Skip to main content

Multi-task Cascaded Convolutional Neural Networks for Face Detection, based on TensorFlow

Project description


Implementation of the MTCNN face detector for TensorFlow. It is written from scratch, using as a reference the implementation of MTCNN from David Sandberg (FaceNet’s MTCNN). It is based on the paper Zhang et al. (2016) [ZHANG2016].


Currently it is only supported python3 onwards. It can be installed with pip:

$ pip3 install mtcnn


The following example illustrates the ease of use of this package:

>>> from mtcnn.mtcnn import MTCNN
>>> import cv2
>>> img = cv2.imread("ivan.jpg")
>>> detector = MTCNN()
>>> print(detector.detect_faces(img))
[{'box': [277, 90, 48, 63], 'keypoints': {'nose': (303, 131), 'mouth_right': (313, 141), 'right_eye': (314, 114), 'left_eye': (291, 117), 'mouth_left': (296, 143)}, 'confidence': 0.99851983785629272}]

The detector returns a list of JSON objects. Each JSON object contains three main keys: ‘box’, ‘confidence’ and ‘keypoints’:

  • The bounding box is formatted as [x, y, width, height] under the key ‘box’.

  • The confidence is the probability for a bounding box to be matching a face.

  • The keypoints are formatted into a JSON object with the keys ‘left_eye’, ‘right_eye’, ‘nose’, ‘mouth_left’, ‘mouth_right’. Each keypoint is identified by a pixel position (x, y).

A good example of usage can be found in the file “” located in the root of this repository.


By default the MTCNN bundles a face detection weights model.

The model is adapted from the Facenet’s MTCNN implementation, merged in a single file located inside the folder ‘data’ relative to the module’s path. It can be overriden by injecting it into the MTCNN() constructor during instantiation.

The model must be numpy-based containing the 3 main keys “pnet”, “rnet” and “onet”, having each of them the weights of each of the layers of the network.



Zhang, K., Zhang, Z., Li, Z., and Qiao, Y. (2016). Joint face detection and alignment using multitask cascaded convolutional networks. IEEE Signal Processing Letters, 23(10):1499–1503.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

mtcnn-0.0.2.tar.gz (2.3 MB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page