Ageify

• 

  Back-end code

• 

  Demonstration of the video on Zoom

Inspiration

I've recently been bombarded with memes talking about networking, job experience, internships, etc all words that I don't really understand. Instead of actually getting job experience or working, I decided to just make myself look older so I can claim experience.

What it does

It allows a person to appear and sound older using a CycleGAN model that is able to transform a person's face and also utilizes a voice transformer that alters the pitch-shift giving the appearance of an older or younger person. It also gives a person a berad (lol!)

How we built it

I utilized something called CycleGAN and leveraged a pre-trained model to adjust a person's face. This was all coded in Python. CycleGAN framework to learn bidirectional mappings between “young” and “old” face domains without requiring paired training examples

Challenges we ran into

One of the biggest challenges was latency. I initially attempted to utilize The Segment Anything Model (SAM) used in this paper https://arxiv.org/pdf/2102.02754. It essentially a transformer style GAN module that is able to change the person's face accurately both becoming younger and older. The problem is that the pth file was 1.6 gigabites and it's humongous --> meaning its slow. This meant I had to adjust to Fast-AgingGAN, leveraging a pre-trained model from https://github.com/HasnainRaz/Fast-AgingGAN/tree/master.

Accomplishments that we're proud of

One of the biggest accomplishments was getting the model to work in real-time. Even with the changed Fast-AgingGAN there was still latency issues that required lots of work.

What we learned

I learned a lot about CycleGAN and creating generative AI.

What's next for HelpINeedToLookOlderQuickForMyZoomMeeting

Changing the model to SAM or possibly another version.

Built With

• cnn
• faceapp
• fast-aginggan
• generative-adversarial-network
• neural-network
• opencv
• python
• pytorch
• pytorch-lightning
• realtime-face-aging