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.
Metric | Value |
---|---|
Type | Detection |
GFLOPs | 3.5456 |
MParams | 7.6828 |
Source framework | MXNet* |
Metric | Value |
---|---|
mAP | 78.7488% |
Image, name - data
, shape - [1x256x256x3]
, format -[BxHxWxC]
where:
B
- batch sizeH
- heightW
- widthC
- channelExpected color order - BGR
.
The converted model has the same parameters as the original model.
WARNING: Please note that the input layout of the converted model is
[BxHxWxC]
.
yolov30_slice_axis1
, shape - 1,18,8,8
. The anchor values are 118,157, 186,248, 285,379
.yolov30_slice_axis2
, shape - 1,18,16,16
. The anchor values are 43,54, 60,75, 80,106
.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 sizeN
- number of detection boxes for cellDB
- size of each detection boxCx
, Cy
- cell indexDetection 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 cellh
,w
- raw height and width of box, apply exponential function and multiply by corresponding anchors to get height and width values relative to cellbox_score
- confidence of detection box, apply sigmoid function to get confidence in [0,1] rangeface_score
- probability that detected object belongs to face class, apply sigmoid function to get confidence in [0,1] rangeyolov30_yolooutputv30_conv0_fwd/YoloRegion
, shape - 1,18,8,8
. The anchor values are 118,157, 186,248, 285,379
.yolov30_yolooutputv31_conv0_fwd/YoloRegion
, shape - 1,18,16,16
. The anchor values are 43,54, 60,75, 80,106
.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 sizeN
- number of detection boxes for cellDB
- size of each detection boxCx
, Cy
- cell indexDetection box has format [x
,y
,h
,w
,box_score
,face_score
], where:
x
,y
) - raw coordinates of box center to the cellh
,w
- raw height and width of box, apply exponential function and multiply by corresponding anchors to get height and width values relative to cellbox_score
- confidence of detection box in [0,1] rangeface_score
- probability that detected object belongs to face class in [0,1] rangeThe original model is distributed under the following license: