Home

Awesome

In-Domain GAN Inversion for Real Image Editing

Based on Seonghyeon Kim's Pytorch Implementation of StyleGAN2

[Paper] [Official Code] [StyleGAN2 Pytorch]

Train Encoder

python train_encoder.py

0k iter
<img src="./imgs/0k.png" width="960">

1M iter
<img src="./imgs/1M.png" width="960">
[encoder checkpoint] [generator checkpoint]

Note: The encoder architecture and loss weights are different from the original implemetation.

Interpolation

interpolate.ipynb

Domain-Guided Encoder (Initial projection)
<img src="./imgs/interpolation_domain_guided_encoder.png" width="360">

In-Domain Inversion (500 steps)
<img src="./imgs/interpolation_idinversion_500steps.png" width="360">

Inperpolation Result
<img src="./imgs/interpolation_results.png" width="960">

Encoder + Model Interpolation

[Paper] [Naver Webtoon Model]

<img src="./imgs/face2webtoon.png" width="960">