Image2StyleGAN implementation

Pipeline

1. Aligned Image
2. Preprocess
3. Find latent vector with encoder
4. Find exact latent by optimize it using gradient descent

Encoder

Encoder extract features via modified Resnet-50
Loss was only MSE, ppl need to be added

0. Column

1. Image

2. Aligned

3. Only Gradient descent result (from 0-vector)

4. Encoder latent

5. Encoder + Gradient descent

With latent vectors you can make a lot of transformation, look at my other repo!

https://github.com/DFGANDP/CelebA-VAE-feature-extractor

References:

https://arxiv.org/abs/1904.03189

http://www.seeprettyface.com/research_notes.html

https://github.com/zaidbhat1234/Image2StyleGAN

https://github.com/AmarSaini/Epoching_StyleGan2_Setup