shaoanlu

I've modified the original auto-encoder into a GAN. https://github.com/shaoanlu/faceswap-GAN btw, I tried random channel_shift on warped images and got worse results (input/output color inconsistency).

There are 2 conceivable problems in my mind about merging the GAN model into the main repo: 1. Code I wrote in the training part is somehow tensowflow-ish that different...

Another relevant one: GeneGAN: Learning Object Transfiguration and Attribute Subspace from Unpaired Data - https://arxiv.org/abs/1705.04932 - https://www.youtube.com/watch?v=KUsVH7P-5vE - https://github.com/Prinsphield/GeneGAN - From the abstract: "Overall, our deterministic generative model learns disentangled...

How about putting multiple person you want to transform into src folder? If the encoder learns good embedding then perhaps this will work.

Great write up with cool results, I like the gifs! Just want to add some comments about the images shown in readme. Siraj in his video did not correctly interpret...

Probably under MIT. But I have some concern about the portrait privacy of the demo gifs in README. Is there anyway I can exclude README file from license?

Most of my experiments were ran on two dataset: 1. **Hinako Sano/Emi Takei images as shown in readme**. This dataset is rather clean and well conditioned (contain little duplication). My...

I ran my experiments on google colab. V2 model took

Adjust `min_face_scale` to proper ratio. Be careful of the output size that if the cropped faces are much smaller than 64x64, it can be detrimental to your model.

Not checked yet but v2 step 11 should be compatible with v2.1.