128 points by marcelohuber2 5 months ago flag hide 17 comments
johnsmith 5 months ago next
This is really cool! I've never seen real-time facial reenactment before.
jane1987 5 months ago next
I agree, it's really amazing! I'm curious about how it works under the hood.
jane1987 5 months ago next
can you share any resources on how can we implement the GAN to do similar project?
user456 5 months ago next
Sure, I'd be happy to help! I'd recommend checking out the Google's deep learning library, TensorFlow, they have a lot of resources and tutorials about GANs.
user123 5 months ago prev next
It looks like this project uses deep learning, that's pretty impressive.
user456 5 months ago next
The project uses a Generative Adversarial Network (GAN) to create the reenactment. It's a common technique in deep learning.
jimbo98 5 months ago prev next
It's crazy to think about the potential applications for this technology. It could be huge in fields like virtual reality and gaming.
lucypop 5 months ago prev next
Definitely! I'm excited to see where this goes. Great job!
sally9 5 months ago prev next
I'm curious about the performance, what kind of hardware is required to run it in real-time?
johnsmith 5 months ago next
It's quite resource-intensive, I'd recommend a GPU with at least 8GB of VRAM for real-time performance. But I am planning to optimize it further
geekydude 5 months ago prev next
How did you deal with the problem of mapping the movements of one face to another? That's always been a big challenge in facial animation.
johnsmith 5 months ago next
We used a technique called
user123 5 months ago next
Yeah, I know that technique, it's a mouth-corner local affine transformations. That's smart! I'm impressed with the technical details you implemented
computerlove 5 months ago prev next
This is really cool technology, but I'm a bit concerned about the ethical implications of using it for fake videos. Have you thought about how to prevent misuse?
johnsmith 5 months ago next
That's a valid concern, and one that we've considered. The model is not intended for any kind of misuse. We're currently working on watermarking the videos generated to prevent misuse
techieguy34 5 months ago prev next
I think this kind of technology has a lot of potential for animations and video games! Great job! Do you have any plans for enhancing the fidelity of lip-syncing and facial expressions ?
johnsmith 5 months ago next
Yes! that's the plan, for the next iteration, we are focusing on improving the lip-syncing and the facial expressions, we are using additional datasets and fine-tuning the network