526互联

openvino头部姿态识别

发布时间 2023-09-27 15:26:54作者: FeiYull

　　网络的输入、输出如下：

Performance

Inputs

  1. name: "data" , shape: [1x3x60x60] - An input image in [1xCxHxW] format. Expected color order is BGR.

Outputs

Output layer names in Inference Engine format:

  1. name: "angle_y_fc", shape: [1, 1] - Estimated yaw (in degrees).
  2. name: "angle_p_fc", shape: [1, 1] - Estimated pitch (in degrees).
  3. name: "angle_r_fc", shape: [1, 1] - Estimated roll (in degrees).

Output layer names in Caffe* format:

  1. name: "fc_y", shape: [1, 1] - Estimated yaw (in degrees).
  2. name: "fc_p", shape: [1, 1] - Estimated pitch (in degrees).
  3. name: "fc_r", shape: [1, 1] - Estimated roll (in degrees).

Each output contains one float value that represents value in Tait-Bryan angles (yaw, pitch or roll).

　　很显然，网络最终层有3个输出（三个欧拉角），所以代码更改如下：

1 angle_y_fc = next(em_it) # yaw
2 angle_p_fc = next(em_it) # pitch
3 angle_r_fc = next(em_it) # roll

　　另外三个欧拉角的是有取值范围的

MetricValue
Supported ranges YAW [-90,90], PITCH [-70,70], ROLL [-70,70]
GFlops 0.105
MParams 1.911
Source framework Caffe*