Accueil Actualités Articles de recherche récents Tutoriels Ensembles de données Wiki SOTA Modèles LLM Classement GPU Événements

Français

Image Reconstruction On Imagenet

Métriques

FID

LPIPS

PSNR

SSIM

Résultats

Résultats de performance de divers modèles sur ce benchmark

Nom du modèle	FID	LPIPS	PSNR	SSIM	Paper Title	Repository
Taming-VQGAN (16x16)	3.64	0.177	19.93	0.542	Taming Transformers for High-Resolution Image Synthesis
Open-Magvit2 (16x16)	1.17	-	21.90	-	Open-MAGVIT2: An Open-Source Project Toward Democratizing Auto-regressive Visual Generation
TiTok-S-128	1.71	-	-	-	An Image is Worth 32 Tokens for Reconstruction and Generation
VQGAN-LC (16x16)	2.62	0.120	23.80	0.589	Scaling the Codebook Size of VQGAN to 100,000 with a Utilization Rate of 99%
OptVQ (16x16x8)	0.91	0.066	27.57	0.729	Preventing Local Pitfalls in Vector Quantization via Optimal Transport
MaskBit (16x16)	1.66	-	-	-	MaskBit: Embedding-free Image Generation via Bit Tokens
IBQ (16x16)	1.00	0.2030	-	-	Taming Scalable Visual Tokenizer for Autoregressive Image Generation	-
ViT-VQGAN (16x16)	1.28	-	-	-	Vector-quantized Image Modeling with Improved VQGAN
MaskGIT-VQGAN (16x16)	2.28	-	-	-	MaskGIT: Masked Generative Image Transformer
RQ-VAE (8x8x16)	1.83	-	-	-	Autoregressive Image Generation using Residual Quantization
OptVQ (16x16x4)	1.00	0.076	26.59	0.717	Preventing Local Pitfalls in Vector Quantization via Optimal Transport
Mo-VQGAN (16x16x4)	1.12	0.113	22.42	0.673	MoVQ: Modulating Quantized Vectors for High-Fidelity Image Generation

0 of 12 row(s) selected.