First, join our discord to connect with the hypergan community.
There are many opportunities to help:
- Share a trained model/json configuration
- Suggest improvements
- Answer an issue
- Help out in the support channel
- Become a developer
- Implement a paper. There is a huge list of GANs left to implement: https://github.com/GKalliatakis/Delving-deep-into-GANs
- Create an API example. Warning: this can be fun.
- Contribute a screenshot, gif or video of your trained network and share it in the discord
- Search for and share good GAN configurations (examples are searchable)
- Do your own custom research. Warning: this is addictive.
- Become a community manager
GANs are very interesting and we welcome and encourage anyone wanting to explore them.
Open an issue describing the feature and a little bit about why you want it. Include a link to a paper if applicable.
Wow you rock - simply send a pull request! Any work contributed will be under the MIT license so everyone can share in the results.
Create a new issue. Please try to keep your titles short and describe how to reproduce the problem if applicable.
We have a lot of issues open if anyone can help answer them. Send a note to hypergan in the discord if you are able to help close any issue.
Create a pull request. Warning: there is a file size github limit on gifs.
Thank you! Issue a pull request
HyperGAN is a community project and it's utility is determined by the community
We foresee developers training models on bleeding edge consumer hardware, then deploying a generator or discriminator to various platforms(phones, tablets, servers, smart toasters).
HyperGAN hopes to make that easy and highly configurable to do with the API, CLI, and UI.
Artists may use GANs to generate paintings, align datasets, or things we cant think of (they are very creative people).
HyperGAN can help artists by being simple to use through the command line and providing a ui.
Researchers may find it useful to compose various parts of different papers to test concepts or add new composable types.
HyperGAN can help researchers by being sharable (and reproducible) using json
configurations.
HyperGAN is in open beta and available to everyone under the MIT license.
GANs are an active area of research and things will likely shift to best accomodate the state of the art.
Email can be sent to hypergan@protonmail.com