JointFace

Joint face detection demo application. The original network is from here: https://github.com/kpzhang93/MTCNN_face_detection_alignment Here we are using the v1 model.

How to generate FPGA network configs and weight files

1. Run the conversion tool to generate FPGA network configs and weight files.

$ python ../../../tool/convertor.py 12net_20x15.ini
$ python ../../../tool/convertor.py 12net_28x21.ini
$ python ../../../tool/convertor.py 12net_40x30.ini
$ python ../../../tool/convertor.py 12net_56x42.ini
$ python ../../../tool/convertor.py 12net_80x60.ini
$ python ../../../tool/convertor.py 12net_112x84.ini
$ python ../../../tool/convertor.py 12net_160x120.ini
$ python ../../../tool/convertor.py 12net_224x168.ini
$ python ../../../tool/convertor.py 12net_320x240.ini
$ python ../../../tool/convertor.py 12net_448x336.ini
$ python ../../../tool/convertor.py 12net_640x480.ini
$ python ../../../tool/convertor.py 24net.ini
$ python ../../../tool/convertor.py 48net.ini

NOTE: The original only have 3 networks, but 12Net is run several times with different input size. Because the limitation of the tool, each input size must have their own configs for the FPGA.

2. Copy the generated FPGA network configs to the root folder.

3. Copy the generated weight files to the FPGA SD card, to the bin/folder. The 12Net generated 11 weight files, but they are all the same. Just copy one of them and rename it to 12Net_weights.bin