Image Generation

Edit

659 papers with code • 58 benchmarks • 41 datasets

Image generation (synthesis) is the task of generating new images from an existing dataset.

Unconditional generation refers to generating samples unconditionally from the dataset, i.e. $p(y)$
Conditional image generation (subtask) refers to generating samples conditionally from the dataset, based on a label, i.e. $p(y|x)$.

In this section, you can find state-of-the-art leaderboards for unconditional generation. For conditional generation, and other types of image generations, refer to the subtasks.

( Image credit: StyleGAN )

Benchmarks

Add a Result

Dataset	Best Model	Paper Title	Compare
CIFAR-10	LSGM (FID)	Score-based Generative Modeling in Latent Space	See all
ImageNet 64x64	DenseFlow-74-10	Densely connected normalizing flows	See all
ImageNet 32x32	DenseFlow-74-10	Densely connected normalizing flows	See all
LSUN Bedroom 256 x 256	ADM (dropout)	Diffusion Models Beat GANs on Image Synthesis	See all
STL-10	Styleformer	Styleformer: Transformer based Generative Adversarial Networks with Style Vector	See all
FFHQ 256 x 256	InsGen	Data-Efficient Instance Generation from Instance Discrimination	See all
FFHQ	StyleGAN2	Analyzing and Improving the Image Quality of StyleGAN	See all
CelebA 256x256	SPN Menick and Kalchbrenner (2019)	Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling	See all
LSUN Churches 256 x 256	CIPS	Image Generators with Conditionally-Independent Pixel Synthesis	See all
MNIST	Locally Masked PixelCNN (8 orders)	Locally Masked Convolution for Autoregressive Models	See all
CelebA 64x64	Styleformer	Styleformer: Transformer based Generative Adversarial Networks with Style Vector	See all
Binarized MNIST	CR-NVAE	Consistency Regularization for Variational Auto-Encoders	See all
CelebA-HQ 128x128	U-Net GAN	A U-Net Based Discriminator for Generative Adversarial Networks	See all
ImageNet 128x128	ADM-G	Diffusion Models Beat GANs on Image Synthesis	See all
CelebA-HQ 256x256	UNCSN	Score Matching Model for Unbounded Data Score	See all
ImageNet 256x256	ADM-G, ADM-U	Diffusion Models Beat GANs on Image Synthesis	See all
CelebA-HQ 1024x1024	StyleGAN	A Style-Based Generator Architecture for Generative Adversarial Networks	See all
CIFAR-100	LeCAM (StyleGAN2 + ADA)	Regularizing Generative Adversarial Networks under Limited Data	See all
CLEVR	GANformer	Generative Adversarial Transformers	See all
Cityscapes	GANformer	Generative Adversarial Transformers	See all
LSUN Cat 256 x 256	ADM (dropout)	Diffusion Models Beat GANs on Image Synthesis	See all
CelebA-HQ 64x64	VAEBM	VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models	See all
CAT 256x256	StyleGAN2 + DA + RLC (Ours)	Regularizing Generative Adversarial Networks under Limited Data	See all
Fashion-MNIST	GLF+perceptual loss (ours)	Generative Latent Flow	See all
Stacked MNIST	VAEBM	VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models	See all
CUB 128 x 128	FineGAN	FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery	See all
Stanford Dogs	FineGAN	FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery	See all
Stanford Cars	FineGAN	FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery	See all
CelebA 128x128	U-Net GAN	A U-Net Based Discriminator for Generative Adversarial Networks	See all
CIFAR-10 (20% data)	DiffAugment-StyleGAN2	Differentiable Augmentation for Data-Efficient GAN Training	See all
CIFAR-10 (10% data)	DiffAugment-StyleGAN2	Differentiable Augmentation for Data-Efficient GAN Training	See all
ObjectsRoom	GENESIS-V2	GENESIS-V2: Inferring Unordered Object Representations without Iterative Refinement	See all
ShapeStacks	GENESIS-V2	GENESIS-V2: Inferring Unordered Object Representations without Iterative Refinement	See all
CelebA 128 x 128	COCO-GAN	COCO-GAN: Generation by Parts via Conditional Coordinating	See all
LSUN Horse 256 x 256	ADM (dropout)	Diffusion Models Beat GANs on Image Synthesis	See all
LSUN Bedroom 64 x 64	WGAN-GP + TT Update Rule	GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium	See all
FFHQ 1024 x 1024	StyleGAN2 ADA+bCR	Training Generative Adversarial Networks with Limited Data	See all
LSUN Bedroom	StyleGAN	A Style-Based Generator Architecture for Generative Adversarial Networks	See all
RC-49	cDRE-F-cSP+RS	Efficient Subsampling for Generating High-Quality Images from Conditional Generative Adversarial Networks	See all
ImageNet 512x512	ADM-G, ADM-U	Diffusion Models Beat GANs on Image Synthesis	See all
Cityscapes-5K 256x512	SB-GAN	Semantic Bottleneck Scene Generation	See all
Cityscapes-25K 256x512	SB-GAN	Semantic Bottleneck Scene Generation	See all
ADE-Indoor	SB-GAN	Semantic Bottleneck Scene Generation	See all
Indian Celebs 256 x 256	MSG-StyleGAN	MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks	See all
Oxford 102 Flowers 256 x 256	MSG-StyleGAN	MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks	See all
LSUN Car 256 x 256	StyleGAN2	Analyzing and Improving the Image Quality of StyleGAN	See all
LSUN Car 512 x 384	StyleGAN2	Analyzing and Improving the Image Quality of StyleGAN	See all
Multi-dSprites	GENESIS	GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representations	See all
GQN	GENESIS	GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representations	See all
Landscapes 256 x 256	CIPS	Image Generators with Conditionally-Independent Pixel Synthesis	See all
Satellite-Buildings 256 x 256	CIPS	Image Generators with Conditionally-Independent Pixel Synthesis	See all
Satellite-Landscapes 256 x 256	CIPS	Image Generators with Conditionally-Independent Pixel Synthesis	See all
LSUN Bedroom 128 x 128	DC-VAE	Dual Contradistinctive Generative Autoencoder	See all
FFHQ 512 x 512	Ours (high res)	Anycost GANs for Interactive Image Synthesis and Editing	See all
FFHQ 128 x 128	LeCAM (StyleGAN2 + ADA)	Regularizing Generative Adversarial Networks under Limited Data	See all
Oxford 102 Flowers 128x128	QSNGAN	Quaternion Generative Adversarial Networks	See all
25% ImageNet 128x128	LeCAM + DA	Regularizing Generative Adversarial Networks under Limited Data	See all
ImageNet - 10% labeled data	LeCAM + DA	Regularizing Generative Adversarial Networks under Limited Data	See all

Show all 58 benchmarks

Collapse benchmarks

Datasets

Subtasks

Face Generation

Text-to-Image Generation

Pose Transfer

Image Interpolation

Facial Inpainting

Layout-to-Image Generation

Pose-Guided Image Generation

User Constrained Thumbnail Generation

Handwritten Word Generation

person reposing

Chinese Landscape Painting Generation

ROI-based image generation

Greatest papers with code

Density estimation using Real NVP

tensorflow/models • • 27 May 2016

Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning.

Ranked #17 on Image Generation on ImageNet 32x32

Density Estimation Image Generation

70,757

Paper
Code

Improved Techniques for Training GANs

tensorflow/models • • NeurIPS 2016

We present a variety of new architectural features and training procedures that we apply to the generative adversarial networks (GANs) framework.

Ranked #11 on Conditional Image Generation on CIFAR-10

Conditional Image Generation Semi-Supervised Image Classification

61,324

Paper
Code

Reformer: The Efficient Transformer

huggingface/transformers • • ICLR 2020

Large Transformer models routinely achieve state-of-the-art results on a number of tasks but training these models can be prohibitively costly, especially on long sequences.

Ranked #2 on Open-Domain Question Answering on SearchQA

Image Generation Language Modelling +1

48,998

Paper
Code

Infinite Nature: Perpetual View Generation of Natural Scenes from a Single Image

google-research/google-research • • 17 Dec 2020

We introduce the problem of perpetual view generation -- long-range generation of novel views corresponding to an arbitrarily long camera trajectory given a single image.

Image Generation Video Generation

18,580

Paper
Code

Rethinking Attention with Performers

google-research/google-research • • ICLR 2021

We introduce Performers, Transformer architectures which can estimate regular (softmax) full-rank-attention Transformers with provable accuracy, but using only linear (as opposed to quadratic) space and time complexity, without relying on any priors such as sparsity or low-rankness.

Ranked #12 on Image Generation on ImageNet 64x64

Image Generation

18,577

Paper
Code

A General and Adaptive Robust Loss Function

google-research/google-research • • CVPR 2019

We present a generalization of the Cauchy/Lorentzian, Geman-McClure, Welsch/Leclerc, generalized Charbonnier, Charbonnier/pseudo-Huber/L1-L2, and L2 loss functions.

Image Generation

18,575

Paper
Code

Self-Attention Generative Adversarial Networks

jantic/DeOldify • • arXiv 2018

In this paper, we propose the Self-Attention Generative Adversarial Network (SAGAN) which allows attention-driven, long-range dependency modeling for image generation tasks.

Ranked #11 on Conditional Image Generation on ImageNet 128x128

Conditional Image Generation

13,889

Paper
Code

GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

jantic/DeOldify • • NeurIPS 2017

Generative Adversarial Networks (GANs) excel at creating realistic images with complex models for which maximum likelihood is infeasible.

Ranked #1 on Image Generation on LSUN Bedroom 64 x 64

Image Generation

13,889

Paper
Code

Training with Quantization Noise for Extreme Model Compression

pytorch/fairseq • • ICLR 2021

A standard solution is to train networks with Quantization Aware Training, where the weights are quantized during training and the gradients approximated with the Straight-Through Estimator.

Image Generation Model Compression

13,407

Paper
Code