ImageNet imagenet-2 Leaderboard
Auto-discovered from papers reporting ImageNet (rFID). Β· Metric: rFID (higher is better) Β· π’ Updated 9 min ago
| # | Model | rFID | Paper |
|---|---|---|---|
| 1 | K=192 | 24.59 | link |
| 2 | + Aux Loss (DINO+CLIP) | 14.75 | link |
| 3 | MIM weight=10 | 13.73 | link |
| 4 | Ο=0.5 | 6.95 | link |
| 5 | K=64 | 5.61 | link |
| 6 | C+HOG | 4.18 | link |
| 7 | + Learnable Tokens, - GAN Loss | 4.15 | link |
| 8 | MIM weight=5 | 4.10 | link |
| 9 | DINO | 4.00 | link |
| 10 | Ο=0.2 | 3.93 | link |
| 11 | + Masking (0.4β0.6) | 3.81 | link |
| 12 | All | 3.64 | link |
| 13 | D+CLIP | 3.60 | link |
| 14 | K=128 | 3.60 | link |
| 15 | MIM weight=1 | 3.60 | link |
| 16 | Ο=0.3 | 3.40 | link |
| 17 | DiTo -S-LPIPS (#D=48.3M) | 3.17 | link |
| 18 | Baseline (e2e, SD-VAE, 256tok, GAN) | 2.74 | link |
| 19 | | | + SSDD -M decoder (#D=48.0M) | 2.01 | link |
| 20 | | | + REPA loss | 1.58 | link |
| 21 | + GAN FT (Full MIMFlow) | 1.47 | link |
| 22 | | | + replace z-norm by KL | 1.32 | link |
| 23 | | | + logit-normal sampling | 1.30 | link |
| 24 | | | + t t -spacing sampler | 1.25 | link |
| 25 | | | + EMA | 1.17 | link |
| 26 | | | + distillation ( SSDD -M) | 1.13 | link |
| 27 | ( Β§ \S ) + GAN loss | 1.07 | link |
| 28 | | | + shared encoder | 1.07 | link |
| 29 | | | + shared pre-training ( SSDD (8) -M) | 1.06 | link |
| 30 | DiTo-XL-LPIPS (DiTo-XL-LPIPS decoder) | 0.78 | link |
| 31 | SD-VAE (SD-VAE decoder) | 0.69 | link |
| 32 | SD-VAE encoder + SD-VAE-XL decoder | 0.64 | link |
| 33 | DiTo-XL-LPIPS encoder + SSDD(8)-M decoder | 0.52 | link |
| 34 | SD-VAE encoder + SSDD(8)-M decoder | 0.47 | link |
| 35 | SD-VAE encoder + SSDD(8)-L decoder | 0.45 | link |
| 36 | SD-VAE encoder + SSDD(8)-H decoder | 0.41 | link |