Image Generation on ImageNet-1K 512x512 (val)
0.19rFIDUVIT-L/4
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| UVIT-L/4tokenizer=VAE [57], generator=UVIT-L/4 [4], parameters=287M, sampling_steps=502024.06 | 0.19 | 18.03 | 76.9 | 4.67 | 213.3 | 1 | — | — | |
| UVIT-H/4tokenizer=VAE [57], generator=UVIT-H/4 [4], parameters=501M, sampling_steps=502024.06 | 0.19 | 15.71 | 101.3 | 4.05 | 263.8 | 0.6 | — | — | |
| DiT-XL/2tokenizer=VAE [57], generator=DiT-XL/2 [49], parameters=675M, sampling_steps=2502024.06 | 0.19 | 12.03 | 105.3 | 3.04 | 240.8 | 0.1 | — | — | |
| UViT-L/4tokenizer=VAE ()‡, P=287M, S=50, T=1.02026.03 | 0.19 | 18.03 | 76.9 | 4.67 | 213.3 | — | — | — | |
| UViT-H/4tokenizer=VAE ()‡, P=501M, S=50, T=0.62026.03 | 0.19 | 15.71 | 101.3 | 4.05 | 263.8 | — | — | — | |
| DiT-XL/2tokenizer=VAE ()‡, P=675M, S=250, T=0.12026.03 | 0.19 | 12.03 | 105.3 | 3.04 | 240.8 | — | — | — | |
| CARD-Btokenizer=SMAP(KL)-B-128, P=234M, S=25, T=11.72026.03 | 0.69 | 4.29 | 155.2 | 2.99 | 253.1 | — | — | — | |
| CARD-Ltokenizer=SMAP(KL)-B-128, P=568M, S=25, T=2.82026.03 | 0.69 | 3.12 | 211.1 | 2.11 | 298.4 | — | — | — | |
| MAGVIT-v2tokenizer=LFQ [68], generator=MAGVIT-v2 [68], parameters=307M, sampling_steps=122024.06 | 1.22 | 4.61 | 192.4 | — | — | 3.5 | — | — | |
| MAGVIT-v2tokenizer=LFQ [68], generator=MAGVIT-v2 [68], parameters=307M, sampling_steps=642024.06 | 1.22 | 3.07 | 213.1 | 1.91 | 324.3 | 1 | — | — | |
| MAGVIT-v2tokenizer=LFQ (Yu et al., 2024a), P=307M, S=12, T=3.52026.03 | 1.22 | 4.61 | 192.4 | — | — | — | — | — | |
| MAGVIT-v2tokenizer=LFQ (Yu et al., 2024a), P=307M, S=64, T=1.02026.03 | 1.22 | 3.07 | 213.1 | 1.91 | 324.3 | — | — | — | |
| TiTok-B-128tokenizer=TiTok-B-128, generator=MaskGIT-ViT [9], parameters=177M, sampling_steps=82024.06 | 1.37 | 3.91 | 182 | 2.49 | 260.4 | 33.3 | — | — | |
| TiTok-B-128tokenizer=TiTok-B-128, generator=MaskGIT-ViT [9], parameters=177M, sampling_steps=642024.06 | 1.37 | 4.17 | 181 | 2.13 | 261.2 | 7.4 | — | — | |
| MaskGIT-ViT (with SMAP-B-128)tokenizer=SMAP-B-128, P=177M, S=8, T=33.32026.03 | 1.37 | 3.91 | 182 | 2.49 | 260.4 | — | — | — | |
| MaskGIT-ViT (with SMAP-B-128)tokenizer=SMAP-B-128, P=177M, S=64, T=7.42026.03 | 1.37 | 4.17 | 181 | 2.13 | 261.2 | — | — | — | |
| TiTok-L-64tokenizer=TiTok-L-64, generator=MaskGIT-ViT [9], parameters=177M, sampling_steps=82024.06 | 1.78 | 3.64 | 179.8 | 2.74 | 221.1 | 41 | — | — | |
| MaskGIT-ViT (with SMAP-L-64)tokenizer=SMAP-L-64, P=177M, S=8, T=41.02026.03 | 1.78 | 3.64 | 179.8 | 2.74 | 221.1 | — | — | — | |
| MaskGIT-ViTtokenizer=MaskGIT-VQGAN [9], generator=MaskGIT-ViT [9], parameters=177M, sampling_steps=122024.06 | 1.97 | 7.32 | 156 | — | — | 3.9 | — | — | |
| MaskGIT-ViTtokenizer=MaskGIT-VQGAN (Chang et al., 2022), P=177M, S=12, T=3.92026.03 | 1.97 | 7.32 | 156 | — | — | — | — | — | |
| BAR-L#params=1.1B, Guidance=classifier-free guidance (linear schedule)2026.02 | — | — | — | — | — | — | 1.09 | 311.1 | |
| DDT#params=675M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 1.28 | 305.1 | |
| DiMR#params=525M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 2.89 | 289.8 | |
| DiT#params=675M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 3.04 | 240.8 | |
| MaskGiT#params=227M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 7.32 | 156 | |
| RAE#params=839M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 1.13 | 259.6 | |
| RAR#params=1.5B, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 1.66 | 295.7 | |
| REPA#params=675M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 2.08 | 274.6 | |
| VAR#params=2.3B, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 2.63 | 303.2 | |
| VQGAN#params=227M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 26.52 | 66.8 | |
| xAR#params=608M, Guidance=classifier-free guidance2026.02 | — | — | — | — | — | — | 1.7 | 281.5 |