face-swap-project

Pre-requisites

• Install required libraries
  $pip install -r requirements.txt

• Download the dlib's facial landmark detection model from the below link and place it under data/ folder
  http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2

Usage

main.py [-h] --detector DETECTOR --image1 IMAGE1 --image2 IMAGE2 --output1 OUTPUT1 --output2 OUTPUT2 --swapper SWAPPER

• Detector: detector mode (0 - dlib detector / 1 - cascade detector)
• Image1: Source Image file path
• Image2: Target Image file path
• Output1: Poisson Blend output file path
• Output2: Simple merge output file path
• Swapper: swapper mode (0-warps the source image to fit target face 1-warps the target image to fit source face)

ex)

Swapper mode 0:
$ python3 src/main.py --detector 0 --image1 image/source/img1.jpg --image2 image/target/img1.jpg --output1 image/result/poisson_blend_target_swap/img1.jpg --output2 image/result/direct_cut_target_swap/img1.jpg --swapper 0

Swapper mode 1:
$ python3 src/main.py --detector 0 --image1 image/source/img1.jpg --image2 image/target/img1.jpg --output1 image/result/poisson_blend_source_swap/img1.jpg --output2 image/result/direct_cut_source_swap/img1.jpg --swapper 1

File Structure

├── data (#pre-trained data files for detection)
├── evaluation (#evaluation code)
├── src (#main code to run face swap)
├── images (#input and output images)
│   ├── source
│   ├── target
│   ├── result
│   │   ├── direct_cut_sorce_warp
│   │   │   ├── img1.jpg
│   │   │   ├── img2.jpg
│   │   │   ├── ..
│   │   ├── direct_cut_target_warp
│   │   │   ├── img1.jpg
│   │   │   ├── img2.jpg
│   │   │   ├── ..
│   │   ├── poisson_blend_sorce_warp
│   │   │   ├── img1.jpg
│   │   │   ├── img2.jpg
│   │   │   ├── ..
│   │   ├── poisson_blend_target_warp
│   │   │   ├── img1.jpg
│   │   │   ├── img2.jpg
│   │   │   ├── ..
├── README.md
├── requirement.txt
├── .DS_Store
└── .gitignore

Evaluation

Source Swap Image Results

Images	Direct Cut SSIM	Direct Cut PSNR	Direct Cut MSE	Poisson Blend SSIM	Poisson Blend PSNR	Poisson Blend MSE
img1.jpg	0.909958	24.159	249.56	0.916568	26.9375	131.622
img2.jpg	0.929526	24.2052	246.923	0.946575	27.9288	104.762
img3.jpg	0.90944	22.5423	362.115	0.91112	23.1357	315.874
img4.jpg	0.925882	22.3367	379.675	0.919438	18.7713	862.887
img5.jpg	0.929575	28.0502	101.872	0.928888	25.9772	164.194
img6.jpg	0.927878	24.984	206.385	0.900615	19.6824	699.58
img7.jpg	0.934676	23.0075	325.335	0.902332	19.035	812.044
img8.jpg	0.94062	27.9016	105.419	0.893054	19.3643	752.745
img9.jpg	0.9439	26.7991	135.885	0.943131	24.4424	233.796
img10.jpg	0.88881	23.1712	313.302	0.86626	17.8595	1064.47
img11.jpg	0.949733	28.5278	91.2651	0.906504	17.1688	1247.97
img12.jpg	0.878535	22.903	333.258	0.855958	16.9716	1305.92
img13.jpg	0.879672	22.4758	367.706	0.88797	25.1339	199.382
img14.jpg	0.920893	24.6128	224.802	0.925137	25.8939	167.374
img15.jpg	0.879356	23.7632	273.374	0.867014	17.4012	1182.92
img16.jpg	0.879322	22.7655	343.979	0.883342	23.5916	284.394
img17.jpg	0.875384	24.3145	240.784	0.853903	15.686	1755.83
img18.jpg	0.876705	20.8263	537.593	0.873504	18.0596	1016.52
Average	0.909993	24.297	268.846	0.898962	21.28	683.46

Target Swap Images

Images	Direct Cut SSIM	Direct Cut PSNR	Direct Cut MSE	Poisson Blend SSIM	Poisson Blend PSNR	Poisson Blend MSE
img1.jpg	0.742468	19.3457	755.978	0.744616	19.8052	680.085
img2.jpg	0.795204	20.7903	542.067	0.800036	21.1549	498.417
img3.jpg	0.677108	18.528	912.59	0.678382	18.6128	894.943
img4.jpg	0.811359	19.4763	733.579	0.8027	20.899	528.667
img5.jpg	0.68118	17.1519	1252.84	0.684977	17.4948	1157.71
img6.jpg	0.801211	19.8956	666.064	0.783414	19.0137	816.033
img7.jpg	0.818234	19.5302	724.534	0.816023	19.7555	687.911
img8.jpg	0.804039	20.606	565.566	0.807582	20.8299	537.142
img9.jpg	0.67203	20.5474	573.248	0.673998	20.4229	589.917
img10.jpg	0.728386	19.0127	816.23	0.728243	19.445	738.884
img11.jpg	0.636317	21.5567	454.366	0.634905	21.6619	443.501
img13.jpg	0.745005	18.6077	896.009	0.755785	19.2468	773.386
img14.jpg	0.778487	23.7072	276.923	0.78105	25.5802	179.913
img15.jpg	0.757574	19.3119	761.887	0.757905	19.658	703.521
img16.jpg	0.774291	19.4979	729.944	0.778078	19.7805	683.957
img17.jpg	0.723573	17.6979	1104.83	0.700135	17.8935	1056.16
img18.jpg	0.755603	17.6214	1124.45	0.762962	18.9528	827.568
Average	0.747181	19.5815	758.3	0.746517	20.0122	693.983

How to run the evaluation of the swapping techniques

1. Go to ./evaluation folder and modify the directory of files you want to evaluate
2. Ensure that the images are as follows source/1.jpg + target/1.jpg = result/1.jpg

source_directory = "./image/source"
target_directory = "./image/target"
direct_cut_directory = "./image/result/direct_cut_source_swap"
poisson_blend_directory = "./image/result/poisson_blend_source_swap"

$ python evaluation/evaluation_script.py