def inference_pose():
print('Thread "pose" started')
stop_watch = StopWatch(window=10)
while True:
while len(det_result_queue) < 1:
time.sleep(0.001)
with det_result_queue_mutex:
ts_input, frame, t_info, mmdet_results = det_result_queue.popleft()
pose_results_list = []
for model_info, pose_history in zip(pose_model_list,
pose_history_list):
model_name = model_info['name']
pose_model = model_info['model']
cat_ids = model_info['cat_ids']
pose_results_last = pose_history['pose_results_last']
next_id = pose_history['next_id']
with stop_watch.timeit(model_name):
# process mmdet results
det_results = process_mmdet_results(
mmdet_results,
class_names=det_model.CLASSES,
cat_ids=cat_ids)
# inference pose model
dataset_name = pose_model.cfg.data['test']['type']
pose_results, _ = inference_top_down_pose_model(
pose_model,
frame,
det_results,
bbox_thr=args.det_score_thr,
format='xyxy',
dataset=dataset_name)
pose_results, next_id = get_track_id(pose_results,
pose_results_last,
next_id,
use_oks=False,
tracking_thr=0.3,
use_one_euro=True,
fps=None)
pose_results_list.append(pose_results)
# update pose history
pose_history['pose_results_last'] = pose_results
pose_history['next_id'] = next_id
t_info += stop_watch.report_strings()
with pose_result_queue_mutex:
pose_result_queue.append((ts_input, t_info, pose_results_list))
event_inference_done.set()
def test_stopwatch():
window_size = 5
test_loop = 10
outer_time = 100
inner_time = 100
stop_watch = StopWatch(window=window_size)
for _ in range(test_loop):
with stop_watch.timeit():
time.sleep(outer_time / 1000.)
with stop_watch.timeit('inner'):
time.sleep(inner_time / 1000.)
_ = stop_watch.report()
_ = stop_watch.report_strings()
def test_stopwatch():
window_size = 5
test_loop = 10
outer_time = 100
inner_time = 100
stop_watch = StopWatch(window=window_size)
for _ in range(test_loop):
with stop_watch.timeit():
time.sleep(outer_time / 1000.)
with stop_watch.timeit('inner'):
time.sleep(inner_time / 1000.)
report = stop_watch.report()
_ = stop_watch.report_strings()
np.testing.assert_allclose(report['_FPS_'],
outer_time + inner_time,
rtol=0.01)
np.testing.assert_allclose(report['inner'], inner_time, rtol=0.01)
def inference_detection():
print('Thread "det" started')
stop_watch = StopWatch(window=10)
min_interval = 1.0 / args.inference_fps
_ts_last = None # timestamp when last inference was done
while True:
while len(input_queue) < 1:
time.sleep(0.001)
with input_queue_mutex:
ts_input, frame = input_queue.popleft()
# inference detection
with stop_watch.timeit('Det'):
mmdet_results = inference_detector(det_model, frame)
t_info = stop_watch.report_strings()
with det_result_queue_mutex:
det_result_queue.append((ts_input, frame, t_info, mmdet_results))
# limit the inference FPS
_ts = time.time()
if _ts_last is not None and _ts - _ts_last < min_interval:
time.sleep(min_interval - _ts + _ts_last)
_ts_last = time.time()
def display():
print('Thread "display" started')
stop_watch = StopWatch(window=10)
# initialize result status
ts_inference = None # timestamp of the latest inference result
fps_inference = 0. # infenrece FPS
t_delay_inference = 0. # inference result time delay
pose_results_list = None # latest inference result
t_info = [] # upstream time information (list[str])
# initialize visualization and output
sunglasses_img = None # resource image for sunglasses effect
text_color = (228, 183, 61) # text color to show time/system information
vid_out = None # video writer
# show instructions
print('Keyboard shortcuts: ')
print('"v": Toggle the visualization of bounding boxes and poses.')
print('"s": Toggle the sunglasses effect.')
print('"b": Toggle the bug-eye effect.')
print('"Q", "q" or Esc: Exit.')
while True:
with stop_watch.timeit('_FPS_'):
# acquire a frame from buffer
ts_input, frame = frame_buffer.get()
# input ending signal
if ts_input is None:
break
img = frame
# get pose estimation results
if len(pose_result_queue) > 0:
with pose_result_queue_mutex:
_result = pose_result_queue.popleft()
_ts_input, t_info, pose_results_list = _result
_ts = time.time()
if ts_inference is not None:
fps_inference = 1.0 / (_ts - ts_inference)
ts_inference = _ts
t_delay_inference = (_ts - _ts_input) * 1000
# visualize detection and pose results
if pose_results_list is not None:
for model_info, pose_results in zip(pose_model_list,
pose_results_list):
pose_model = model_info['model']
bbox_color = model_info['bbox_color']
dataset_name = pose_model.cfg.data['test']['type']
# show pose results
if args.show_pose:
img = vis_pose_result(pose_model,
img,
pose_results,
radius=4,
thickness=2,
dataset=dataset_name,
kpt_score_thr=args.kpt_thr,
bbox_color=bbox_color)
# sunglasses effect
if args.sunglasses:
if dataset_name == 'TopDownCocoDataset':
left_eye_idx = 1
right_eye_idx = 2
elif dataset_name == 'AnimalPoseDataset':
left_eye_idx = 0
right_eye_idx = 1
else:
raise ValueError(
'Sunglasses effect does not support'
f'{dataset_name}')
if sunglasses_img is None:
# The image attributes to:
# https://www.vecteezy.com/free-vector/glass
# Glass Vectors by Vecteezy
sunglasses_img = cv2.imread(
'demo/resources/sunglasses.jpg')
img = apply_sunglasses_effect(img, pose_results,
sunglasses_img,
left_eye_idx,
right_eye_idx)
# bug-eye effect
if args.bugeye:
if dataset_name == 'TopDownCocoDataset':
left_eye_idx = 1
right_eye_idx = 2
elif dataset_name == 'AnimalPoseDataset':
left_eye_idx = 0
right_eye_idx = 1
else:
raise ValueError('Bug-eye effect does not support'
f'{dataset_name}')
img = apply_bugeye_effect(img, pose_results,
left_eye_idx, right_eye_idx)
# delay control
if args.display_delay > 0:
t_sleep = args.display_delay * 0.001 - (time.time() - ts_input)
if t_sleep > 0:
time.sleep(t_sleep)
t_delay = (time.time() - ts_input) * 1000
# show time information
t_info_display = stop_watch.report_strings() # display fps
t_info_display.append(f'Inference FPS: {fps_inference:>5.1f}')
t_info_display.append(f'Delay: {t_delay:>3.0f}')
t_info_display.append(
f'Inference Delay: {t_delay_inference:>3.0f}')
t_info_str = ' | '.join(t_info_display + t_info)
cv2.putText(img, t_info_str, (20, 20), cv2.FONT_HERSHEY_DUPLEX,
0.3, text_color, 1)
# collect system information
sys_info = [
f'RES: {img.shape[1]}x{img.shape[0]}',
f'Buffer: {frame_buffer.qsize()}/{frame_buffer.maxsize}'
]
if psutil_proc is not None:
sys_info += [
f'CPU: {psutil_proc.cpu_percent():.1f}%',
f'MEM: {psutil_proc.memory_percent():.1f}%'
]
sys_info_str = ' | '.join(sys_info)
cv2.putText(img, sys_info_str, (20, 40), cv2.FONT_HERSHEY_DUPLEX,
0.3, text_color, 1)
# save the output video frame
if args.out_video_file is not None:
if vid_out is None:
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
fps = args.out_video_fps
frame_size = (img.shape[1], img.shape[0])
vid_out = cv2.VideoWriter(args.out_video_file, fourcc, fps,
frame_size)
vid_out.write(img)
# display
cv2.imshow('mmpose webcam demo', img)
keyboard_input = cv2.waitKey(1)
if keyboard_input in (27, ord('q'), ord('Q')):
break
elif keyboard_input == ord('s'):
args.sunglasses = not args.sunglasses
elif keyboard_input == ord('b'):
args.bugeye = not args.bugeye
elif keyboard_input == ord('v'):
args.show_pose = not args.show_pose
cv2.destroyAllWindows()
if vid_out is not None:
vid_out.release()
event_exit.set()