tohinz/ConSinGAN
A PyTorch implementation of a GAN model that trains on a single image to perform unconditional generation, editing, and animation.
Not currently ranked — collecting fresh signals.
star history
ConSinGAN is an official implementation of a WACV-2021 paper that trains GANs on individual images rather than large datasets. The model trains iteratively at increasing resolutions, adding convolutional layers as it scales up, and trains only recent layers while using smaller learning rates for earlier layers. It supports unconditional generation, image harmonization, animation, and editing tasks.
Frequently asked
- What is tohinz/ConSinGAN?
- A PyTorch implementation of a GAN model that trains on a single image to perform unconditional generation, editing, and animation.
- Is ConSinGAN open source?
- Yes — tohinz/ConSinGAN is open source, released under the MIT license.
- What language is ConSinGAN written in?
- tohinz/ConSinGAN is primarily written in Python.
- How popular is ConSinGAN?
- tohinz/ConSinGAN has 444 stars on GitHub.
- Where can I find ConSinGAN?
- tohinz/ConSinGAN is on GitHub at https://github.com/tohinz/ConSinGAN.