Real time face detection using Yolov5 deployed on android

You can get dataset from kaggle competition Face Mask Detection. This dataset contains Images and Annotation folder, which we convert to YOLOV5 required form.

The format of annotation required is:

<class> <x> <y> <width> <height>
ex 0 2.0 12.0 240 320

We convert xml files to text files. Then we can split the dataset into Train, Test, Val

Create Folders as shown below

Images
├── Train
├── Test
└──  Val

Annotations
├── Train
├── Test
└──  Val

Then change the directories path in pre proess data.py and run the file.

Lets start by cloning yolov5 and installing requirements

$git clone https://github.com/ultralytics/yolov5
$pip install -r requirements.txt

Create a data.yaml file and change the path, nc (number of classes) and names

Organize Directories as shown below

Project
├── yolov5
├── Train
    ├── images
    ├── labels
├── Test
    ├── images
    ├── labels
├── Val
    ├── images
    ├── labels
├── mask_yolovs.yaml
└── data.yaml 

Run the command

# Train yolov5 on our dataset
$python yolov5/train.py --batch 16 --epoch 5 --data data.yaml --cfg mask_yolov5s.yaml --weights yolov5s.pt --name masks_yolov5_result

Test the model on single image

$cd yolov5
$python detect.py --source test.png --weights runs/train/mask_yolov5_result/weights/best.pt

Export ptl file for android app run export.py

Before running the code edit export.py line 80 '.torchscript.pt' to '.torchscript.ptl

$python export.py --weights runs/train/mask_yolov5_result/weights/best.pt --batch 1

Create a classes.txt file and write down the class name in newline

Go to this directory

"\ObjectDetection\app\src\main\assets"

Add your .ptl and txt files and build the app in android studio