Distribution Alignment on ImageNet (val)
2.39FDvalBase
Evaluation results
| Method | Links | |
|---|---|---|
| BaseModel=RAE, Embedder=Incep., cfg=1, λ=--, R1=--2026.06 | 2.39 | |
| DRL (λ*)Model=RAE, Embedder=Incep., cfg=1, λ=1, R1=10−52026.06 | 2.52 | |
| DRL (λ=1, R1=0)Model=RAE, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 2.72 | |
| DRL (λ*)Model=JiT, Embedder=Incep., cfg=1, λ=5, R1=10−52026.06 | 3.28 | |
| DRL (λ=1, R1=0)Model=SiT, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 3.63 | |
| DRL (λ*)Model=SiT, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 3.63 | |
| DRL (λ=1, R1=0)Model=REPA, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 3.66 | |
| DRL (λ*)Model=REPA, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 3.66 | |
| DRL (λ=1, R1=0)Model=JiT, Embedder=Incep., cfg=1, λ=1, R1=02026.06 | 3.73 | |
| BaseModel=REPA, Embedder=Incep., cfg=1, λ=--, R1=--2026.06 | 5.81 | |
| DRL (λ*)Model=RAE, Embedder=DINOv3-L, cfg=1, λ=10, R1=10−52026.06 | 6.1 | |
| BaseModel=JiT, Embedder=Incep., cfg=1, λ=--, R1=--2026.06 | 6.47 | |
| DRL (λ=1, R1=0)Model=RAE, Embedder=DINOv3-L, cfg=1, λ=1, R1=02026.06 | 7.14 | |
| BaseModel=RAE, Embedder=DINOv3-L, cfg=1, λ=--, R1=--2026.06 | 8.04 | |
| BaseModel=SiT, Embedder=Incep., cfg=1, λ=--, R1=--2026.06 | 8.13 | |
| DRL (λ*)Model=RAE, Embedder=SigLIP-L, cfg=1, λ=10, R1=10−52026.06 | 11.15 | |
| DRL (λ=1, R1=0)Model=RAE, Embedder=SigLIP-L, cfg=1, λ=1, R1=02026.06 | 11.31 | |
| BaseModel=RAE, Embedder=SigLIP-L, cfg=1, λ=--, R1=--2026.06 | 11.88 | |
| DRL (λ*)Model=JiT, Embedder=DINOv3-L, cfg=1, λ=20, R1=10−52026.06 | 13.24 | |
| DRL (λ*)Model=REPA, Embedder=SigLIP-L, cfg=1, λ=5, R1=10−52026.06 | 14.88 | |
| DRL (λ=1, R1=0)Model=REPA, Embedder=SigLIP-L, cfg=1, λ=1, R1=02026.06 | 16.56 | |
| DRL (λ*)Model=REPA, Embedder=DINOv3-L, cfg=1, λ=10, R1=10−52026.06 | 17.37 | |
| DRL (λ*)Model=SiT, Embedder=SigLIP-L, cfg=1, λ=5, R1=10−52026.06 | 18.94 | |
| DRL (λ*)Model=SiT, Embedder=DINOv3-L, cfg=1, λ=20, R1=10−52026.06 | 21.19 | |
| DRL (λ=1, R1=0)Model=SiT, Embedder=SigLIP-L, cfg=1, λ=1, R1=02026.06 | 22.11 | |
| DRL (λ*)Model=JiT, Embedder=SigLIP-L, cfg=1, λ=20, R1=10−52026.06 | 23.04 | |
| DRL (λ=1, R1=0)Model=JiT, Embedder=DINOv3-L, cfg=1, λ=1, R1=02026.06 | 25.08 | |
| DRL (λ=1, R1=0)Model=REPA, Embedder=DINOv3-L, cfg=1, λ=1, R1=02026.06 | 30.81 | |
| DRL (λ=1, R1=0)Model=JiT, Embedder=SigLIP-L, cfg=1, λ=1, R1=02026.06 | 31.36 | |
| BaseModel=REPA, Embedder=SigLIP-L, cfg=1, λ=--, R1=--2026.06 | 32.13 | |
| DRL (λ*)Model=RAE, Embedder=DINOv2-L, cfg=1, λ=10, R1=10−52026.06 | 33.88 | |
| BaseModel=JiT, Embedder=DINOv3-L, cfg=1, λ=--, R1=--2026.06 | 35.11 | |
| BaseModel=JiT, Embedder=SigLIP-L, cfg=1, λ=--, R1=--2026.06 | 40.94 | |
| DRL (λ=1, R1=0)Model=RAE, Embedder=DINOv2-L, cfg=1, λ=1, R1=02026.06 | 42.47 | |
| BaseModel=SiT, Embedder=SigLIP-L, cfg=1, λ=--, R1=--2026.06 | 42.91 | |
| DRL (λ=1, R1=0)Model=SiT, Embedder=DINOv3-L, cfg=1, λ=1, R1=02026.06 | 44.88 | |
| BaseModel=RAE, Embedder=DINOv2-L, cfg=1, λ=--, R1=--2026.06 | 49.27 | |
| DRL (λ*)Model=REPA, Embedder=DINOv2-L, cfg=1, λ=5, R1=10−52026.06 | 56.12 | |
| DRL (λ*)Model=JiT, Embedder=DINOv2-L, cfg=1, λ=20, R1=10−52026.06 | 57.29 | |
| BaseModel=REPA, Embedder=DINOv3-L, cfg=1, λ=--, R1=--2026.06 | 64.46 | |
| DRL (λ*)Model=SiT, Embedder=DINOv2-L, cfg=1, λ=10, R1=10−52026.06 | 70.97 | |
| DRL (λ=1, R1=0)Model=REPA, Embedder=DINOv2-L, cfg=1, λ=1, R1=02026.06 | 72.52 | |
| BaseModel=SiT, Embedder=DINOv3-L, cfg=1, λ=--, R1=--2026.06 | 88.53 | |
| DRL (λ=1, R1=0)Model=SiT, Embedder=DINOv2-L, cfg=1, λ=1, R1=02026.06 | 108.56 | |
| DRL (λ=1, R1=0)Model=JiT, Embedder=DINOv2-L, cfg=1, λ=1, R1=02026.06 | 109 | |
| BaseModel=JiT, Embedder=DINOv2-L, cfg=1, λ=--, R1=--2026.06 | 158.85 | |
| BaseModel=REPA, Embedder=DINOv2-L, cfg=1, λ=--, R1=--2026.06 | 168.91 | |
| BaseModel=SiT, Embedder=DINOv2-L, cfg=1, λ=--, R1=--2026.06 | 249.99 |