tslgithub / facedetector

a demo to detect face

RetinaFace in PyTorch

support

run on ubuntu20, ubuntu18, python3

Installation

sudo apt-get install python3-pip3
pip3 install -r requirements.txt

download weight

https://drive.usercontent.google.com/download?id=14KX6VqF69MdSPk3Tr9PlDYbq7ArpdNUW&export=download&authuser=0&confirm=t&uuid=1c047ae2-05e4-4f33-b66e-e0d73a711f5b&at=APZUnTWD5sWebJKA6TbMpsDvAHFJ:1700533565125

• save the weight in folder named weigths

run script

python3 script.py  images/
python3 script.py  images/1.jpg

• use defaults vis_thres is 0.6
• result will be saved in folder named result

evaluation

1. download the val images in https://drive.google.com/file/d/1GUCogbp16PMGa39thoMMeWxp7Rp5oM8Q/view

2. unzip the WIDER_val.zip

3. download the val annotation label in http://shuoyang1213.me/WIDERFACE/support/bbx_annotation/wider_face_split.zip

4. unzip wider_face_split.zip in folder of WIDER_val, and rename wider_face_split as annotation

python3 metric.py         # use default iou_thres=0.5
python3 metric.py  0.7    # use  iou_thres=0.7

• 0.7 means iou_thres=0.7, or defaults is 0.5
• result will be saved in folder named result

unit test

1. download test dataset in https://drive.google.com/file/d/1HIfDbVEWKmsYKJZm4lchTBDLW5N7dY5T/view
2. unzip the test dataset
3. run test:

python3 script.py WIDER_test/images/0--Parade
python3 script.py WIDER_test/images/1--Handshaking
...
...
...
python3 script.py WIDER_test/images/61--Street_Battle