Awesome

the OpenVINO implemententation of LFFD

paper:LFFD: A Light and Fast Face Detector for Edge Devices

official github: LFFD

ncnn implementation is here

MNN implementation is here

Prerequirements

openvino's version: openvino_2019.1.148, I used opencv3.4.3

Please refer the official OPenVINO‘s DOC to install openvino. In that documentation, you will find how to convert the official mxnet model to openvino. And,before you convert the mxnet model ,you need to modify the symbol.json as follows:

• First ,follow the author's original github to build the devolopment environment.
• Modify symbol_10_320_20L_5scales_v2.py (your_path/A-Light-and-Fast-Face-Detector-for-Edge-Devices\face_detection\symbol_farm) in function loss_branch,Note out（注释掉） the line 57(predict_score = mxnet.symbol.slice_axis(predict_score, axis=1, begin=0, end=1) in function get_net_symbol, Note out（注释掉）the line 99(data = (data - 127.5) / 127.5,preprocess).
• Next,in this path , by doing "python symbol_10_320_20L_5scales_v2.py ",generate the symbol.json. symbol_10_560_25L_8scales_v1.py do the same thing .

Inference time on CPU i7 7700

The time is average time with 100 loops. set the mode CPU. When setting it GPU, it uses the intel graphic gpu

• v2

• v1