K210 YOLO_V2 FACE DETECTION

My multi-class detection of Yolo V2 based on mobilenet has been successful, So I released this one-class detection code based on Yolo V2. You can use my code to train face detection model or other model.

FDDB FACE

Download dataset

wget http://tamaraberg.com/faceDataset/originalPics.tar.gz
mkdir FDDB
tar -zxvf originalPics.tar.gz -C FDDB
wget http://vis-www.cs.umass.edu/fddb/FDDB-folds.tgz
tar -zxvf FDDB-folds.tgz -C FDDB

Make DataSet List

Please modify the make_fddb_list.py line 7 fddb_path = '/media/zqh/Datas/DataSet/FDDB' to your dataset path, Then:

python3 make_fddb_list.py

Make Anchor List

First extract all annotations.

python3 get_all_annotations.py data/fddb_ann.list

It will save all annotations to tmp/all.txt, Then use kmeans find the anchors:

python3 make_anchor_list.py tmp/all.txt data/fddb_anchors.list

NOTE：

output file name should be ${datasetname}_anchors.list
if use other dataset, should use python3 make_anchor_list.py tmp/all.txt data/fddb_anchors.list --is_random=True
when --is_random=True, may be unsuccessful , just repeat it several times.

Now generate the data/fddb_anchors.list

Train Faces detection

make train MODEL=mobile_yolo MAXEP=20 ILR=0.001 DATASET=fddb CLSNUM=1 IAA=False CLASSIFIER=True BATCH=32

we can continue training

make train MODEL=mobile_yolo MAXEP=40 ILR=0.0005 DATASET=fddb CLSNUM=1 IAA=True CLASSIFIER=False BATCH=32 CKPT=log/xxxxx

and you can use tensorboard --logdir log to look plot

Freeze Graph

make freeze MODEL=mobile_yolo CKPT=log/xxxxxxx CLSNUM=1 DATASET=fddb

now we have Freeze_save.pb

Inferernce

make inference CLSNUM=1 IMG=data/timg.jpg DATASET=fddb

Convert PB to .C Code

make kmodel_convert

Compile C Code

cp xxxxx/kendryte-model-compiler/build/gencode_output.c kpu_yolo_v2

If you modify the anchor value, you have to modify the kpu_yolo_v2/main.c line 47,then you can use the kendryte model compiler to compile. and use kendryte standalone sdk version: v0.5.4

Download .bin

My KD233 board was damaged by squirrels :(

So now I am using the Maix Go, Then use the new Kflash

liciflash kpu_yolo_v2.bin -B goE -p /dev/ttyUSB1 -b 2000000

NOTE liciflash is alias..

Final

VOC

I won't open source this part, If you want to trian Multi class object detection, you must to modify the some code.

Make DataSet List

first use yolo scripts:

wget https://pjreddie.com/media/files/VOCtrainval_11-May-2012.tar
wget https://pjreddie.com/media/files/VOCtrainval_06-Nov-2007.tar
wget https://pjreddie.com/media/files/VOCtest_06-Nov-2007.tar
tar xf VOCtrainval_11-May-2012.tar
tar xf VOCtrainval_06-Nov-2007.tar
tar xf VOCtest_06-Nov-2007.tar
wget https://pjreddie.com/media/files/voc_label.py
python3 voc_label.py
cat 2007_train.txt 2007_val.txt 2012_*.txt > train.txt

now you have train.txt, then :

cd xxxxxxxxx/k210-yolo-v2/
python3 make_voc_list.py xxxxxxxxx/train.txt

Train

make train MODEL=mobile_yolo MAXEP=50 ILR=0.0005 DATASET=voc CLSNUM=20 IAA=False CLASSIFIER=True BATCH=64

habibzadeh / k210_yolo_v2_face Goto Github PK

k210_yolo_v2_face's Introduction

K210 YOLO_V2 FACE DETECTION

FDDB FACE

Download dataset

Make DataSet List

Make Anchor List

Train Faces detection

Freeze Graph

Inferernce

Convert PB to .C Code

Compile C Code

Download .bin

Final

VOC

Make DataSet List

Train

Final

k210_yolo_v2_face's People

Contributors

Watchers

Recommend Projects

Recommend Topics

Recommend Org