Image Classification on ImageNet-1K (val) (Top-1 and Top-5 Scores)
83.8Top-1 AccuracyDerf
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Derfbackbone=ViT-L2025.12 | 83.8 | — | |
| DyTbackbone=ViT-L2025.12 | 83.6 | — | |
| LayerNormbackbone=ViT-L2025.12 | 83.1 | — | |
| Derfbackbone=ViT-B2025.12 | 82.8 | — | |
| DyTbackbone=ViT-B2025.12 | 82.5 | — | |
| LayerNormbackbone=ViT-B2025.12 | 82.3 | — | |
| Koo-Fu CLIPf(·) (Prototype transformation)=1K (Sw:1K), class. (Search Space)=1K2026.02 | 79.05 | 95.54 | |
| Koo-Fu CLIPf(·) (Prototype transformation)=21K (Sw:21K), class. (Search Space)=1K2026.02 | 78.4 | 95.31 | |
| SupervisedBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 76.5 | — | |
| ResNet-50Precision=Mixed precision DFP16, Batch-size=1024, Epochs=902018.02 | 75.77 | 92.84 | |
| ResNet-50Precision=Baseline FP32, Batch-size=1024, Epochs=902018.02 | 75.7 | 92.78 | |
| CLIPf(·) (Prototype transformation)=None (-), class. (Search Space)=1K2026.02 | 75.08 | 93.42 | |
| MV-MRBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 74.5 | 92.1 | |
| BYOLBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 74.3 | 91.6 | |
| OBOWBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 73.8 | — | |
| Barlow TwinsBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 73.2 | 91 | |
| VICRegBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 73.2 | 91.1 | |
| InfoMin AugBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 73 | 91.1 | |
| SimSiamBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 71.3 | — | |
| MV-MRBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=10%2023.03 | 69.9 | 89.5 | |
| Barlow TwinsBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=10%2023.03 | 69.7 | 89.3 | |
| VICRegBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=10%2023.03 | 69.5 | 89.5 | |
| GoogLeNet-v1Precision=Mixed precision DFP16, Batch-size=1024, Epochs=802018.02 | 69.34 | 89.31 | |
| GoogLeNet-v1Precision=Baseline FP32, Batch-size=1024, Epochs=802018.02 | 69.26 | 89.31 | |
| BYOLBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=10%2023.03 | 68.8 | 89 | |
| VGG-16Precision=Baseline FP32, Batch-size=256, Epochs=602018.02 | 68.23 | 88.47 | |
| VGG-16Precision=Mixed precision DFP16, Batch-size=256, Epochs=602018.02 | 68.12 | 88.18 | |
| PIRLBackbone=ResNet-50, Evaluation Protocol=Linear2023.03 | 63.6 | — | |
| Koo-Fu CLIPf(·) (Prototype transformation)=1K (Sw:1K), class. (Search Space)=14K2026.02 | 60.52 | 71.39 | |
| Koo-Fu CLIPf(·) (Prototype transformation)=21K (Sw:21K), class. (Search Space)=14K2026.02 | 60.09 | 71.23 | |
| CLIPf(·) (Prototype transformation)=None (-), class. (Search Space)=14K2026.02 | 57.79 | 70.03 | |
| AlexNetPrecision=Baseline FP32, Batch-size=1024, Epochs=882018.02 | 57.43 | 80.65 | |
| AlexNetPrecision=Mixed precision DFP16, Batch-size=1024, Epochs=882018.02 | 56.94 | 80.06 | |
| SupervisedBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=10%2023.03 | 56.4 | 80.4 | |
| MV-MRBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=1%2023.03 | 56.1 | 79.4 | |
| Barlow TwinsBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=1%2023.03 | 55 | 79.2 | |
| VICRegBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=1%2023.03 | 54.8 | 79.4 | |
| BYOLBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=1%2023.03 | 53.2 | 78.4 | |
| Koo-Fu CLIPf(·) (Prototype transformation)=1K (Sw:1K), class. (Search Space)=21K2026.02 | 46.84 | 75.7 | |
| Koo-Fu CLIPf(·) (Prototype transformation)=21K (Sw:21K), class. (Search Space)=21K2026.02 | 45.32 | 74.76 | |
| CLIPf(·) (Prototype transformation)=None (-), class. (Search Space)=21K2026.02 | 42.73 | 71.28 | |
| SupervisedBackbone=ResNet-50, Evaluation Protocol=Semi-Supervised, Label Percentage=1%2023.03 | 25.4 | 48.4 |