The inference result is not good

[yerang823]

The inference result is not good, is this a model that only works well for trained faces?

Result source, target, result

Furthermore, The results you posted are also different from the result image of the paper. Result source, target, my result

Here is your result on paper

Please provide detailed answers on the above results.

[e4s2022]

Hi, thanks for your posts.

1. The model was trained on aligned faces, did you align you provided example first?

2. The results differ a bit on the bang near the left eyebrow, which is caused by the different operating orders during the mask composition process. You can try to set hair_first = False here.

Hope this can answer your question.

[yerang823]

Thank you for your detailed answer.

The problem was that i used not aligned images. but attribute informations of images are different from target image. is there any solutions?

Thank you, in advance.

[e4s2022]

Did you refer to the slightly different expression? For this, I recommend trying some recent re-enactment work, e.g., MegaPortraits: One-shot Megapixel Neural Head Avatars,

I will update this repo if I find some better pre-trained models. BTW, any contribution is welcome from the community.

[yerang823]

Thank you for your reference link and answer. And I will wait for your further updates. Have a nice day !

[anleleaaa]

Hello, I want to ask you, how long does it take to change the big concept of the face after loading the initial model, can it be done in real time?

[BlueWinters]

Thank you for your detailed answer.

The problem was that i used not aligned images. but attribute informations of images are different from target image. is there any solutions?

Thank you, in advance.

is there any code to show the alignment process?

[yc-cui]

Thank you for your reference link and answer. And I will wait for your further updates. Have a nice day !

Would you be so kind as to update us on any recent progress?