Simplely implement the paper 'Semantic Image Inpainting with Deep Generative Models'
This code simplely implement the paper Semantic Image Inpainting with Deep Generative Models. The results of the paper have good results of face inpainting.
The method of the paper is divided into two stages,
First, train the DCGAN to get the pretrained model(generator, discriminator).
Second, use the pretrained model of DCGAN from the first stage to train the input(z) of the generator, a little similar with neural style transfer.
======================
tensorflow 1.4.0
python 3.5
pillow
numpy
scipy
======================
In the first stage, we select the CelebA as the dataset of the DCGAN to get the pretrained model, and remain 1000 as test data in the second stage.
| Raw | incompleted | completed |
|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
| Inpainting |
|---|
 |
 |
 |
 |
col 1: raw image, col 2: incompleted image, col 3: generated image by generator, col 4: completed image