cv_dnn_face

Sample scripts using OpenCV DNN based face detection and recognition.

Environment

• Ubuntu 20.04.4 LTS
• Python 3.8.10

requirements

• opencv-contrib-python==4.5.4.60 (checked with 4.5.4)
• numpy

model files

Download following onnx model files and locate in model direcory. model/yunet.onnx model/face_recognizer_fast.onnx

install

$ python setup.py install

demo scripts

detector.py

You can show face detection demo.

$ python detector.py -h
usage: detector.py [-h] img_file

face_recognizer

positional arguments:
  img_file    image_file, movie_file or camera_number

optional arguments:
  -h, --help  show this help message and exit

recognizer.py

You can show face recognition demo. This assumes you have face features in aligned_faces directory.

$ python recognizer.py -h
usage: recognizer.py [-h] img_file

face_recognizer

positional arguments:
  img_file    image_file, video_file or camera_number

optional arguments:
  -h, --help  show this help message and exit

How to generate face samples and face dataset in aligned_faces directory.

crop.py

This will crop faces with margin.

$ python crop.py -h
usage: crop.py [-h] [--clockwise] dir

face cropper from images.

positional arguments:
  dir          image source dir

optional arguments:
  -h, --help   show this help message and exit
  --clockwise  rotate image clockwize

crop_from_video.py

This will crop faces with margin from video.

$ python crop_from_video.py  -h
usage: crop_from_video.py [-h] [--dst_dir DST_DIR] [--interval INTERVAL] name

crop face from video

positional arguments:
  name                 video file

optional arguments:
  -h, --help           show this help message and exit
  --dst_dir DST_DIR    dst dir
  --interval INTERVAL  frame interval to process

move_similar_faces.py

You might want to remove too similar faces to reduce dataset.

$ python move_similar_faces.py -h
usage: move_similar_faces.py [-h] [--th TH] [-r] src_dir dst_dir

face cropper from images.

positional arguments:
  src_dir     image source dir
  dst_dir     destination dir

optional arguments:
  -h, --help  show this help message and exit
  --th TH     if face_distance is smaller than threshold, skips
  -r          recursive file search

generate_aligned_faces.py

This will generate feature data file (.npy), and face image without margin (.jpg)

$ python generate_aligned_faces.py -h
usage: generate aligned face images from an image [-h] [--recursive] image

positional arguments:
  image        input image file path (./image.jpg)

optional arguments:
  -h, --help   show this help message and exit
  --recursive

Note: recognizer.py assumes feature file name with label. biden.npy means label is biden.

Download

---

• yunet.onnx
• face_recognizer_fast.onnx

SeeAlso

https://github.com/opencv/opencv/blob/4.x/samples/dnn/README.md https://github.com/opencv/opencv/blob/4.x/samples/dnn/face_detect.py

This was inspired by following URL https://qiita.com/UnaNancyOwen/items/f3db189760037ec680f3