Dataset
Dataset Name | Training Images | Validation Images | Class Labels | License |
---|---|---|---|---|
Open Images Dataset | 20,000 [1] | 1,975 | 0: Human Face | Annotations under CC BY 4.0, Images under CC BY 2.0 |
Model Zoo
DeGirum’s face detection model zoo offers models trained with the Ultralytics repository.
Model Architecture | Input Size | Precision | Runtime | Device Type | mAP 50-95 | mAP 50 | FPS |
---|---|---|---|---|---|---|---|
yolov5nu_relu6_face | 640x640 | INT8 | N2X | ORCA1 | 0.5677 | 0.7874 | 117.6 |
yolov5nu_relu6_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5683 | 0.7882 | – |
yolov5nu_relu6_face | 640x640 | INT8 | OPENVINO | CPU | 0.5570 | 0.7764 | – |
yolov5nu_relu6_face | 640x640 | FP32 | OPENVINO | CPU | 0.5695 | 0.7898 | – |
yolov5nu_silu_face | 640x640 | INT8 | N2X | ORCA1 | 0.5591 | 0.7784 | – |
yolov5nu_silu_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5590 | 0.7774 | – |
yolov5nu_silu_face | 640x640 | INT8 | OPENVINO | CPU | 0.5613 | 0.7802 | – |
yolov5nu_silu_face | 640x640 | FP32 | OPENVINO | CPU | 0.5677 | 0.7839 | – |
yolov8n_relu6_face | 640x640 | INT8 | N2X | ORCA1 | 0.5679 | 0.7869 | 106.8 |
yolov8n_relu6_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5686 | 0.7876 | – |
yolov8n_relu6_face | 640x640 | INT8 | OPENVINO | CPU | 0.5409 | 0.7586 | – |
yolov8n_relu6_face | 640x640 | FP32 | OPENVINO | CPU | 0.5694 | 0.7871 | – |
yolov8n_silu_face | 640x640 | INT8 | N2X | ORCA1 | 0.5514 | 0.7647 | – |
yolov8n_silu_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5493 | 0.7642 | – |
yolov8n_silu_face | 640x640 | INT8 | OPENVINO | CPU | 0.5594 | 0.7754 | – |
yolov8n_silu_face | 640x640 | FP32 | OPENVINO | CPU | 0.5692 | 0.7799 | – |
yolov5su_relu6_face | 640x640 | INT8 | N2X | ORCA1 | 0.5797 | 0.7967 | 43.9 |
yolov5su_relu6_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5797 | 0.7974 | – |
yolov5su_relu6_face | 640x640 | INT8 | OPENVINO | CPU | 0.5483 | 0.7653 | – |
yolov5su_relu6_face | 640x640 | FP32 | OPENVINO | CPU | 0.5821 | 0.7972 | – |
yolov5su_silu_face | 640x640 | INT8 | N2X | ORCA1 | 0.5766 | 0.7951 | – |
yolov5su_silu_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5751 | 0.7955 | – |
yolov5su_silu_face | 640x640 | INT8 | OPENVINO | CPU | 0.5691 | 0.7887 | – |
yolov5su_silu_face | 640x640 | FP32 | OPENVINO | CPU | 0.5784 | 0.7923 | – |
yolov8s_relu6_face | 640x640 | INT8 | N2X | ORCA1 | 0.5803 | 0.7971 | 39.0 |
yolov8s_relu6_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5806 | 0.7973 | – |
yolov8s_relu6_face | 640x640 | INT8 | OPENVINO | CPU | 0.5465 | 0.7617 | – |
yolov8s_relu6_face | 640x640 | FP32 | OPENVINO | CPU | 0.5817 | 0.7959 | – |
yolov8s_silu_face | 640x640 | INT8 | N2X | ORCA1 | 0.5789 | 0.7958 | – |
yolov8s_silu_face | 640x640 | INT8 | TFLITE | EDGETPU | 0.5784 | 0.7968 | – |
yolov8s_silu_face | 640x640 | INT8 | OPENVINO | CPU | 0.5711 | 0.7836 | – |
yolov8s_silu_face | 640x640 | FP32 | OPENVINO | CPU | 0.5853 | 0.8009 | – |
A subset of face training images are used for training. ↩︎