def recognize_from_video():
# net initialize
detector = ailia.Net(
DETECTOR_MODEL_PATH, DETECTOR_WEIGHT_PATH, env_id=args.env_id
)
estimator = ailia.Net(
ESTIMATOR_MODEL_PATH, ESTIMATOR_WEIGHT_PATH, env_id=args.env_id
)
capture = webcamera_utils.get_capture(args.video)
# create video writer if savepath is specified as video format
if args.savepath != SAVE_IMAGE_PATH:
f_h = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
f_w = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
writer = webcamera_utils.get_writer(args.savepath, f_h, f_w)
else:
writer = None
while(True):
ret, frame = capture.read()
if (cv2.waitKey(1) & 0xFF == ord('q')) or not ret:
break
_, img128, scale, pad = but.resize_pad(frame[:, :, ::-1])
input_data = img128.astype('float32') / 255.
input_data = np.expand_dims(np.moveaxis(input_data, -1, 0), 0)
# inference
# Person detection
detector_out = detector.predict([input_data])
detections = but.detector_postprocess(detector_out)
count = len(detections) if detections[0].size > 0 else 0
# Pose estimation
landmarks = []
flags = []
if count > 0:
img, affine, _ = but.estimator_preprocess(
frame, detections, scale, pad
)
flags, normalized_landmarks, _ = estimator.predict([img])
landmarks = but.denormalize_landmarks(normalized_landmarks, affine)
# postprocessing
display_result(frame, count, landmarks, flags)
cv2.imshow('frame', frame)
# save results
if writer is not None:
writer.write(frame)
capture.release()
cv2.destroyAllWindows()
if writer is not None:
writer.release()
logger.info('Script finished successfully.')
pass
def recognize_pose(frame, detector, estimator, out_frame=None):
_, img128, scale, pad = bput.resize_pad(frame[:, :, ::-1])
input_data = img128.astype('float32') / 255.
input_data = np.expand_dims(np.moveaxis(input_data, -1, 0), 0)
# inference
# Person detection
detector_out = detector.predict([input_data])
detections = bput.detector_postprocess(
detector_out,
anchor_path="../../pose_estimation/blazepose/anchors.npy",
min_score_thresh=0.5)
count = len(detections) if detections[0].size > 0 else 0
# display bbox
if args.bbox:
detections2 = bput.denormalize_detections(detections[0].copy(), scale,
pad)
display_hand_box(out_frame, detections2)
# Pose estimation
landmarks = []
flags = []
if count > 0:
imgs, affine, _ = bput.estimator_preprocess(frame, detections, scale,
pad)
#flags, normalized_landmarks, _ = estimator.predict([imgs])
#print(flags.shape)
#print(normalized_landmarks.shape)
flags = np.zeros((imgs.shape[0]))
normalized_landmarks = np.zeros((imgs.shape[0], 31, 4))
for i in range(imgs.shape[0]):
flag, normalized_landmark, _ = estimator.predict(
[imgs[i:i + 1, :, :, :]])
flags[i] = flag
normalized_landmarks[i] = normalized_landmark
landmarks = bput.denormalize_landmarks(normalized_landmarks, affine)
# postprocessing
display_result_pose(out_frame, count, landmarks, flags)
def pose_estimate(net, det_net, img):
h, w = img.shape[:2]
src_img = img
logger.debug(f'input image shape: {img.shape}')
_, img224, scale, pad = but.resize_pad(img)
img224 = img224.astype('float32') / 255.
img224 = np.expand_dims(img224, axis=0)
detector_out = det_net.predict([img224])
detections = but.detector_postprocess(detector_out)
count = len(detections) if detections[0].size != 0 else 0
# Pose estimation
imgs = []
if 0 < count:
imgs, affine, _ = but.estimator_preprocess(
src_img, detections, scale, pad
)
flags = []
landmarks = []
for i, img in enumerate(imgs):
img = np.expand_dims(img, axis=0)
output = net.predict([img])
normalized_landmarks, f, _, _, _ = output
normalized_landmarks = postprocess(normalized_landmarks)
flags.append(f[0])
landmarks.append(normalized_landmarks[0])
if len(imgs)>=1:
landmarks = np.stack(landmarks)
landmarks = but.denormalize_landmarks(landmarks, affine)
return flags, landmarks
def recognize_from_image():
# net initialize
detector = ailia.Net(DETECTOR_MODEL_PATH,
DETECTOR_WEIGHT_PATH,
env_id=args.env_id)
estimator = ailia.Net(ESTIMATOR_MODEL_PATH,
ESTIMATOR_WEIGHT_PATH,
env_id=args.env_id)
# input image loop
for image_path in args.input:
# prepare input data
logger.info(image_path)
src_img = cv2.imread(image_path)
_, img128, scale, pad = but.resize_pad(src_img[:, :, ::-1])
input_data = img128.astype('float32') / 255.
input_data = np.expand_dims(np.moveaxis(input_data, -1, 0), 0)
# inference
logger.info('Start inference...')
if args.benchmark:
logger.info('BENCHMARK mode')
for _ in range(5):
start = int(round(time.time() * 1000))
# Person detection
detector_out = detector.predict([input_data])
detections = but.detector_postprocess(detector_out)
count = len(detections) if detections[0].size > 0 else 0
# Pose estimation
landmarks = []
flags = []
if count > 0:
img, affine, _ = but.estimator_preprocess(
src_img, detections, scale, pad)
flags, normalized_landmarks, _ = estimator.predict([img])
landmarks = but.denormalize_landmarks(
normalized_landmarks, affine)
end = int(round(time.time() * 1000))
logger.info(f'\tailia processing time {end - start} ms')
else:
# Person detection
detector_out = detector.predict([input_data])
detections = but.detector_postprocess(detector_out)
count = len(detections) if detections[0].size != 0 else 0
# Pose estimation
landmarks = []
flags = []
if count > 0:
img, affine, _ = but.estimator_preprocess(
src_img, detections, scale, pad)
flags, normalized_landmarks, _ = estimator.predict([img])
landmarks = but.denormalize_landmarks(normalized_landmarks,
affine)
# postprocessing
logger.info(f'person_count={count}')
display_result(src_img, count, landmarks, flags)
savepath = get_savepath(args.savepath, image_path)
logger.info(f'saved at : {savepath}')
cv2.imwrite(savepath, src_img)
logger.info('Script finished successfully.')