mobilefacedet-v1-mxnet

Use Case and High-Level Description

MobileFace Detection V1 is a Light and Fast Face Detector for Edge Devices (LFFD) model based on Yolo V3 architecture and trained with MXNet*. For details see the repository and paper.

Specification

Metric

Value

Type

Detection

GFLOPs

3.5456

MParams

7.6828

Source framework

MXNet*

Accuracy

Metric

Value

mAP

78.7488%

Input

Original model

Image, name - data, shape - 1, 256, 256, 3, format -B, H, W, C, where:

  • B - batch size

  • H - height

  • W - width

  • C - channel

Expected color order - BGR.

Converted model

The converted model has the same parameters as the original model.

WARNING: Please note that the input layout of the converted model is B, H, W, C.

Output

Original model

  1. The array of detection summary info, name - yolov30_slice_axis1, shape - 1, 18, 8, 8. The anchor values are 118,157,  186,248,  285,379.

  2. The array of detection summary info, name - yolov30_slice_axis2, shape - 1, 18, 16, 16. The anchor values are 43,54,  60,75,  80,106.

  3. The array of detection summary info, name - yolov30_slice_axis3, shape - 1, 18, 32, 32. The anchor values are 10,12,  16,20,  23,29.

For each case format is B, N*DB, Cx, Cy, where:

  • B - batch size

  • N - number of detection boxes for cell

  • DB - size of each detection box

  • Cx, Cy - cell index

Detection box has format [x, y, h, w, box_score, face_score], where:

  • (x, y) - raw coordinates of box center, apply sigmoid function to get coordinates relative to the cell

  • h,w - raw height and width of box, apply exponential function and multiply by corresponding anchors to get height and width values relative to cell

  • box_score - confidence of detection box, apply sigmoid function to get confidence in [0, 1] range

  • face_score - probability that detected object belongs to face class, apply sigmoid function to get confidence in [0, 1] range

Converted model

  1. The array of detection summary info, name - yolov30_yolooutputv30_conv0_fwd/YoloRegion, shape - 1, 18, 8, 8. The anchor values are 118,157,  186,248,  285,379.

  2. The array of detection summary info, name - yolov30_yolooutputv31_conv0_fwd/YoloRegion, shape - 1, 18, 16, 16. The anchor values are 43,54,  60,75,  80,106.

  3. The array of detection summary info, name - yolov30_yolooutputv32_conv0_fwd/YoloRegion, shape - 1, 18, 32, 32. The anchor values are 10,12,  16,20,  23,29.

For each case format is B, N*DB, Cx, Cy, where:

  • B - batch size

  • N - number of detection boxes for cell

  • DB - size of each detection box

  • Cx, Cy - cell index

Detection box has format [x, y, h, w, box_score, face_score], where:

  • (x, y) - raw coordinates of box center to the cell

  • h, w - raw height and width of box, apply exponential function and multiply by corresponding anchors to get height and width values relative to cell

  • box_score - confidence of detection box in [0, 1] range

  • face_score - probability that detected object belongs to face class in [0, 1] range

Download a Model and Convert it into OpenVINO™ IR Format

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

An example of using the Model Downloader:

omz_downloader --name <model_name>

An example of using the Model Converter:

omz_converter --name <model_name>

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities: