def process_video_frame(video_frame):
video_frame = video_frame[:, :, :3]
video_frame = scipy.misc.imresize(video_frame, (240, 320))
image_v = np.expand_dims((video_frame.astype('float') / 255.0) - 0.5, 0)
keypoint_coord3d_tf, scale_tf, center_tf, keypoints_scoremap_tf = network_elements
keypoint_coord3d_v, scale_v, center_v, keypoints_scoremap_v = sess.run(
[keypoint_coord3d_tf, scale_tf, center_tf, keypoints_scoremap_tf],
feed_dict={image_tf: image_v})
keypoints_scoremap_v = np.squeeze(keypoints_scoremap_v)
keypoint_coord3d_v = np.squeeze(keypoint_coord3d_v)
# post processing
coord_hw_crop = detect_keypoints(np.squeeze(keypoints_scoremap_v))
coord_hw = trafo_coords(coord_hw_crop, center_v, scale_v, 256)
plot_hand_2d(coord_hw, video_frame)
score_label = process_keypoints(keypoint_coord3d_v)
if score_label is not None:
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(video_frame, score_label, (10, 200), font, 1.0,
(255, 0, 0), 2, cv2.LINE_AA)
return video_frame
def process_img(known_finger_poses, img_name):
image_raw = scipy.misc.imread(img_name)[:, :, :3]
image_raw = np.array(Image.fromarray(image_raw).resize((320, 240)))
image_v = np.expand_dims((image_raw.astype('float') / 255.0) - 0.5, 0)
#if args.plot_fingers == 1:
scale_v, center_v, keypoints_scoremap_v, \
keypoint_coord3d_v = sess.run([scale_tf, center_tf, keypoints_scoremap_tf,\
keypoint_coord3d_tf], feed_dict = {image_tf: image_v})
keypoints_scoremap_v = np.squeeze(keypoints_scoremap_v)
keypoint_coord3d_v = np.squeeze(keypoint_coord3d_v)
# post processing
coord_hw_crop = detect_keypoints(np.squeeze(keypoints_scoremap_v))
coord_hw = trafo_coords(coord_hw_crop, center_v, scale_v, 256)
plot_hand_2d(coord_hw, image_raw)
# Classifying based on Geometry
#if args.solve_by == 0:
score_label = predict_by_geometry(keypoint_coord3d_v, known_finger_poses,
0.55)
# Classifying based on Neural networks
# elif args.solve_by == 1:
# score_label = predict_by_neural_network(keypoint_coord3d_v, known_finger_poses,
# args.pb_file, args.threshold)
# Classifying based on SVM
# elif args.solve_by == 2:
# score_label = predict_by_svm(keypoint_coord3d_v, known_finger_poses, args.svc_file)
# save processed image
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(image_raw, score_label, (10, 200), font, 1.0, (255, 0, 0), 2,
cv2.LINE_AA)
file_save_path = os.path.join('../images_out',
"{}.png".format(int(time.time())))
mpimg.imsave(file_save_path, image_raw)
return score_label
image_raw = scipy.misc.imresize(image_raw, (240, 320))
image_v = np.expand_dims((image_raw.astype('float') / 255.0) - 0.5, 0)
if args.plot_fingers == 1:
scale_v, center_v, keypoints_scoremap_v, \
keypoint_coord3d_v = sess.run([scale_tf, center_tf, keypoints_scoremap_tf,\
keypoint_coord3d_tf], feed_dict = {image_tf: image_v})
keypoints_scoremap_v = np.squeeze(keypoints_scoremap_v)
keypoint_coord3d_v = np.squeeze(keypoint_coord3d_v)
# post processing
coord_hw_crop = detect_keypoints(np.squeeze(keypoints_scoremap_v))
coord_hw = trafo_coords(coord_hw_crop, center_v, scale_v, 256)
plot_hand_2d(coord_hw, image_raw)
else:
keypoint_coord3d_v = sess.run(keypoint_coord3d_tf,
feed_dict={image_tf: image_v})
# Classifying based on Geometry
if args.solve_by == 0:
score_label = predict_by_geometry(keypoint_coord3d_v,
known_finger_poses,
args.threshold)
# Classifying based on Neural networks
elif args.solve_by == 1:
score_label = predict_by_neural_network(keypoint_coord3d_v,
known_finger_poses,
args.pb_file,