#ModelISPaper
1FrequencyBooster-H (CFG=2.9, interval=[0.1,1.0])319.04link
2DDT**310.60link
3REPA**306.30link
4eMIGM-H305.99link
5eMIGM-L304.16link
6MAR-H303.70link
7REPA-E**302.90link
8𝔼: +LPIPS298.42link
9𝔻: +Fusion296.22link
10MAR-L296.00link
11β„‚: +Wide Decoder287.34link
12PixelFlow282.10link
13MAR-B281.70link
14eMIGM-B278.97link
15Exp. 3 (Encoder Layers=10, Decoder Layers=2, Hidden Size=768Γ—2)278.33link
16𝔹: +iREPA263.07link
17RAE**262.60link
18Exp. 5 (iREPA=0.05, LPIPS=0.1)258.24link
19Exp. 6 (iREPA=0.05, LPIPS=0.01)256.55link
20eMIGM-S254.48link
21Exp. 4 (iREPA=0.05, LPIPS=0.5)247.71link
22Exp. 2 (Encoder Layers=10, Decoder Layers=2, Hidden Size=768)233.44link
23Exp. 1 (Encoder Layers=12, Decoder Layers=0, Hidden Size=768)233.40link
24FrequencyBooster-B/16 (Enc=10, Dec=2)232.70link
25eMIGM-XS224.91link
26𝔸: Baseline-B/16176.48link
27+ GAN FT (Full MIMFlow)130.00link
28REPA SiT-XL/2 + Ξ”FM (BS=256)129.89link
29CFM (Ξ»=0.0)115.83link
30REPA SiT-XL/2 (BS=256)115.83link
31+ Masking (0.4–0.6)105.20link
32D+CLIP92.30link
33K=12892.30link
34MIM weight=192.30link
35K=6491.80link
36+ Aux Loss (DINO+CLIP)91.00link
37DINO91.00link
38All88.60link
39Οƒ=0.388.60link
40Baseline (e2e, SD-VAE, 256tok, GAN)86.00link
41MIM weight=585.60link
42Οƒ=0.281.50link
43Οƒ=0.575.60link
44K=19268.00link
45+ Learnable Tokens, - GAN Loss63.10link
46MIM weight=1061.80link
47C+HOG56.00link
48REPA SiT-B/2 + Ξ”FM (BS=256)43.44link
49REPA SiT-B/2 (BS=256)38.04link
ImageNet imagenet Leaderboard