Face and key point detection: yolo5face practice

Github:https://github.com/deepcam-cn/yolov5-face

Reading guide ： It took a day to reproduce the project , Mainly in data set format , The author is actually train2yolo It's amazing that Widerface Training set （12000+ Pictures ） Save one by one , If you don't do this, you will report an error , The reason is that the data cache cannot be read . hamal , liver ！

---

---

One 、 Settings

Step by step

Cloning of warehouse , Get into conda Environmental Science

git clone https://github.com/deepcam-cn/yolov5-face.git

conda activate pytorch-cifar

test model
Download the author's pre trained YOLO5m-face Model , And test the effect of a single picture . function detectface as follows ：

      "args": [
          // // // to run train2yolo:
          // "data/widerface/train",
          // "data/widerface/train/labels"

          // // // to run val2yolo:
          // "data/widerface",
          // "data/widerface/val/labels"

          // // to run detect_face
          // "--img-size", "640",
          // "--weights", "weights/yolov5m-face.pt",


          // to train on widerface
          // "CUDA_VISIBLE_DEVICES"," 0, ",
          "--data", "data/widerface.yaml",
          "--cfg", "models/yolov5s.yaml",
          "--weights", "weights/yolov5s.pt"
      ]

Above is launch.json Configuration parameters for , Corresponding to several tasks . In which operation train2yolo.py, Great slowness ; But I have to do .

Two 、 Training

2.1 Prepare the data ：

Recommend going to the official website http://shuoyang1213.me/WIDERFACE/
download widerface Data sets , Download separately train、val、test Three bags . I refer to RetinaFace The data set downloaded is not comprehensive .

Because the official website did not find the annotation text of face key points , At the beginning git link YOLO5face You can download with keypoint Of annotation file . The whole file structure is as follows ：

(base) [email protected]:NewProjects$ ln -s ~/Datasets/widerface  yolov5-face/data/
//

yolov5-face
│   README.md
│   ...   
│data
| └─widerface
│         └───test
|                 └───images
|                 └───labels
│         └───train
|                 └───images
|                 └───labels
|                      └───0..._1_5.jpg
|                      └───0..._1_5.txt
│         └───val

The catalogue is a little ugly , But for the sake of explanation , It's also hard work .

Download pre trained yolo5 Test model , Because the project is based on yolo The version is older （ newest 6.0 edition ） For a long time , Download link reference ：
https://github.com/ultralytics/yolov5/releases/tag/v4.0

Accurately read the data path , stay widerface.yaml It is amended as follows ：

train: /home/wqt/NewProjects/yolov5-face/data/widerface/train/labels  
val: /home/wqt/NewProjects/yolov5-face/data/widerface/val/labels   

# number of classes
nc: 1

# class names
names: [ 'face']

function train, No mistake , The Buddha bless ！

autoanchor: Analyzing anchors... anchors/target = 4.02, Best Possible Recall (BPR) = 0.9997
Image sizes 800 train, 800 test
Using 4 dataloader workers
Logging results to runs/train/exp12
Starting training for 250 epochs...

     Epoch   gpu_mem       box       obj       cls  landmark     total   targets  img_size
     0/249     4.65G    0.1032   0.03953         0    0.1396    0.2823        25       800: 100%|███████████████████████████████████████████████| 801/801 [03:12<00:00,  4.16it/s]

     Epoch   gpu_mem       box       obj       cls  landmark     total   targets  img_size
     1/249     5.01G   0.08159   0.03674         0   0.02959    0.1479        15       800: 100%|███████████████████████████████████████████████| 801/801 [03:03<00:00,  4.38it/s]

Training was quite fast , One epoch Just a few minutes , common 250epoch.

of YOLO data format

The original widerface It is marked as follows ：

0--Parade/0_Parade_marchingband_1_849.jpg
1
449 330 122 149 0 0 0 0 0 0 

Turn into YOLO Format , as follows ：

0--Parade/0_Parade_marchingband_1_849.jpg
449 330 122 149 488.906 373.643 0.0 542.089 376.442 0.0 515.031 412.83 0.0 485.174 425.893 0.0 538.357 431.491 0.0 0.82

namely ：bbox (xywh) + (xi, yi, flag)

COCO keypoints Turn into YOLO Format (17 Key points of human body ), The marks corresponding to the following pictures are as follows ：

0 0.686445 0.531960 0.082891 0.323967 0.667188 0.399061 1.000000 0.670312 0.396714 2.000000 0.000000 0.000000 0.000000 0.678125 0.394366 2.000000 0.000000 0.000000 0.000000 0.689063 0.415493 2.000000 0.696875 0.415493 2.000000 0.682813 0.469484 2.000000 0.671875 0.483568 2.000000 0.671875 0.516432 2.000000 0.656250 0.504695 2.000000 0.695312 0.530516 2.000000 0.706250 0.523474 2.000000 0.698438 0.610329 2.000000 0.709375 0.603286 2.000000 0.710938 0.680751 2.000000 0.717187 0.671362 2.000000

Trained 12h after , The measured results are as follows ：

  Epoch   gpu_mem       box       obj       cls  landmark     total   targets  img_size
   249/249     5.01G   0.03923   0.02193         0   0.00727   0.06844        18       800: 
               Class      Images     Targets           P           R      [email protected]  [email protected]:.95: 
                 all    3.22e+03    3.96e+04       0.576       0.723       0.715        0.34

That is to say WiderFace We got it [email protected]=71.5%,[email protected]=0.34

Actually measured 1：

Actually measured 2：

Use the zoom effect , There are still some errors in the detection , For example, the points on both sides of the corner of the mouth .