[Open-to-the-community] HugGAN sprint

HugGAN sprint :art: :artist: :paintbrush:


Hi all,

Happy to invite you to HugGAN - a community event in which we’ll train and showcase generative adversarial networks (GANs for short), taking place from April 4th to April 15th. With compute provided by Paperspace (read: GPUs), you can train a GAN on your own data, get to know how they work through several talks by famous generative AI artists, and leverage the :hugs: ecosystem (using tools such as :hugs: Datasets, :hugs: Accelerate and :hugs: Spaces).

:point_right: We’ll be organizing this event on GitHub.

What it is about :question:

The goal of the event is to make people more familiar with GANs, a type of generative models introduced by Goodfellow et al. in 2014, and showcase them to the community. We hope to add many image datasets and GAN-based models to the :hugs: Hub, and showcase them as a :hugs: Space.

The main components of the HugGAN sprint consist of:

  • sharing image datasets, useful for training generative models, with the community using the brand-new (and awesome) ImageFolder and push_to_hub features. There are already 25+ datasets contributed, including this one :rocket:
  • training GANs using example scripts provided by :hugs:, or any custom model picked by you, using free compute :computer: :money_mouth_face: provided by Paperspace
  • get to know how GANs work, through multiple community talks and discussions by famous AI artists (announced soon!)
  • get to know the :hugs: ecosystem through Datasets, Accelerate, using either PyTorch/Keras
  • showcasing the magic of a GAN as a :hugs: Space

How does it work :gear:

Participants have two weeks to train and showcase cool GANs using compute and tools provided by :hugs:.

  • All datasets, models and Spaces are uploaded to the HugGAN organization on the hub.
  • Each demo will be evaluated by the team, giving prizes to the most original ones.

During the event, you will have the opportunity to not only get to know GANs, but also how to leverage the tools provided by HuggingFace to make training neural networks a breeze!

What do I need to do to participate :clipboard:

To participate, please follow these steps:

  • fill out this short google form
  • create a Hugging Face Hub account here if you haven’t already, and join the huggan organization
  • join our discord here - when joining the event’s discord channel please make sure to click on the :hugs: emoji under the first message to access all relevant information.

Paperspace kindly offered to provide a limited about of GPUs for participants if needed - if you would like to have access to a GPU, please join the discord for more information.

This sprint should be especially interesting to those interested in generative art and using artificial intelligence for creative applications. Your technical skills will help you select the best image dataset, and possibly build the best GAN demo that ever existed.

More in-detail information will be announced in the discord channel. We are looking forward to seeing you there!

What do I get :gift:

  • enjoy a bit of the Hugging Face vibe
  • learn how to build and train GANs, or create generative art in general
  • learn how to use :hugs: tools, including Datasets, Accelerate and the push_to_hub functionality
  • free compute to build a powerful model and demo it to the community
  • Hugging Face SWAG if you manage to have one of the best and most creative demos
  • team up and make friends in the discord :handshake:

Open-sourcely yours,

Merve, Nate, Niels, & Omar