Image Generation On Lsun Churches 256 X 256

评估指标

FID

评测结果

各个模型在此基准测试上的表现结果

		Paper Title	Repository
VE (erel=0.01)	26.46	Gotta Go Fast When Generating Data with Score-Based Models
VE (erel=0.02)	26.46	Gotta Go Fast When Generating Data with Score-Based Models
TransGAN	8.94	TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up
PNDM	8.69	Pseudo Numerical Methods for Diffusion Models on Manifolds
Denoising Diffusion Probabilistic Model	7.89	Denoising Diffusion Probabilistic Models
PGGAN	6.42	Progressive Growing of GANs for Improved Quality, Stability, and Variation
LFM	5.54	Flow Matching in Latent Space
DDGAN	5.25	Tackling the Generative Learning Trilemma with Denoising Diffusion GANs
MSG-StyleGAN	5.2	MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks
WaveDiff	5.06	Wavelet Diffusion Models are fast and scalable Image Generators
SWAGAN-Bi	4.97	SWAGAN: A Style-based Wavelet-driven Generative Model
StyleGAN	4.21	A Style-Based Generator Architecture for Generative Adversarial Networks
Unleashing Transformers	4.07	Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes
INR-GAN-bil	4.04	Adversarial Generation of Continuous Images
TDPM+ (TTrunc=99)	3.98	Truncated Diffusion Probabilistic Models and Diffusion-based Adversarial Auto-Encoders
Polarity-StyleGAN2	3.92	Polarity Sampling: Quality and Diversity Control of Pre-Trained Generative Networks via Singular Values
StyleGAN2	3.86	Analyzing and Improving the Image Quality of StyleGAN
CLR-GAN	3.43	CLR-GAN: Improving GANs Stability and Quality via Consistent Latent Representation and Reconstruction	-
StyleNAT	3.4	StyleNAT: Giving Each Head a New Perspective
Diffusion StyleGAN2	3.17	Diffusion-GAN: Training GANs with Diffusion

0 of 27 row(s) selected.