A lot of people may not find it strange to know PS and use it as a graphics software to be a self of the opposite sex.But isn't it more interesting that we are today, generated with artificial intelligence?

This kind of AI function implementation involving images, friends who understand deep learning will know some.It is often implemented with the very popular GAN generative adversarial network, or DCGAN.And the use of AI to make the opposite sex self mentioned in this article is to use a gender conversion function of PaddleGAN.In other words, it is to use the face attribute editing ability in PaddleGAN to realize gender transformation of any face!

I. Environment preparation

Download PaddleGAN

git clone https://github.com/PaddlePaddle/PaddleGANorgit clone https://gitee.com/PaddlePaddle/PaddleGAN

Install PaddleGAN locally

#Enter PaddleGAN directorycd /home/aistudio/PaddleGAN#install dependenciespip install -r requirements.txt#Install PaddleGAN locallypython setup.py develop

Install dlib

pip install dlib

If the arrangement of dlib fails, please go to the official website to download the corresponding whl version and install it manually.

Second, Extract the Latent Code of the original image

Here we use Pixel2Style2Pixel to extract Latent Code, you only need to change two parameters:

1. input_image: the original image path, that is, the photo path where hidden features need to be extracted

2. output_path: The storage path of the hidden features of the original image, which needs to be used in the attribute editing and generation module later

Put the face photo of the face you want to change gender in the input_image part, please note that it is best to take a selfie and a big head photo, the effect is better without glasses~

Command method:

cd applications/python -u tools/pixel2style2pixel.py  --input_image '/home/aistudio/face test set/yingbb2.png' --output_path '/home/aistudio/psp_output/yingbb2'--model_type ffhq-inversion--seed 233--size 1024--style_dim 512--n_mlp 8--channel_multiplier 2

Programmatically:

import paddlefrom ppgan.apps import Pixel2Style2PixelPredictorpredictor = Pixel2Style2PixelPredictor(output_path="outputs",weight_path=None,model_type="ffhq-inversion",seed=233,size=1024,style_dim=512,n_mlp=8,channel_multiplier=2)predictor.output_path = 'outputs'predictor.run('inputs/src.png')

Three, generate the opposite sex you

Edit the Latent Code in a specific direction, and use StyleGAN V2 to generate a new face from the edited vector.

Command method:

python -u tools/styleganv2editing.py--latent '/home/aistudio/psp_output/11/dst.npy'--output_path '/home/aistudio/final_output/11'--model_type ffhq-config-f--size 1024--style_dim 512--n_mlp 8--channel_multiplier 2--direction_name gender  --direction_offset 5 #Convert positive numbers from female to male, and negative numbers from male to female, 

Programmatically

import paddlefrom ppgan.apps import StyleGANv2EditingPredictorpredictorEditing = StyleGANv2EditingPredictor(output_path="outputs",weight_path=None,model_type="ffhq-config-f",seed=None,size=1024,style_dim=512,n_mlp=8,channel_multiplier=2,direction_path=None)dstNpyPath="outputs/dst.npy"predictorEditing.output_path = 'outputs'direction_offset = 5 #positive numbers are converted from female to male, and negative numbers are converted from male to femalepredictorEditing.run(dstNpyPath, "gender", direction_offset)

Four, the final conversion result

Okay, I'll share it here today. If you have any ideas, please leave a message for discussion.