README
1 Getting Started
1.1 face_recognition
download from dilb :whms
pip install dlib-19.17.99-cp37-cp37m-win_amd64.whl
pip install face_recognition
1.2 skimage
conda install scikit-image
1.3 tqdm
pip install tqdm
1.4 scipy
download numpy+mkl and scipy from [here](Python Extension Packages for Windows - Christoph Gohlke (uci.edu)).
pip install numpy-1.21.5+mkl-cp37-cp37m-win_amd64.whl
pip install scipy-1.7.3-cp37-cp37m-win_amd64.whl
1.5 pytorch
Just follow the steps on the official website to install.
2 Usage
Place the video files in the video folder and write the names of the videos to be processed in the List_of_testing_videos.txt
file.
python face_cut.py -r 512 -t 500 -m 0 -i 5 -s 60 # -b 10
-r
:Resolution of the captured face
-t
:The range of tolerable original face resolution(faces with resolution greater than r-t will be captured and resize to r)
-m
:Crop and alignment,0-VGGface,1-FFHQ
-i
:Initial face capture interval
-s
:Number of skipped frames when no face is detected
-b
:Threshold value for out-of-focus blur detection
The two parameters Max_int = 40 and Min_int = 5 in the file refer to the maximum and minimum values of the dynamic acquisition interval. Increasing Min_int and Max_int decreases the number of face acquisitions and increases the difference between face images.