Head pose estimation network based on simple, handmade CNN architecture. Angle regression layers are convolutions + ReLU + batch norm + fully connected with one output.
Biwi Kinect Head Pose Database
| Metric | Value |
|---|---|
| Supported ranges | YAW [-90,90], PITCH [-70,70], ROLL [-70,70] |
| GFlops | 0.105 |
| MParams | 1.911 |
| Source framework | Caffe* |
| Angle | Mean ± standard deviation of absolute error |
|---|---|
| yaw | 5.4 ± 4.4 |
| pitch | 5.5 ± 5.3 |
| roll | 4.6 ± 5.6 |
Output layer names in Inference Engine format:
Output layer names in Caffe* format:
Each output contains one float value that represents value in Tait-Bryan angles (yaw, pitсh or roll).
[*] Other names and brands may be claimed as the property of others.