = {}
next_frame_id = 0
next_frame_id_to_show = 0
mode_info = {mode.current: ModeInfo()}
event = threading.Event()
callback_exceptions = []
# ----------------------------------------------- 6. Doing inference -----------------------------------------------
log.info("Starting inference...")
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
print(
"To switch between min_latency/user_specified modes, press TAB key in the output window"
)
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(cap.get(cv2.CAP_PROP_FRAME_WIDTH) / 4),
round(cap.get(cv2.CAP_PROP_FRAME_HEIGHT) / 8)))
while (cap.isOpened() \
or completed_request_results \
or len(empty_requests) < len(exec_nets[mode.current].requests)) \
and not callback_exceptions:
if next_frame_id_to_show in completed_request_results:
frame, output, start_time, is_same_mode = completed_request_results.pop(
next_frame_id_to_show)
next_frame_id_to_show += 1
if is_same_mode:
mode_info[mode.current].frames_count += 1
objects = get_objects(output, net, (input_height, input_width),
frame.shape[:-1], args.prob_threshold,
args.keep_aspect_ratio)
objects = filter_objects(objects, args.iou_threshold,
args.prob_threshold)
if len(objects) and args.raw_output_message:
log.info(
" Class ID | Confidence | XMIN | YMIN | XMAX | YMAX | COLOR "
)
origin_im_size = frame.shape[:-1]
presenter.drawGraphs(frame)
for obj in objects:
# Validation bbox of detected object
obj['xmax'] = min(obj['xmax'], origin_im_size[1])
obj['ymax'] = min(obj['ymax'], origin_im_size[0])
obj['xmin'] = max(obj['xmin'], 0)
obj['ymin'] = max(obj['ymin'], 0)
color = (min(obj['class_id'] * 12.5,
255), min(obj['class_id'] * 7,
255), min(obj['class_id'] * 5, 255))
det_label = labels_map[obj['class_id']] if labels_map and len(labels_map) >= obj['class_id'] else \
str(obj['class_id'])
if args.raw_output_message:
log.info(
"{:^9} | {:10f} | {:4} | {:4} | {:4} | {:4} | {} ".
format(det_label, obj['confidence'], obj['xmin'],
obj['ymin'], obj['xmax'], obj['ymax'], color))
cv2.rectangle(frame, (obj['xmin'], obj['ymin']),
(obj['xmax'], obj['ymax']), color, 2)
cv2.putText(
frame, "#" + det_label + ' ' +
str(round(obj['confidence'] * 100, 1)) + ' %',
(obj['xmin'], obj['ymin'] - 7), cv2.FONT_HERSHEY_COMPLEX,
0.6, color, 1)
# Draw performance stats over frame
if mode_info[mode.current].frames_count != 0:
fps_message = "FPS: {:.1f}".format(mode_info[mode.current].frames_count / \
(perf_counter() - mode_info[mode.current].last_start_time))
mode_info[
mode.current].latency_sum += perf_counter() - start_time
latency_message = "Latency: {:.1f} ms".format((mode_info[mode.current].latency_sum / \
mode_info[mode.current].frames_count) * 1e3)
put_highlighted_text(frame, fps_message, (15, 20),
cv2.FONT_HERSHEY_COMPLEX, 0.75,
(200, 10, 10), 2)
put_highlighted_text(frame, latency_message, (15, 50),
cv2.FONT_HERSHEY_COMPLEX, 0.75,
(200, 10, 10), 2)
mode_message = "{} mode".format(mode.current.name)
put_highlighted_text(frame, mode_message,
(10, int(origin_im_size[0] - 20)),
cv2.FONT_HERSHEY_COMPLEX, 0.75, (10, 10, 200),
2)
if not args.no_show:
cv2.imshow("Detection Results", frame)
#print(frame.shape)
cv2.imwrite("prob_threshold-0.2.png", frame)
key = cv2.waitKey(wait_key_time)
if key in {ord("q"), ord("Q"), 27}: # ESC key
break
if key == 9: # Tab key
prev_mode = mode.current
mode.next()
await_requests_completion(exec_nets[prev_mode].requests)
empty_requests.clear()
empty_requests.extend(exec_nets[mode.current].requests)
mode_info[prev_mode].last_end_time = perf_counter()
mode_info[mode.current] = ModeInfo()
else:
presenter.handleKey(key)
elif empty_requests and cap.isOpened():
start_time = perf_counter()
ret, frame = cap.read()
if not ret:
if args.loop_input:
cap.open(input_stream)
else:
cap.release()
continue
request = empty_requests.popleft()
# resize input_frame to network size
in_frame = preprocess_frame(frame, input_height, input_width,
nchw_shape, args.keep_aspect_ratio)
# Start inference
request.set_completion_callback(
py_callback=async_callback,
py_data=(request, next_frame_id, mode.current, frame,
start_time, completed_request_results, empty_requests,
mode, event, callback_exceptions))
request.async_infer(inputs={input_blob: in_frame})
next_frame_id += 1
else:
event.wait()
if callback_exceptions:
raise callback_exceptions[0]
for mode_value in mode_info.keys():
log.info("")
log.info("Mode: {}".format(mode_value.name))
end_time = mode_info[mode_value].last_end_time if mode_value in mode_info \
and mode_info[mode_value].last_end_time is not None \
else perf_counter()
log.info("FPS: {:.1f}".format(mode_info[mode_value].frames_count / \
(end_time - mode_info[mode_value].last_start_time)))
log.info("Latency: {:.1f} ms".format((mode_info[mode_value].latency_sum / \
mode_info[mode_value].frames_count) * 1e3))
print(presenter.reportMeans())
for exec_net in exec_nets.values():
await_requests_completion(exec_net.requests)
def main():
args = build_argparser().parse_args()
cap = open_images_capture(args.input, args.loop)
place_recognition = PlaceRecognition(args.model, args.device,
args.gallery_folder,
args.gallery_size)
compute_embeddings_times = []
search_in_gallery_times = []
frames_processed = 0
presenter = monitors.Presenter(args.utilization_monitors, 0)
video_writer = cv2.VideoWriter()
metrics = PerformanceMetrics()
while True:
start_time = perf_counter()
frame = cap.read()
if frame is None:
if frames_processed == 0:
raise ValueError("Can't read an image from the input")
break
elapsed, probe_embedding = time_elapsed(
place_recognition.compute_embedding, frame)
compute_embeddings_times.append(elapsed)
elapsed, (sorted_indexes, distances) = time_elapsed(
place_recognition.search_in_gallery, probe_embedding)
search_in_gallery_times.append(elapsed)
image, key = visualize(
frame, [str(place_recognition.impaths[i]) for i in sorted_indexes],
distances[sorted_indexes],
place_recognition.input_size,
np.mean(compute_embeddings_times),
np.mean(search_in_gallery_times),
imshow_delay=3,
presenter=presenter,
no_show=args.no_show)
metrics.update(start_time)
if frames_processed == 0:
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(image.shape[1], image.shape[0])):
raise RuntimeError("Can't open video writer")
frames_processed += 1
if video_writer.isOpened() and (args.output_limit <= 0 or
frames_processed <= args.output_limit):
video_writer.write(image)
if key == 27:
break
metrics.log_total()
for rep in presenter.reportMeans():
log.info(rep)
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")
base_height = args.height_size
fx = args.fx
frames_processed = 0
delay = 1
esc_code = 27
p_code = 112
space_code = 32
mean_time = 0
presenter = monitors.Presenter(args.utilization_monitors, 0)
while frame is not None:
current_time = cv2.getTickCount()
input_scale = base_height / frame.shape[0]
scaled_img = cv2.resize(frame,
dsize=None,
fx=input_scale,
fy=input_scale)
if fx < 0: # Focal length is unknown
fx = np.float32(0.8 * frame.shape[1])
inference_result = inference_engine.infer(scaled_img)
poses_3d, poses_2d = parse_poses(inference_result, input_scale, stride,
fx, is_video)
edges = []
def main():
metrics = PerformanceMetrics()
args = build_argparser().parse_args()
# Plugin initialization for specified device and load extensions library if specified
log.info("Creating Inference Engine")
ie = IECore()
# Read IR
log.info("Loading network files:\n\t{}".format(args.model_pnet))
p_net = ie.read_network(args.model_pnet)
assert len(p_net.input_info.keys()
) == 1, "Pnet supports only single input topologies"
assert len(p_net.outputs) == 2, "Pnet supports two output topologies"
log.info("Loading network files:\n\t{}".format(args.model_rnet))
r_net = ie.read_network(args.model_rnet)
assert len(r_net.input_info.keys()
) == 1, "Rnet supports only single input topologies"
assert len(r_net.outputs) == 2, "Rnet supports two output topologies"
log.info("Loading network files:\n\t{}".format(args.model_onet))
o_net = ie.read_network(args.model_onet)
assert len(o_net.input_info.keys()
) == 1, "Onet supports only single input topologies"
assert len(o_net.outputs) == 3, "Onet supports three output topologies"
log.info("Preparing input blobs")
pnet_input_blob = next(iter(p_net.input_info))
rnet_input_blob = next(iter(r_net.input_info))
onet_input_blob = next(iter(o_net.input_info))
log.info("Preparing output blobs")
for name, blob in p_net.outputs.items():
if blob.shape[1] == 2:
pnet_cls_name = name
elif blob.shape[1] == 4:
pnet_roi_name = name
else:
raise RuntimeError("Unsupported output layer for Pnet")
for name, blob in r_net.outputs.items():
if blob.shape[1] == 2:
rnet_cls_name = name
elif blob.shape[1] == 4:
rnet_roi_name = name
else:
raise RuntimeError("Unsupported output layer for Rnet")
for name, blob in o_net.outputs.items():
if blob.shape[1] == 2:
onet_cls_name = name
elif blob.shape[1] == 4:
onet_roi_name = name
elif blob.shape[1] == 10:
onet_pts_name = name
else:
raise RuntimeError("Unsupported output layer for Onet")
cap = open_images_capture(args.input, args.loop)
next_frame_id = 0
log.info('Starting inference...')
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
presenter = None
video_writer = cv2.VideoWriter()
while True:
start_time = perf_counter()
origin_image = cap.read()
if origin_image is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
presenter = monitors.Presenter(args.utilization_monitors, 55,
(round(origin_image.shape[1] / 4),
round(origin_image.shape[0] / 8)))
if args.output and not video_writer.open(
args.output, cv2.VideoWriter_fourcc(*'MJPG'), cap.fps(),
(origin_image.shape[1], origin_image.shape[0])):
raise RuntimeError("Can't open video writer")
next_frame_id += 1
rgb_image = cv2.cvtColor(origin_image, cv2.COLOR_BGR2RGB)
oh, ow, _ = rgb_image.shape
scales = utils.calculate_scales(rgb_image)
# *************************************
# Pnet stage
# *************************************
log.info("Loading Pnet model to the plugin")
t0 = cv2.getTickCount()
pnet_res = []
for scale in scales:
hs = int(oh * scale)
ws = int(ow * scale)
image = preprocess_image(rgb_image, ws, hs)
p_net.reshape({pnet_input_blob:
[1, 3, ws,
hs]}) # Change weidth and height of input blob
exec_pnet = ie.load_network(network=p_net, device_name=args.device)
p_res = exec_pnet.infer(inputs={pnet_input_blob: image})
pnet_res.append(p_res)
image_num = len(scales)
rectangles = []
for i in range(image_num):
roi = pnet_res[i][pnet_roi_name]
cls = pnet_res[i][pnet_cls_name]
_, _, out_h, out_w = cls.shape
out_side = max(out_h, out_w)
rectangle = utils.detect_face_12net(cls[0][1], roi[0], out_side,
1 / scales[i], ow, oh,
score_threshold[0],
iou_threshold[0])
rectangles.extend(rectangle)
rectangles = utils.NMS(rectangles, iou_threshold[1], 'iou')
# Rnet stage
if len(rectangles) > 0:
log.info("Loading Rnet model to the plugin")
r_net.reshape({rnet_input_blob:
[len(rectangles), 3, 24,
24]}) # Change batch size of input blob
exec_rnet = ie.load_network(network=r_net, device_name=args.device)
rnet_input = []
for rectangle in rectangles:
crop_img = rgb_image[int(rectangle[1]):int(rectangle[3]),
int(rectangle[0]):int(rectangle[2])]
crop_img = preprocess_image(crop_img, 24, 24)
rnet_input.extend(crop_img)
rnet_res = exec_rnet.infer(inputs={rnet_input_blob: rnet_input})
roi = rnet_res[rnet_roi_name]
cls = rnet_res[rnet_cls_name]
rectangles = utils.filter_face_24net(cls, roi, rectangles, ow, oh,
score_threshold[1],
iou_threshold[2])
# Onet stage
if len(rectangles) > 0:
log.info("Loading Onet model to the plugin")
o_net.reshape({onet_input_blob:
[len(rectangles), 3, 48,
48]}) # Change batch size of input blob
exec_onet = ie.load_network(network=o_net, device_name=args.device)
onet_input = []
for rectangle in rectangles:
crop_img = rgb_image[int(rectangle[1]):int(rectangle[3]),
int(rectangle[0]):int(rectangle[2])]
crop_img = preprocess_image(crop_img, 48, 48)
onet_input.extend(crop_img)
onet_res = exec_onet.infer(inputs={onet_input_blob: onet_input})
roi = onet_res[onet_roi_name]
cls = onet_res[onet_cls_name]
pts = onet_res[onet_pts_name]
rectangles = utils.filter_face_48net(cls, roi, pts, rectangles, ow,
oh, score_threshold[2],
iou_threshold[3])
# display results
for rectangle in rectangles:
# Draw detected boxes
cv2.putText(origin_image,
'confidence: {:.2f}'.format(rectangle[4]),
(int(rectangle[0]), int(rectangle[1])),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0))
cv2.rectangle(origin_image, (int(rectangle[0]), int(rectangle[1])),
(int(rectangle[2]), int(rectangle[3])), (255, 0, 0),
1)
# Draw landmarks
for i in range(5, 15, 2):
cv2.circle(origin_image,
(int(rectangle[i + 0]), int(rectangle[i + 1])), 2,
(0, 255, 0))
infer_time = (cv2.getTickCount() -
t0) / cv2.getTickFrequency() # Record infer time
cv2.putText(origin_image,
'summary: {:.1f} FPS'.format(1.0 / infer_time), (5, 15),
cv2.FONT_HERSHEY_COMPLEX, 0.5, (0, 0, 200))
if video_writer.isOpened() and (args.output_limit <= 0 or next_frame_id
<= args.output_limit - 1):
video_writer.write(origin_image)
if not args.no_show:
cv2.imshow('MTCNN Results', origin_image)
key = cv2.waitKey(1)
if key in {ord('q'), ord('Q'), 27}:
break
presenter.handleKey(key)
metrics.update(start_time, origin_image)
metrics.print_total()
def main():
args = build_argparser().parse_args()
metrics = PerformanceMetrics()
log.info('Initializing Inference Engine...')
ie = IECore()
plugin_config = get_plugin_configs(args.device, args.num_streams,
args.num_threads)
cap = open_images_capture(args.input, args.loop)
start_time = perf_counter()
frame = cap.read()
if frame is None:
raise RuntimeError("Can't read an image from the input")
log.info('Loading network...')
model = get_model(ie, args, frame.shape[1] / frame.shape[0])
hpe_pipeline = AsyncPipeline(ie,
model,
plugin_config,
device=args.device,
max_num_requests=args.num_infer_requests)
log.info('Starting inference...')
hpe_pipeline.submit_data(frame, 0, {
'frame': frame,
'start_time': start_time
})
next_frame_id = 1
next_frame_id_to_show = 0
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(frame.shape[1] / 4), round(frame.shape[0] / 8)))
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")
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
while True:
if hpe_pipeline.callback_exceptions:
raise hpe_pipeline.callback_exceptions[0]
# Process all completed requests
results = hpe_pipeline.get_result(next_frame_id_to_show)
if results:
(poses, scores), frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(poses) and args.raw_output_message:
print_raw_results(poses, scores)
presenter.drawGraphs(frame)
frame = draw_poses(frame, poses, args.prob_threshold)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Pose estimation results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
next_frame_id_to_show += 1
continue
if hpe_pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
break
# Submit for inference
hpe_pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
hpe_pipeline.await_any()
hpe_pipeline.await_all()
# Process completed requests
while hpe_pipeline.has_completed_request():
results = hpe_pipeline.get_result(next_frame_id_to_show)
if results:
(poses, scores), frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
if len(poses) and args.raw_output_message:
print_raw_results(poses, scores)
presenter.drawGraphs(frame)
frame = draw_poses(frame, poses, args.prob_threshold)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Pose estimation results', frame)
key = cv2.waitKey(1)
ESC_KEY = 27
# Quit.
if key in {ord('q'), ord('Q'), ESC_KEY}:
break
presenter.handleKey(key)
next_frame_id_to_show += 1
else:
break
metrics.print_total()
print(presenter.reportMeans())
input_source = int(args.input)
except ValueError:
input_source = args.input
cap = cv.VideoCapture(input_source)
if not cap.isOpened():
assert "{} not exist".format(input_source)
color_coeff = np.load(coeffs).astype(np.float32)
assert color_coeff.shape == (
313,
2), "Current shape of color coefficients does not match required shape"
imshowSize = (640, 480)
graphSize = (imshowSize[0] // 2, imshowSize[1] // 4)
presenter = monitors.Presenter(args.utilization_monitors,
imshowSize[1] * 2 - graphSize[1], graphSize)
while True:
log.debug("#############################")
hasFrame, original_frame = cap.read()
if not hasFrame:
break
(h_orig, w_orig) = original_frame.shape[:2]
log.debug("Preprocessing frame")
if original_frame.shape[2] > 1:
frame = cv.cvtColor(cv.cvtColor(original_frame, cv.COLOR_BGR2GRAY),
cv.COLOR_GRAY2RGB)
else:
frame = cv.cvtColor(original_frame, cv.COLOR_GRAY2RGB)
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 main():
args = build_argparser().parse_args()
if args.labels:
with open(args.labels) as f:
labels = [line.strip() for line in f]
else:
labels = None
log.info('OpenVINO Runtime')
log.info('\tbuild: {}'.format(get_version()))
core = Core()
if 'MYRIAD' in args.device:
myriad_config = {'MYRIAD_ENABLE_HW_ACCELERATION': 'YES'}
core.set_property('MYRIAD', myriad_config)
decoder_target_device = 'CPU'
if args.device != 'CPU':
encoder_target_device = args.device
else:
encoder_target_device = decoder_target_device
models = [
IEModel(args.m_encoder,
core,
encoder_target_device,
model_type='Action Recognition Encoder',
num_requests=(3 if args.device == 'MYRIAD' else 1))
]
if args.architecture_type == 'en-de':
if args.m_decoder is None:
raise RuntimeError(
'No decoder for encoder-decoder model type (-m_de) provided')
models.append(
IEModel(args.m_decoder,
core,
decoder_target_device,
model_type='Action Recognition Decoder',
num_requests=2))
seq_size = models[1].input_shape[1]
elif args.architecture_type == 'en-mean':
models.append(DummyDecoder(num_requests=2))
seq_size = args.decoder_seq_size
elif args.architecture_type == 'i3d-rgb':
seq_size = models[0].input_shape[1]
presenter = monitors.Presenter(args.utilization_monitors, 70)
result_presenter = ResultRenderer(
no_show=args.no_show,
presenter=presenter,
output=args.output,
limit=args.output_limit,
labels=labels,
label_smoothing_window=args.label_smoothing)
cap = open_images_capture(args.input, args.loop)
run_pipeline(cap,
args.architecture_type,
models,
result_presenter.render_frame,
args.raw_output_message,
seq_size=seq_size,
fps=cap.fps())
for rep in presenter.reportMeans():
log.info(rep)
def main():
metrics = PerformanceMetrics()
args = build_argparser().parse_args()
log.info('Initializing Inference Engine...')
ie = IECore()
plugin_config = get_plugin_configs(args.device, args.num_streams,
args.num_threads)
log.info('Loading network...')
model = SegmentationModel(ie, args.model)
pipeline = AsyncPipeline(ie,
model,
plugin_config,
device=args.device,
max_num_requests=args.num_infer_requests)
cap = open_images_capture(args.input, args.loop)
next_frame_id = 0
next_frame_id_to_show = 0
log.info('Starting inference...')
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
visualizer = Visualizer(args.colors)
presenter = None
video_writer = cv2.VideoWriter()
while True:
if pipeline.is_ready():
# Get new image/frame
start_time = perf_counter()
frame = cap.read()
if frame is None:
if next_frame_id == 0:
raise ValueError("Can't read an image from the input")
break
if next_frame_id == 0:
presenter = monitors.Presenter(
args.utilization_monitors, 55,
(round(frame.shape[1] / 4), round(frame.shape[0] / 8)))
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")
# Submit for inference
pipeline.submit_data(frame, next_frame_id, {
'frame': frame,
'start_time': start_time
})
next_frame_id += 1
else:
# Wait for empty request
pipeline.await_any()
if pipeline.callback_exceptions:
raise pipeline.callback_exceptions[0]
# Process all completed requests
results = pipeline.get_result(next_frame_id_to_show)
if results:
objects, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
frame = visualizer.overlay_masks(frame, objects)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Segmentation Results', frame)
key = cv2.waitKey(1)
if key == 27 or key == 'q' or key == 'Q':
break
presenter.handleKey(key)
next_frame_id_to_show += 1
pipeline.await_all()
# Process completed requests
while pipeline.has_completed_request():
results = pipeline.get_result(next_frame_id_to_show)
if results:
objects, frame_meta = results
frame = frame_meta['frame']
start_time = frame_meta['start_time']
frame = visualizer.overlay_masks(frame, objects)
presenter.drawGraphs(frame)
metrics.update(start_time, frame)
if video_writer.isOpened() and (
args.output_limit <= 0
or next_frame_id_to_show <= args.output_limit - 1):
video_writer.write(frame)
if not args.no_show:
cv2.imshow('Segmentation Results', frame)
key = cv2.waitKey(1)
next_frame_id_to_show += 1
else:
break
metrics.print_total()
print(presenter.reportMeans())
def main():
log.basicConfig(format='[ %(levelname)s ] %(message)s',
level=log.INFO,
stream=sys.stdout)
args = build_argparser().parse_args()
# Plugin initialization for specified device and load extensions library if specified.
log.info('Creating Inference Engine...')
ie = IECore()
if args.cpu_extension and 'CPU' in args.device:
ie.add_extension(args.cpu_extension, 'CPU')
# Read IR
log.info('Loading Mask-RCNN network')
mask_rcnn_net = ie.read_network(
args.mask_rcnn_model,
os.path.splitext(args.mask_rcnn_model)[0] + '.bin')
log.info('Loading encoder part of text recognition network')
text_enc_net = ie.read_network(
args.text_enc_model,
os.path.splitext(args.text_enc_model)[0] + '.bin')
log.info('Loading decoder part of text recognition network')
text_dec_net = ie.read_network(
args.text_dec_model,
os.path.splitext(args.text_dec_model)[0] + '.bin')
model_required_inputs = {'image'}
old_model_required_inputs = {'im_data', 'im_info'}
if set(mask_rcnn_net.input_info) == model_required_inputs:
old_model = False
required_output_keys = {'boxes', 'labels', 'masks', 'text_features.0'}
n, c, h, w = mask_rcnn_net.input_info['image'].input_data.shape
elif set(mask_rcnn_net.input_info) == old_model_required_inputs:
old_model = True
required_output_keys = {
'boxes', 'scores', 'classes', 'raw_masks', 'text_features'
}
n, c, h, w = mask_rcnn_net.input_info['im_data'].input_data.shape
args.alphabet = ' 0123456789abcdefghijklmnopqrstuvwxyz'
args.tr_threshold = 0
else:
raise RuntimeError(
'Demo supports only topologies with the following input keys: '
f'{model_required_inputs} or {old_model_required_inputs}.')
assert required_output_keys.issubset(mask_rcnn_net.outputs.keys()), \
f'Demo supports only topologies with the following output keys: {required_output_keys}' \
f'Found: {mask_rcnn_net.outputs.keys()}.'
assert n == 1, 'Only batch 1 is supported by the demo application'
log.info('Loading IR to the plugin...')
mask_rcnn_exec_net = ie.load_network(network=mask_rcnn_net,
device_name=args.device,
num_requests=2)
text_enc_exec_net = ie.load_network(network=text_enc_net,
device_name=args.device)
text_dec_exec_net = ie.load_network(network=text_dec_net,
device_name=args.device)
hidden_shape = text_dec_net.input_info[
args.trd_input_prev_hidden].input_data.shape
del mask_rcnn_net
del text_enc_net
del text_dec_net
input_source = args.input_source
if os.path.isdir(input_source):
cap = FolderCapture(input_source)
else:
try:
input_source = int(args.input_source)
cap = cv2.VideoCapture(input_source)
cap.set(cv2.CAP_PROP_BUFFERSIZE, 1)
except ValueError:
cap = cv2.VideoCapture(input_source)
if not cap.isOpened():
raise RuntimeError('Failed to open "{}"'.format(input_source))
ret, frame = cap.read()
if not ret:
raise RuntimeError("Can't read an image from the input")
if args.no_track:
tracker = None
else:
tracker = StaticIOUTracker()
visualizer = Visualizer(['__background__', 'text'],
show_boxes=args.show_boxes,
show_scores=args.show_scores)
render_time = 0
presenter = monitors.Presenter(args.utilization_monitors, 45,
(frame.shape[1] // 4, frame.shape[0] // 8))
log.info('Starting inference...')
print(
"To close the application, press 'CTRL+C' here or switch to the output window and press ESC key"
)
while ret:
if not args.keep_aspect_ratio:
# Resize the image to a target size.
scale_x = w / frame.shape[1]
scale_y = h / frame.shape[0]
input_image = cv2.resize(frame, (w, h))
else:
# Resize the image to keep the same aspect ratio and to fit it to a window of a target size.
scale_x = scale_y = min(h / frame.shape[0], w / frame.shape[1])
input_image = cv2.resize(frame, None, fx=scale_x, fy=scale_y)
input_image_size = input_image.shape[:2]
input_image = np.pad(input_image,
((0, h - input_image_size[0]),
(0, w - input_image_size[1]), (0, 0)),
mode='constant',
constant_values=0)
# Change data layout from HWC to CHW.
input_image = input_image.transpose((2, 0, 1))
input_image = input_image.reshape((n, c, h, w)).astype(np.float32)
input_image_info = np.asarray(
[[input_image_size[0], input_image_size[1], 1]], dtype=np.float32)
# Run the net.
inf_start = time.time()
if old_model:
outputs = mask_rcnn_exec_net.infer({
'im_data': input_image,
'im_info': input_image_info
})
else:
outputs = mask_rcnn_exec_net.infer({'image': input_image})
# Parse detection results of the current request
if old_model:
boxes = outputs['boxes']
scores = outputs['scores']
classes = outputs['classes'].astype(np.uint32)
raw_masks = outputs['raw_masks']
text_features = outputs['text_features']
else:
boxes = outputs['boxes'][:, :4]
scores = outputs['boxes'][:, 4]
classes = outputs['labels'].astype(np.uint32)
raw_masks = outputs['masks']
text_features = outputs['text_features.0']
# Filter out detections with low confidence.
detections_filter = scores > args.prob_threshold
scores = scores[detections_filter]
classes = classes[detections_filter]
boxes = boxes[detections_filter]
raw_masks = raw_masks[detections_filter]
text_features = text_features[detections_filter]
boxes[:, 0::2] /= scale_x
boxes[:, 1::2] /= scale_y
masks = []
for box, cls, raw_mask in zip(boxes, classes, raw_masks):
if old_model:
raw_mask = raw_mask[cls, ...]
mask = segm_postprocess(box, raw_mask, frame.shape[0],
frame.shape[1])
masks.append(mask)
texts = []
for feature in text_features:
feature = text_enc_exec_net.infer({'input': feature})['output']
feature = np.reshape(feature,
(feature.shape[0], feature.shape[1], -1))
feature = np.transpose(feature, (0, 2, 1))
hidden = np.zeros(hidden_shape)
prev_symbol_index = np.ones((1, )) * SOS_INDEX
text = ''
text_confidence = 1.0
for i in range(MAX_SEQ_LEN):
decoder_output = text_dec_exec_net.infer({
args.trd_input_prev_symbol:
prev_symbol_index,
args.trd_input_prev_hidden:
hidden,
args.trd_input_encoder_outputs:
feature
})
symbols_distr = decoder_output[args.trd_output_symbols_distr]
symbols_distr_softmaxed = softmax(symbols_distr, axis=1)[0]
prev_symbol_index = int(np.argmax(symbols_distr, axis=1))
text_confidence *= symbols_distr_softmaxed[prev_symbol_index]
if prev_symbol_index == EOS_INDEX:
break
text += args.alphabet[prev_symbol_index]
hidden = decoder_output[args.trd_output_cur_hidden]
texts.append(text if text_confidence >= args.tr_threshold else '')
inf_end = time.time()
inf_time = inf_end - inf_start
render_start = time.time()
if len(boxes) and args.raw_output_message:
log.info('Detected boxes:')
log.info(
' Class ID | Confidence | XMIN | YMIN | XMAX | YMAX '
)
for box, cls, score, mask in zip(boxes, classes, scores, masks):
log.info(
'{:>10} | {:>10f} | {:>8.2f} | {:>8.2f} | {:>8.2f} | {:>8.2f} '
.format(cls, score, *box))
# Get instance track IDs.
masks_tracks_ids = None
if tracker is not None:
masks_tracks_ids = tracker(masks, classes)
presenter.drawGraphs(frame)
# Visualize masks.
frame = visualizer(frame, boxes, classes, scores, masks, texts,
masks_tracks_ids)
# Draw performance stats.
inf_time_message = 'Inference and post-processing time: {:.3f} ms'.format(
inf_time * 1000)
render_time_message = 'OpenCV rendering time: {:.3f} ms'.format(
render_time * 1000)
cv2.putText(frame, inf_time_message, (15, 15),
cv2.FONT_HERSHEY_COMPLEX, 0.5, (200, 10, 10), 1)
cv2.putText(frame, render_time_message, (15, 30),
cv2.FONT_HERSHEY_COMPLEX, 0.5, (10, 10, 200), 1)
# Print performance counters.
if args.perf_counts:
perf_counts = mask_rcnn_exec_net.requests[0].get_perf_counts()
log.info('Performance counters:')
print('{:<70} {:<15} {:<15} {:<15} {:<10}'.format(
'name', 'layer_type', 'exet_type', 'status', 'real_time, us'))
for layer, stats in perf_counts.items():
print('{:<70} {:<15} {:<15} {:<15} {:<10}'.format(
layer, stats['layer_type'], stats['exec_type'],
stats['status'], stats['real_time']))
if not args.no_show:
# Show resulting image.
cv2.imshow('Results', frame)
render_end = time.time()
render_time = render_end - render_start
if not args.no_show:
key = cv2.waitKey(args.delay)
esc_code = 27
if key == esc_code:
break
presenter.handleKey(key)
ret, frame = cap.read()
print(presenter.reportMeans())
cv2.destroyAllWindows()
cap.release()
def run_demo(args):
cap = open_images_capture(args.input, args.loop)
log.info('OpenVINO Inference Engine')
log.info('\tbuild: {}'.format(get_version()))
ie = IECore()
log.info('Reading Object Detection model {}'.format(args.model_od))
detector_person = Detector(ie,
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(ie,
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)