Image Classification on ImageNet-1K (test) (Accuracy & Calibration Metrics)
79.36Top-1 AccuracyHO-HE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HO-HEBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 79.36 | — | — | — | |
| Diversity (α = 1)Backbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 79.14 | — | — | — | |
| HO-HEBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 78.97 | — | — | — | |
| SBSim (α = 0)Backbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 78.76 | — | — | — | |
| Diversity (α = 1)Backbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 78.72 | — | — | — | |
| RandomBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 78.68 | — | — | — | |
| RealScoreBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 78.67 | — | — | — | |
| SBSim (α = 0)Backbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 78.58 | — | — | — | |
| RandomBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 78.52 | — | — | — | |
| RealScoreBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 78.52 | — | — | — | |
| CSAMBackbone=ResNet-50, Optimizer=CSAM2025.05 | 77.95 | 1.18 | 1.09 | 1.19 | |
| EC-DynamicDynamic Depth=true, Dynamic Width=true, Dynamic Resolution=true, #MACs (B)=2.62026.07 | 77.6 | — | — | — | |
| SAMBackbone=ResNet-50, Optimizer=SAM2025.05 | 77.32 | 1.52 | 1.54 | 1.44 | |
| DR-ResNetDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=3.22026.07 | 77 | — | — | — | |
| SGDBackbone=ResNet-50, Optimizer=SGD2025.05 | 76.97 | 3.39 | 1.8 | 3.31 | |
| EC-DynamicDynamic Depth=true, Dynamic Width=true, Dynamic Resolution=true, #MACs (B)=2.12026.07 | 76.8 | — | — | — | |
| SkipNetDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=3.62026.07 | 76.2 | — | — | — | |
| ConvNet-AIGDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=3.12026.07 | 76.2 | — | — | — | |
| Channel SelectionDynamic Depth=false, Dynamic Width=true, Dynamic Resolution=false, #MACs (B)=2.52026.07 | 76.2 | — | — | — | |
| Channel SelectionDynamic Depth=false, Dynamic Width=true, Dynamic Resolution=false, #MACs (B)=2.32026.07 | 76.1 | — | — | — | |
| ResNet-50Dynamic Depth=false, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=4.12026.07 | 76 | — | — | — | |
| CSAMBackbone=ViT-S/16, Optimizer=CSAM2025.05 | 75.91 | 1.58 | 1.34 | 1.54 | |
| MSDNetDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=2.12026.07 | 75.7 | — | — | — | |
| SAMBackbone=ViT-S/16, Optimizer=SAM2025.05 | 75.42 | 1.76 | 1.66 | 1.73 | |
| ConvNet-AIGDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=2.62026.07 | 75.3 | — | — | — | |
| DR-ResNetDynamic Depth=true, Dynamic Width=false, Dynamic Resolution=false, #MACs (B)=2.32026.07 | 75.3 | — | — | — | |
| CLIP-AlignBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 75.23 | — | — | — | |
| RANetDynamic Depth=true, Dynamic Width=true, Dynamic Resolution=false, #MACs (B)=2.02026.07 | 75.2 | — | — | — | |
| HO-HEBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 74.76 | — | — | — | |
| Diversity (α = 1)Backbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 74.25 | — | — | — | |
| SBSim (α = 0)Backbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 74.14 | — | — | — | |
| HO-HEBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 74.02 | — | — | — | |
| Diversity (α = 1)Backbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.79 | — | — | — | |
| RandomBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 73.65 | — | — | — | |
| RealScoreBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 73.62 | — | — | — | |
| RandomBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.54 | — | — | — | |
| RealScoreBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.53 | — | — | — | |
| HO-HEBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.33 | — | — | — | |
| Diversity (α = 1)Backbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.25 | — | — | — | |
| SBSim (α = 0)Backbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 73.23 | — | — | — | |
| HO-HEBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 73.14 | — | — | — | |
| CLIP-AlignBackbone=ViT-B/16, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 72.4 | — | — | — | |
| Diversity (α = 1)Backbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 72.25 | — | — | — | |
| SBSim (α = 0)Backbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 72.18 | — | — | — | |
| SBSim (α = 0)Backbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 72.18 | — | — | — | |
| RealScoreBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 72.16 | — | — | — | |
| RandomBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 72.13 | — | — | — | |
| HO-HEBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 72.09 | — | — | — | |
| RandomBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 72.03 | — | — | — | |
| RealScoreBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 72.01 | — | — | — | |
| SBSim (α = 0)Backbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 71.97 | — | — | — | |
| RealScoreBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 71.93 | — | — | — | |
| RandomBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 71.9 | — | — | — | |
| Diversity (α = 1)Backbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 71.8 | — | — | — | |
| HO-HEBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 71.49 | — | — | — | |
| Diversity (α = 1)Backbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 71.42 | — | — | — | |
| AdamWBackbone=ViT-S/16, Optimizer=AdamW2025.05 | 71.35 | 9.72 | 3.66 | 9.72 | |
| CLIP-AlignBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 70.98 | — | — | — | |
| SBSim (α = 0)Backbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 70.95 | — | — | — | |
| RealScoreBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 70.92 | — | — | — | |
| RandomBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 70.9 | — | — | — | |
| CLIP-AlignBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+3M2026.07 | 70.88 | — | — | — | |
| CSAMBackbone=ViT-S/32, Optimizer=CSAM2025.05 | 70.01 | 2.88 | 0.92 | 2.78 | |
| CLIP-AlignBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 69.44 | — | — | — | |
| RandomBackbone=ResNet-50, Training Protocol=Scratch, Synthetic Samples=+02026.07 | 69.27 | — | — | — | |
| SAMBackbone=ViT-S/32, Optimizer=SAM2025.05 | 69.21 | 3.04 | 1.18 | 3.05 | |
| CLIP-AlignBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+1M2026.07 | 69.18 | — | — | — | |
| CLIP-AlignBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=3M2026.07 | 67.57 | — | — | — | |
| AdamWBackbone=ViT-S/32, Optimizer=AdamW2025.05 | 65.03 | 9.11 | 2.63 | 9.11 | |
| RandomBackbone=ViT-B/16, Training Protocol=Scratch, Synthetic Samples=+02026.07 | 64.22 | — | — | — | |
| CLIP-AlignBackbone=ResNet-50, Training Protocol=Fine-tune, Synthetic Samples=1M2026.07 | 62.57 | — | — | — |