face-replace

by marsbroshok

marsbroshok /face-replace

Snapchat -like replace face in the video by a face mask

131 Stars 42 Forks Last release: Not found 6 Commits 0 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

Face Replacement Filter in Video

This is a proof-of-concept Python script inspired by MSQRD and Snapchat 3D Face Placement filters.

To play with the script you should have a still image with sample face and video file with someone's else face in it (or you can use your web camera with your own face;) ).

You can find sample still image and video in the "demo" directory.

Basic Idea

  1. Detected where is a face in the still image
  2. Extract face features' landmarks (like eyes, nose and mouth coordinates)
  3. Detect a face in the video frame and extract its landmarks too
  4. Align landmarks coordinates from steps 2 and 3
  5. Use results of affine transform to warp still-image face around video frame's face

Final result could be saved to the video file or just shown on the screen.

Example

Original video, no face replaced:

Final video, Arni's face is in place of mine:

Technical Notes

  • All image processing is done on the greyscale frames. If input video is in color, it will be converted to greyscale.
  • Web cam and Video in-out were tested on OS X thus video saved with QuickTime codec.
  • To make processing faster all input frames are down-sampled to the size 420x240 px.

Requirements

This script was tested in the following environment:

How To

Before you begin, download and extract Dlib face shape predictor ('shapepredictor68facelandmarks.dat'). Put it to ./models/shapepredictor68facelandmarks.dat

Usage:

python ./change-face-in-video.py [-h] STILLFACE VIDEOIN VIDEOOUT

where arguments are:

  STILLFACE   the full path to jpg file with face
  VIDEOIN     the full path to input video file where face will be changed. If
              "0" is provided then the web cam will be used
  VIDEOOUT    the full path to output video file with the new face. If "0" is
              provided then process video will be shown on the screen, but not
              saved.

For example this command will created a new video

demo_arni.mov
with face replacement from image
arni.jpg
:

python ./change-face-in-video.py ./demo/arni.jpg ./demo/demo_orig.mov ./demo/demo_arni.mov

This command will use image

arni.jpg
again, but for the video stream from web camera:

python ./change-face-in-video.py ./demo/arni.jpg 0 ./demo/demo_arni.mov

Credits

One Millisecond Face Alignment with an Ensemble of Regression Trees by Vahid Kazemi and Josephine Sullivan, CVPR 2014


Alexander Usoltsev, 2016

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.