Find short PyTorch GAN implementations (50-100 lines) to understand how a basic GAN works without reading dense papers.
Follow the curated paper list to build a structured reading path through GAN research history up to 2017.
Explore applied GAN techniques like image-to-image translation, super-resolution, and face generation through linked papers.
Use CycleGAN references to learn how to translate between two visual styles without paired training data.
This is a reading list with no code to install, just browse the links directly.
This repository is a curated reading list of academic papers, tutorials, videos, and code implementations covering Generative Adversarial Networks, commonly called GANs. A GAN is a type of machine learning setup where two neural networks are trained together in competition: one network (the generator) tries to produce realistic-looking data, while the other (the discriminator) tries to tell real data apart from fake. Over many training rounds, the generator gets better at fooling the discriminator, which is how the system learns to produce things like photorealistic images. The list is organized into several sections. The tutorials and blogs section includes beginner-friendly introductions, including an official NIPS 2016 tutorial by Ian Goodfellow, one of the original GAN inventors. The code section links to minimal working implementations (some as short as 50 or 100 lines of PyTorch code) that reproduce results from key papers, useful for anyone who wants to understand the mechanics without reading a dense paper first. The papers section is the largest part of the repository, split into theory-focused work and applied work. The applied vision section covers uses like generating photorealistic faces, translating images from one style to another (for example, turning a sketch into a photo, or converting images from summer to winter), super-resolution (making low-quality images sharper), and generating images from text descriptions. Some highlighted recommendations include CycleGAN, which can translate between two visual styles without needing paired training examples, and work on generating realistic frontal-face views from angled photos. The maintainer stopped adding new papers in November 2017, noting that GANs had moved from a niche research topic to mainstream machine learning by that point, making an exhaustive list impractical. The list reflects research up to that date and is offered as a historical reference. Others are invited to fork and continue it. The full list of papers is longer than what was shown.
← nightrome on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.