Facial Expression Editing Task

[zhou-frank]

Hello, I have read your paper and run the model that you released.But it is only generating face with new identity.How to generate the input identity face image? Where do I need to change the code?

[wxh001qq]

@zhou-frank Hi，have you run this code without bug? I find the dataset name is 025.jpeg, but the raw data is FirstF_000.jpeg, should i raname the raw dataset ,than is run correct？

[zhou-frank]

I didn’t encounter this situation

[Sampson-Lee]

Hi, @zhou-frank

I encountered this phenomenon, where the generated image can not preserve the identity. I think it is a shortcoming of ExprGAN.

Here is the original images/generated images in OULU:

Here is the original images/generated images in CelebA:

[yyssmm]

I happen to have an error: No such file or directory: '../data/OULU/P008/Surprise/028.jpeg'
Where can I get the test dataset ? Thank you!

[voa18105]

@Sampson-Lee did you overcome this problem? I have a same one now, it generates faces itself and changes their emotions

[farzinh]

@wxh001qq
I developed a simple script that can do this for you, you can access it from here:
https://gist.github.com/farzinh/f2d4040bd4c2e02cc3bff316714eaa4b
Copy and paste it in a txt file and save it as whatEverYouLike.py file for this you need python3.
go to terminal and type:
python3 whatEverYouLike.py the path of oulu_anno.pickle file

[Sampson-Lee]

@voa18105 I have not solved it

[SophieChang66]

Hi, @zhou-frank

I encountered this phenomenon, where the generated image can not preserve the identity. I think it is a shortcoming of ExprGAN.

Here is the original images/generated images in OULU:

Here is the original images/generated images in CelebA:

hello, can you share me that how to run the code sucessfully or how to repair bug of the code? I got many bugs when i ran it. Then I find the Flags.is_train is not useful, after repair the bug

[jcjavaismgood]

I happen to have an error: No such file or directory: '../data/OULU/P008/Surprise/028.jpeg'
Where can I get the test dataset ? Thank you!

you pick c dataset，it should be B

[pongkun]

Hi, @zhou-frank

I encountered this phenomenon, where the generated image can not preserve the identity. I think it is a shortcoming of ExprGAN.

Here is the original images/generated images in OULU:

Here is the original images/generated images in CelebA:

Hello, I tried to reproduce the model using pytorch, but the results were fuzzy. Does the first stage require training on the FAR dataset?