Skip to content

shoutOutYangJie/Face_Recognition_using_pytorch

Repository files navigation

Face_Recognition_using_pytorch

Using MTCNN and MobileFaceNet on Face Recognition

This repo is easy to use and convenient to adapt to your own project. You just need one GPU supporting CUDA and do some simple installations, and then you can easily run it successfully. MTCNN pretrained and mobileFaceNet pretrained model have been available in this project. As detection of face, MTCNN can be replaced with any faster method. General performance is real-time.

image

requirements

PyTorch1.0

torchvision

opencv-python

PIL

numpy

notes that make sure your computer has one camera, because code uses cv2.videoCapture(0) to get image.

how to use

please going to root dir.

firstly for getting your name, run

python get_save_features.py --name {$YOUR_NAME$}

notes that when cvWindow arises, please left click it to make it being current window, and press 'c' to crop your face, and press 'q' to quit program.

And then you will find your feature vector under the "users/{$YOUR_NAME$}" dir.

secondly, run

python verification_multi_face.py

acknowledgement

MTCNN[1] comes from TropComplique, that is a great work containing pretrained model.

mobileFaceNet[2] derives from Xiaoccer. I followed this project to train the model on CASIA dataset, but only can get 97.3% on the lfw dataset.

reference

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

[2] MobileFaceNets: Efficient CNNs for Accurate Real-time Face Verification on Mobile Devices

About

Using MTCNN and MobileFaceNet on Face Recognition

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published