HyperAI초신경

홈 뉴스 연구 논문 튜토리얼 데이터셋 백과사전 SOTA LLM 모델 GPU 랭킹 컨퍼런스

한국어

HyperAI초신경

Image Generation On Celeba 256X256

평가 지표

bpd

평가 결과

이 벤치마크에서 각 모델의 성능 결과

모델 이름	bpd	Paper Title	Repository
SPN Menick and Kalchbrenner (2019)	0.61	Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling	-
LSGM	0.70	Score-based Generative Modeling in Latent Space
StyleSwin	-	StyleSwin: Transformer-based GAN for High-resolution Image Generation
NCP-VAE	-	A Contrastive Learning Approach for Training Variational Autoencoder Priors	-
MaCow (Unf)	0.95	MaCow: Masked Convolutional Generative Flow
Efficient-VDVAE	0.51	Efficient-VDVAE: Less is more
HiT-B	-	Improved Transformer for High-Resolution GANs
StyleALAE	-	Adversarial Latent Autoencoders
Glow (Kingma and Dhariwal, 2018)	1.03	Glow: Generative Flow with Invertible 1x1 Convolutions
Residual Flow	0.992	Residual Flows for Invertible Generative Modeling
Locally Masked PixelCNN	0.74	Locally Masked Convolution for Autoregressive Models
GLF+perceptual loss (ours)	-	Generative Latent Flow
MSP	-	Latent Space Factorisation and Manipulation via Matrix Subspace Projection
VQGAN	-	Taming Transformers for High-Resolution Image Synthesis
ANF Huang et al. (2020)	0.72	Augmented Normalizing Flows: Bridging the Gap Between Generative Flows and Latent Variable Models
NVAE w/ flow	0.70	NVAE: A Deep Hierarchical Variational Autoencoder
MaCow (Var)	0.67	MaCow: Masked Convolutional Generative Flow

0 of 17 row(s) selected.