hwenjun18 / faceai-master

FaceAI

A python library built to realize face detection with Deep Learning using simple and few lines of code.

---

Now FaceAI is an original version and only support MTCCN to detect face. Eventually, FaceAI will offer others various kind of applications about face.

Update Record

• V0.3.0 : Add 3D face restruction module.
• V0.2.0 : Add download pre-trained automatically from google drive.
• V0.1.0 : Add landmarks detection module.
• V0.0.1 : Initialize the file structure of FaceAI and add MTCCN model.

Dependencies

• Tensorflow 1.10.0 (and later versions)

• Numpy 1.13.1 (and later versions)

• OpenCV

• Matplotlib

• easydict

• PIL

• scipy

Demo

• Face Detection input image:

  

  output image:

  

  code:

  from faceai.Detection import *
  import os
  import cv2
  
  def main():
      execution_path = os.getcwd()
      model_path = ['MTCNN_model/PNet_landmark/PNet-30', 'MTCNN_model/RNet_landmark/RNet-22',
              'MTCNN_model/ONet_landmark/ONet-22']
      input_path = os.path.join(execution_path, "image.jpg")
      output_path = os.path.join(execution_path, "imagenew.jpg")
      # image_arr = cv2.imread(input_path)
      # image_arr = cv2.cvtColor(image_arr, cv2.COLOR_BGR2RGB)
      facedetector = FacesDetection()
      facedetector.setModelTypeAsMTCNN()
      facedetector.setModelPath(model_path)
      facedetector.loadModel(min_face_size=24)
  
      img,dets = facedetector.detectFacesFromImage(input_image=input_path,box_mark=True)
      cv2.imwrite("imagenew.jpg", img)
      print('the number of faces: {:0>3d}'.format((len(dets))))
  
  if __name__ == '__main__':
      main()
  

• Face Landmarks Detection input image:

  

  output image:

  

  code:

  from faceai.Detection import *
  from faceai.Alignment import LandmarksDetection
  import os
  import cv2
  
  def main():
      execution_path = os.getcwd()
      model_path = ['MTCNN_model/PNet_landmark/PNet-30', 'MTCNN_model/RNet_landmark/RNet-22',
                  'MTCNN_model/ONet_landmark/ONet-22']
      input_path = os.path.join(execution_path, "land.jpg")
      output_path = os.path.join(execution_path, "landnew.jpg")
  
      facedetector = FacesDetection()
      facedetector.setModelTypeAsMTCNN()
      facedetector.setModelPath(model_path)
      facedetector.loadModel(min_face_size=48)
  
      img,infs = facedetector.detectFacesFromImage(input_image=input_path,box_mark=False)
      dets = []
      for inf in infs:
          dets.append(inf["detection_details"])
      print('the number of faces: {:0>3d}'.format((len(infs))))
  
      landsdetector = LandmarksDetection()
      landsdetector.setModelTypeAsDAN()
      landsdetector.setModelPath("./Model/Model")
      landsdetector.loadModel()
  
      img,lands = landsdetector.detectLandmarksFromImage(img,dets,points_mark = True)
      cv2.imshow("t",img)
      cv2.imwrite("landnew.jpg", img)
  
  if __name__ == '__main__':
      main()
  

• 3D Face Restruction

  input image:

  

  output image: dense image pose image depth image

  code:

  from faceai.Detection import FacesDetection
  from faceai.Alignment import LandmarksDetection
  from faceai.ThrDFace import ThreeDimRestructure
  import os
  import cv2
  
  def main():
    execution_path = os.getcwd()
    input_path = os.path.join(execution_path, "3d.jpg")
    output_path = os.path.join(execution_path, "landnew.jpg")
  
    facedetector = FacesDetection()
    facedetector.setModelTypeAsMTCNN()
    facedetector.loadModel(detection_speed='fast',min_face_size=12)
  
    img,infs = facedetector.detectFacesFromImage(input_image=input_path,box_mark=False)
    dets = []
    for inf in infs:
        dets.append(inf["detection_details"])
    print('the number of faces: {:0>3d}'.format((len(infs))))
  
    landsdetector = ThreeDimRestructure()
    landsdetector.setModelTypeAsPRNet()
    landsdetector.loadModel()
  
    img_3d = landsdetector.restructure3DFaceFromImage(img,dets=dets,depth=True,pose=True)
    for img_ in img_3d:
        cv2.imshow("vertices",img_['img_show']['vertices'])
        cv2.imshow("pose", img_['img_show']['pose'])
        cv2.imshow("depth", img_['img_show']['depth'])
        cv2.imwrite("3d_vertices.jpg", img_['img_show']['vertices'])
        cv2.imwrite("3d_pose.jpg", img_['img_show']['pose'])
        cv2.imwrite("3d_depth.jpg", img_['img_show']['depth'])
  
  if __name__ == '__main__':
    main()
  

Installation

To install ImageAI, run the python installation instruction below in the command line:

pip install faceai-0.1.2-py2.py3-none-any.whl

Pre-trained model

if you can not download model ,please download by hand,and code will remind you where you should uncompress these file.

*mtcnn pre-trained model will download at C:/Users/Administrator/.faceai/mtcnn

• DAN

• MTCNN

References

• Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks

• Deep Alignment Network: A convolutional neural network for robust face alignment

• Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network [code]