Swap faces in the video or images in realtime using only one photo of both people

That's is course work done by Denis Tsyupa, Vasiliy Kozlov, Vadim Chernishov at HSE University

It needs CUDA Compatible GPU with at least 4GB of RAM to work

Download weights from here and place them into weights directory

https://www.kaggle.com/drjerk/one-shot-deepfake-weights or https://drive.google.com/drive/folders/1Dhudzp0ZvefUWsZv7h6aJk6qCTKnU-nW?usp=sharing

https://www.kaggle.com/drjerk/swap-faces-usage-demo - simple usage demo

python process_image.py - for image processing

usage: process_image.py [-h] --input-image INPUT_IMAGE --output-image OUTPUT_IMAGE --source-image SOURCE_IMAGE [--refl-coef REFL_COEF] [--fallback-point-detector]

One shot face swapper

optional arguments:

-h, --help show this help message and exit

--input-image INPUT_IMAGE

--output-image OUTPUT_IMAGE

--source-image SOURCE_IMAGE

--refl-coef REFL_COEF

--fallback-point-detector

python main.py - for video processing, supports reading from webcam / video file and writing to webcam (using akvcam) / video file / showing in window

python main.py --source-image images/swapme.jpg --target-image images/you.jpg --verbose

or

python main.py --source-image images/swapme.jpg --verbose --disable-faceid

for video processing demo (Q button to quit)

usage: main.py [-h] [--input-video INPUT_VIDEO] [--output-video OUTPUT_VIDEO] [--output-camera OUTPUT_CAMERA] [--camera-width CAMERA_WIDTH] [--camera-height CAMERA_HEIGHT] --source-image SOURCE_IMAGE [--target-image TARGET_IMAGE] [--verbose] [--refl-coef REFL_COEF] [--fallback-point-detector] [--disable-faceid]

One shot face swapper

optional arguments:

-h, --help show this help message and exit

--input-video INPUT_VIDEO

--output-video OUTPUT_VIDEO

--output-camera OUTPUT_CAMERA

--camera-width CAMERA_WIDTH

--camera-height CAMERA_HEIGHT

--source-image SOURCE_IMAGE

--target-image TARGET_IMAGE

--verbose (show simple gui demo window)

--refl-coef REFL_COEF

--fallback-point-detector

--disable-faceid

Some swapping results

Sources

https://github.com/TreB1eN/InsightFace_Pytorch - face comparison model

https://github.com/1adrianb/face-alignment - facial landmarks model used to pseudolabel dataset

https://github.com/webcamoid/akvcam - virtual camera integration

https://www.kaggle.com/yukia18/sub-rals-biggan-with-auxiliary-classifier - good biggan implementation

https://github.com/Vasiliikoz/CourseProject - fast facial landmarks model