Spaces:
Running
Interest in collaboration
Hello there,
we are a young non-profit art ensemble from Berlin and one of our videos will include two people turning old over the course of 10s. I myself am a computer scientist, but I am not entirely sure how to set this up on my local machine. Also I am slightly confused because I found the repo below on github. Did you build on top of that? Or do you just happen to have quite some similarities? Will we need the repo in combination with your app or will the app work as a standalone? And what license do you grant?
Looking forward to getting in touch with you!
https://github.com/ry-lu/pytorch-face-reaging-network?tab=readme-ov-file
Hi, the GitHub is mine and the license is MIT, fee free to use my code whenever for whatever. Although not required, it would be nice if you could star/reference the work if used!
The files in this hugging face space contain a pretrained model and can be copied to local to work as a standalone app. I believe you just need to install the requirements.txt and run the app.py.
The GitHub repository gives the model training code if you want to train a similar model yourself but is not required to run this app. I hope the final project turns out well!
It's so great to get a response and so quickly. Thanks a lot! And yes, we will definitely credit your work in a suitable way.
I actually got it running and am currently working on making it run directly on video. If you'll allow a few more questions:
- Do you have a theory why the images turn out a bit blurry? I only see a note about that in your README.
- Are you in contact with the authors of the paper? We're a mostly educational, but since we want to submit our work at festivals and competitions, I'm still trying to get an explicit agreement to that usage of Disney IP which I am not even sure if it is patented or pending a patent.
- Is there also some code for generating the dataset? We might need higher resolution so I've been thinking about retraining at 1080x1080.
Check us out on insta, if you're interested: lean_art_official
Oh, I tried applying the model to videos too, but you may encounter flickering. I think they can be removed through post processing though unsure of the specifics, good luck!
For the questions:
A possible explanation is the blur layers inside the model , but it could also be the model has not been trained enough. The pre trained model provided was made with limited resources.
I had no affiliation/contact with Disney, and the code was produced using their idea in the paper and my own technical judgment. I am unfamiliar with how copyright laws work but I will just say AI copyright is weird.
There was, but I lost it 😅. I think getting higher resolution images should be pretty straightforward in theory: generate high res synthetic images and apply SAM. Upscaling the original dataset could work too.