def get_default_network(
pose_resnet_config: PoseResNetModelConfig) -> Pose2DNetResnet:
model = pose_resnet.get_pose_net(pose_resnet_config)
# logger.info('=> loading model from {}'.format(cfg.pose_estimator.model_state_file))
model.load_state_dict(torch.load(pose_resnet_config.model_state_file))
model = model.cuda()
model.eval()
return Pose2DNetResnet(model, SkeletonStickman)
buffer_size = 20
action_names = [Action.IDLE.name, Action.WALK.name, Action.WAVE.name]
use_action_recognition = True
use_quick_n_dirty = False
# Input Provider
input_provider = WebcamProvider(camera_number=0,
image_size=image_size,
fps=fps)
# input_provider = ImgDirProvider(
# "/media/disks/beta/records/real_cam/2019_03_13_Freilichtmuseum_Dashcam_01/full",
# image_size=image_size, fps=fps)
fps_tracker = FPSTracker(average_over_seconds=1)
# Pose Network
pose_model = pose_resnet.get_pose_net(pose_resnet_config)
logger.info('=> loading model from {}'.format(
pose_resnet_config.model_state_file))
pose_model.load_state_dict(torch.load(pose_resnet_config.model_state_file))
pose_model = pose_model.cuda()
pose_model.eval()
pose_net = Pose2DNetResnet(pose_model, skeleton_type)
pose_tracker = PoseTracker(image_size=image_size,
skeleton_type=skeleton_type)
# Action Network
action_model = ShuffleNetV2(input_size=32, n_class=3)
state_dict = torch.load(ehpi_model_state_file)
action_model.load_state_dict(state_dict)
action_model.cuda()