Use Case and High-Level Description

Face detector for driver monitoring and similar scenarios. The network features a default MobileNet backbone that includes depth-wise convolutions to reduce the amount of computation for the 3x3 convolution block.



Metric Value
AP (head height >10px) 37.4%
AP (head height >32px) 84.8%
AP (head height >64px) 93.1%
AP (head height >100px) 94.1%
Min head size 90x90 pixels on 1080p
GFlops 2.835
MParams 1.053
Source framework Caffe*

Average Precision (AP) is defined as an area under the precision/recall curve. Numbers are on Wider Face validation subset.


Image, name: input, shape: 1, 3, 384, 672 in the format B, C, H, W, where:

  • B - batch size
  • C - number of channels
  • H - image height
  • W - image width

Expected color order is BGR.


The net outputs blob with shape: 1, 1, 200, 7 in the format 1, 1, N, 7, where N is the number of detected bounding boxes. The results are sorted by confidence in decreasing order. Each detection has the format [image_id, label, conf, x_min, y_min, x_max, y_max], where:

  • image_id - ID of the image in the batch
  • label - predicted class ID (1 - face)
  • conf - confidence for the predicted class
  • (x_min, y_min) - coordinates of the top left bounding box corner
  • (x_max, y_max) - coordinates of the bottom right bounding box corner

Legal Information

[*] Other names and brands may be claimed as the property of others.