def main():
print('Tensorflow version: %s' % tf.__version__)
assert tf.__version__.startswith(
'2.'), "Tensorflow version 2.x must be used!"
if args.output_dir:
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
model = args.model # mobilenet resnet50
stride = args.stride # 8, 16, 32 (max 16 for mobilenet)
quant_bytes = args.quant_bytes # float
multiplier = args.multiplier # only for mobilenet
posenet = load_model(model, stride, quant_bytes, multiplier)
filenames = [
f.path for f in os.scandir(args.image_dir)
if f.is_file() and f.path.endswith(('.png', '.jpg'))
]
start = time.time()
for f in filenames:
img = cv2.imread(f)
pose_scores, keypoint_scores, keypoint_coords = posenet.estimate_multiple_poses(
img)
img_poses = posenet.draw_poses(img, pose_scores, keypoint_scores,
keypoint_coords)
posenet.print_scores(f, pose_scores, keypoint_scores, keypoint_coords)
cv2.imwrite(
os.path.join(args.output_dir, os.path.relpath(f, args.image_dir)),
img_poses)
print('Average FPS:', len(filenames) / (time.time() - start))
def main():
print('Tensorflow version: %s' % tf.__version__)
assert tf.__version__.startswith(
'2.'), "Tensorflow version 2.x must be used!"
model = args.model # mobilenet resnet50
stride = args.stride # 8, 16, 32 (max 16 for mobilenet)
quant_bytes = args.quant_bytes # float
multiplier = args.multiplier # only for mobilenet
posenet = load_model(model, stride, quant_bytes, multiplier)
num_images = args.num_images
filenames = [
f.path for f in os.scandir(args.image_dir)
if f.is_file() and f.path.endswith(('.png', '.jpg'))
]
if len(filenames) > num_images:
filenames = filenames[:num_images]
images = {f: cv2.imread(f) for f in filenames}
start = time.time()
for i in range(num_images):
image = images[filenames[i % len(filenames)]]
posenet.estimate_multiple_poses(image)
print('Average FPS:', num_images / (time.time() - start))
def main():
print('Tensorflow version: %s' % tf.__version__)
assert tf.__version__.startswith(
'2.'), "Tensorflow version 2.x must be used!"
model = args.model # mobilenet resnet50
stride = args.stride # 8, 16, 32 (max 16 for mobilenet, min 16 for resnet50)
quant_bytes = args.quant_bytes # float
multiplier = args.multiplier # only for mobilenet
posenet = load_model(model, stride, quant_bytes, multiplier)
if args.file is not None:
cap = cv2.VideoCapture(args.file)
else:
cap = cv2.VideoCapture(args.cam_id)
cap.set(3, args.cam_width)
cap.set(4, args.cam_height)
start = time.time()
frame_count = 0
while True:
res, img = cap.read()
if not res:
raise IOError("webcam failure")
pose_scores, keypoint_scores, keypoint_coords = posenet.estimate_multiple_poses(
img)
overlay_image = draw_skel_and_kp(img,
pose_scores,
keypoint_scores,
keypoint_coords,
min_pose_score=0.15,
min_part_score=0.1)
cv2.imshow('posenet', overlay_image)
frame_count += 1
if cv2.waitKey(1) & 0xFF == ord('q'):
break
print('Average FPS: ', frame_count / (time.time() - start))
def main():
print('Tensorflow version: %s' % tf.__version__)
assert tf.__version__.startswith(
'2.'), "Tensorflow version 2.x must be used!"
model = args.model # mobilenet resnet50
stride = args.stride # 8, 16, 32 (max 16 for mobilenet, min 16 for resnet50)
quant_bytes = args.quant_bytes # float
multiplier = args.multiplier # only for mobilenet
posenet = load_model(model, stride, quant_bytes, multiplier)
# for inspiration, see: https://www.programcreek.com/python/example/72134/cv2.VideoWriter
if args.input_file is not None:
cap = cv2.VideoCapture(args.input_file)
else:
raise IOError("video file not found")
fps = cap.get(cv2.CAP_PROP_FPS)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
video_writer = cv2.VideoWriter(args.output_file, fourcc, fps,
(width, height))
max_pose_detections = 20
# Scaling the input image reduces the quality of the pose detections!
# The speed gain is about the square of the scale factor.
posenet_input_height = 540 # scale factor for the posenet input
posenet_input_scale = 1.0 # posenet_input_height / height # 1.0
posenet_input_width = int(width * posenet_input_scale)
print("posenet_input_scale: %3.4f" % (posenet_input_scale))
start = time.time()
frame_count = 0
ret, frame = cap.read()
while ret:
if posenet_input_scale == 1.0:
frame_rescaled = frame # no scaling
else:
frame_rescaled = \
cv2.resize(frame, (posenet_input_width, posenet_input_height), interpolation=cv2.INTER_LINEAR)
pose_scores, keypoint_scores, keypoint_coords = posenet.estimate_multiple_poses(
frame_rescaled, max_pose_detections)
keypoint_coords_upscaled = keypoint_coords / posenet_input_scale
overlay_frame = draw_skel_and_kp(frame,
pose_scores,
keypoint_scores,
keypoint_coords_upscaled,
min_pose_score=0.15,
min_part_score=0.1)
frame_count += 1
# This is uncompressed video. cv2 has no way to write compressed videos, so we'll have to use ffmpeg to
# compress it afterwards! See:
# https://stackoverflow.com/questions/25998799/specify-compression-quality-in-python-for-opencv-video-object
video_writer.write(overlay_frame)
ret, frame = cap.read()
print('Average FPS: ', frame_count / (time.time() - start))
video_writer.release()
cap.release()
def main():
print("Tensorflow version: %s" % tf.__version__)
assert tf.__version__.startswith(
"2."), "Tensorflow version 2.x must be used!"
if args.image_dir == args.output_dir:
print(
"[WARNING] input dir is the same as output dir -- the pictures will be overwritten"
)
print("Do you wish to continue?: y/n")
if input() != "y":
exit()
# get input folder name to concatenate with out file name
image_dir_name = os.path.basename(args.image_dir).split("./", 1)[0]
start_date = datetime.now().strftime("--%H-%M--%d-%m-%Y")
# paths for posenet out pictures and .csv
output_pic_dir_name = args.output_dir + "/" + image_dir_name + start_date
output_csv_dir_name = args.output_dir + "/" + image_dir_name + start_date + ".csv"
if not os.path.exists(output_pic_dir_name):
os.makedirs(output_pic_dir_name)
model = args.model # mobilenet resnet50
stride = args.stride # 8, 16, 32 (max 16 for mobilenet, min 16 for resnet50)
quant_bytes = args.quant_bytes # float
multiplier = args.multiplier # only for mobilenet
label = args.pose # type of pose for pictures in folder
posenet = load_model(model, stride, quant_bytes, multiplier)
# file_count = su.count_files(args.image_dir)
filenames = [
f.path for f in os.scandir(args.image_dir)
if f.is_file() and f.path.endswith((".png", ".jpg"))
]
# prepare .csv file for points coords
csv_file = open(output_csv_dir_name, "ab")
np.savetxt(csv_file, su.csv_column_names, delimiter=",", fmt="%s")
start = time.time()
for f in filenames:
img = cv2.imread(f)
pose_scores, keypoint_scores, keypoint_coords = posenet.estimate_multiple_poses(
img)
img_poses = posenet.draw_poses(img, pose_scores, keypoint_scores,
keypoint_coords)
cv2.imwrite(
os.path.join(output_pic_dir_name,
os.path.relpath(f, args.image_dir)),
img_poses,
)
one_row = np.concatenate(
(
[pose_scores[0]
], # certainity score for pose for pose 0 in 10 total
keypoint_coords[0, :7, 0], # X coords
keypoint_coords[0, :7, 1], # Y
[label], # pose type
# TODO: picture number (first implement it in vid2pic)
),
axis=0,
).reshape(su.csv_column_names.shape)
np.savetxt(csv_file, one_row, delimiter=",")
print("Average FPS:", len(filenames) / (time.time() - start))
csv_file.close()