cardinalblue/mtcnn-pytorch

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

This project is belongs to the original creator (https://github.com/khrlimam/mtcnn-pytorch). I just fixed some problems that may have been caused by newer version of PyTorch.

How to install

Install the package with pip:

https

pip install git+https://github.com/cardinalblue/mtcnn-pytorch.git

ssh

pip install git+ssh://git@github.com/cardinalblue/mtcnn-pytorch.git

How to use

from torch_mtcnn import detect_faces
from PIL import Image

image = Image.open('image.jpg')
bounding_boxes, landmarks = detect_faces(image)

You can use the bounding_boxes to crop the image (assuming there is only 1 face):

bounding_boxes = list(map(int, bounding_boxes[0]))
img_1 = cv2.imread(path)
img_1[ bounding_boxes[1] : bounding_boxes[3], bounding_boxes[0] : bounding_boxes[2]]

Autocrop

I have included a utility function get_faces. This function gets the image path and returns all of the faces in the image:

from torch_mtcnn import get_faces

faces = get_faces('img.jpg')

Requirements

Please see the requirements.txt

Credit

The original project belongs to https://github.com/khrlimam/mtcnn-pytorch.