search

genforce / interfacegan

[CVPR 2020] Interpreting the Latent Space of GANs for Semantic Face Editing
https://genforce.github.io/interfacegan/
MIT License
1.5k stars 282 forks source link

Results were not satisfying? #79

Open alan-ai-learner opened 3 years ago

alan-ai-learner commented 3 years ago

Hey, @ShenYujun I followed your interfacegan paper like I did everything that we need to do. But the results I'm getting have seemed like a different person, not the same image on which I'm trying to edit. that is the latent vector generated by StyleGAN is not that accurate. or the equivalent latent space I'm creating using the StyleGAN encoder isn't the same as the original image which I'm trying to edit. Then I try to use the StyleGAN2 for encoding the image and its resultant image latent vector almost similar as the original one and far better than StyleGAN encoder. But the problem is I'm not able to use the StyleGAN2 encoded latent code for editing using your code. So is there any way to use StyleGAN2 generated latent vectors for editing using the interfacegan? Or any other solution, I spent lots of resources to came this far to implement interface gan to edit custom images. And one more thing my goal is to create an application same as FaceApp(https://www.faceapp.com/), any way to achieve the same thing as FaceApp. Thanks in advance!

fisakhan commented 3 years ago

@alan-ai-learner can you please provide some suggestion. I also want to do similar task.

alan-ai-learner commented 3 years ago

@alan-ai-learner can you please provide some suggestion. I also want to do similar task.

@fisakhan put your questions here will try to answer them.

fisakhan commented 3 years ago

here? where? My question is that How can I generate multiple different images of the same identity? Different images might have different pose, expression etc. but the identity should remain the same (or look like the same identity). The implementation in this repo either changes the identity or generates an extremely poor quality image.

alan-ai-learner commented 3 years ago

yeah, that's the problem I faced as well, I spend like 4 months and lots of resources on it but in the end, I haven't found the results satisfactory, that's why I raised this issue. Btw they published one more paper called SeFa, check that one maybe that can help.\

On Mon, 5 Jul 2021 at 11:43, fisakhan @.***> wrote:

here? where? My question is that How can I generate multiple different images of the same identity? Different images might have different pose, expression etc. but the identity should remain the same (or look like the same identity). The implementation in this repo either changes the identity or generates an extremely poor quality image.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/genforce/interfacegan/issues/79#issuecomment-873832479, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMZS64BSUC3USU2DFD25ZFTTWFEPFANCNFSM44YTDKYA .

fisakhan commented 3 years ago

Thanks @alan-ai-learner