Image Reconstruction on ImageNet

Metric: FID (lower is better)

LeaderboardDataset

Loading chart...

Results

Submit a result

#	Model↕	FID▲	Extra Data	Paper	Date↕	Code
1	MGVQ (16x16x8)	0.49	No	MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tok...	2025-07-10	Code
2	MGVQ (16x16x4)	0.64	No	MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tok...	2025-07-10	Code
3	GigaTok-XL-XXL	0.79	No	GigaTok: Scaling Visual Tokenizers to 3 Billion ...	2025-04-11	Code
4	OptVQ (16x16x8)	0.91	No	Preventing Local Pitfalls in Vector Quantization...	2024-12-19	Code
5	OptVQ (16x16x4)	1	No	Preventing Local Pitfalls in Vector Quantization...	2024-12-19	Code
6	IBQ (16x16)	1	No	Taming Scalable Visual Tokenizer for Autoregress...	2024-12-03	Code
7	Mo-VQGAN (16x16x4)	1.12	No	MoVQ: Modulating Quantized Vectors for High-Fide...	2022-09-19	Code
8	Open-Magvit2 (16x16)	1.17	No	Open-MAGVIT2: An Open-Source Project Toward Demo...	2024-09-06	Code
9	ViT-VQGAN (16x16)	1.28	No	Vector-quantized Image Modeling with Improved VQ...	2021-10-09	Code
10	MaskBit (16x16)	1.66	No	MaskBit: Embedding-free Image Generation via Bit...	2024-09-24	Code
11	TiTok-S-128	1.71	No	An Image is Worth 32 Tokens for Reconstruction a...	2024-06-11	Code
12	RQ-VAE (8x8x16)	1.83	No	Autoregressive Image Generation using Residual Q...	2022-03-03	Code
13	MaskGIT-VQGAN (16x16)	2.28	No	MaskGIT: Masked Generative Image Transformer	2022-02-08	Code
14	VQGAN-LC (16x16)	2.62	No	Scaling the Codebook Size of VQGAN to 100,000 wi...	2024-06-17	Code
15	Taming-VQGAN (16x16)	3.64	No	Taming Transformers for High-Resolution Image Sy...	2020-12-17	Code