Generative Adversarial Networks (GANs) logopeterindia.net
Knowledge Hub

Generative Adversarial Networks (GANs): Papers, Code, Architectures & Applications

Generative adversarial networks — the idea, landmark papers, implementations, applications and learning resources.

48 topics•6 categories•Updated October 2026

What Is a GAN?

7 entries

A generative adversarial network trains two networks against each other: a generator that turns random noise into samples, and a discriminator that tries to tell real data from generated data. The competition pushes the generator toward realistic output. GANs produce sharp images in a single forward pass, but training can be unstable, suffer mode collapse, and is harder to scale than diffusion, which has largely replaced GANs for text-to-image generation while GANs remain useful for fast generation and image-to-image tasks.

01Generative adversarial network (Wikipedia)Overview of the method and its variants. 02Generative model (Wikipedia)Where GANs sit among generative models. 03StyleGAN (Wikipedia)NVIDIA's style-based GAN family. 04From GAN to WGAN (Lilian Weng)Clear walkthrough of GAN theory and its fixes. 05Google Machine Learning: GANsShort course on GAN structure and training. 06Dive into Deep Learning: GANsTextbook chapter with code. 07Deep Learning (Goodfellow, Bengio, Courville)The textbook by the inventor of GANs and colleagues.

Landmark Papers

11 entries

The architectures that defined the field.

08Generative Adversarial Networks (Goodfellow et al., 2014)The original paper. 09Conditional GANsCondition generation on labels. 10DCGANConvolutional architecture guidelines for stable training. 11Wasserstein GANWasserstein distance for more stable training. 12Improved Training of Wasserstein GANsGradient penalty (WGAN-GP). 13Progressive Growing of GANsGrow resolution during training (ProGAN). 14BigGANLarge-scale GAN training for high-fidelity images. 15StyleGANStyle-based generator architecture. 16StyleGAN2Analyzing and improving image quality. 17StyleGAN3 (Alias-Free GAN)Alias-free generation. 18Diffusion Models Beat GANs on Image SynthesisThe paper that marked the shift toward diffusion.

Applications: Translation, Super-Resolution & Restoration

6 entries

GANs for image-to-image tasks.

19pix2pixImage-to-image translation with conditional GANs. 20CycleGANUnpaired image-to-image translation. 21CycleGAN and pix2pix (PyTorch)Reference implementations. 22SRGANPhoto-realistic single-image super-resolution. 23Real-ESRGANPractical image and video restoration and upscaling. 24GFPGANFace restoration using GAN priors.

Code, Tutorials & Courses

11 entries

Build and train GANs.

25StyleGAN3NVIDIA's official implementation. 26StyleGAN2-ADA (PyTorch)Training with limited data. 27FFHQ datasetHigh-quality face dataset used for StyleGAN. 28PyTorch-GANCollection of PyTorch GAN implementations. 29Keras-GANCollection of Keras GAN implementations. 30PyTorch DCGAN exampleOfficial example. 31PyTorch DCGAN tutorialGenerate faces step by step. 32TensorFlow DCGAN tutorialGenerate handwritten digits. 33GAN ZooList of named GAN variants. 34Really Awesome GANCurated GAN papers and code. 35Generative Adversarial Networks Specialization (Coursera)DeepLearning.AI course.

Deepfakes & Content Provenance

3 entries

Risks and countermeasures.

36Deepfake (Wikipedia)Synthetic media and its risks. 37C2PAOpen standard for content provenance and authenticity. 38Content CredentialsTamper-evident provenance metadata for media.

Related Pages

10 entries

Neighbouring pages on this site.

39Generative AIGenerative AI hub. 40Variational AutoencodersVAEs: papers, code and resources. 41Diffusion ModelsDiffusion and flow-matching models. 42Diffusion Language ModelsDiffusion applied to text. 43Deep LearningDeep learning hub. 44Deep Learning FrameworksFrameworks for deep learning. 45Computer VisionComputer vision hub. 46Generative AI CompaniesGenerative AI companies. 47Machine LearningMachine learning hub. 48Artificial IntelligenceAI hub.

Frequently Asked Questions

What is a GAN?

A generative model made of two networks, a generator and a discriminator, trained against each other so the generator learns to produce realistic samples.

GANs, VAEs or diffusion models?

GANs generate sharp samples in one pass but are harder to train; VAEs train stably and give a latent space; diffusion models give the best quality and diversity for text-to-image and video but need many sampling steps.

Are GANs still used?

Yes, for fast generation, super-resolution, face restoration, image-to-image translation and as perceptual components inside other systems, although diffusion dominates text-to-image generation.

Why are GANs hard to train?

The two networks must stay balanced; problems include mode collapse and unstable gradients, addressed by Wasserstein losses, gradient penalties, spectral normalisation and careful architecture choices.

How current is this list?

Links were checked in October 2026. Archived projects are labelled; sites that block automated checks are kept and may need a manual click.