Classification on CIFAR100
93.45AccuracyFP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FPBackbone=ViT-B, Task=Fine-tuning2025.03 | 93.45 | — | |
| HOTBackbone=ViT-B, Task=Fine-tuning2025.03 | 92.99 | — | |
| LoRABackbone=ViT-B, Task=Fine-tuning2025.03 | 92.61 | — | |
| HOT + LoRABackbone=ViT-B, Task=Fine-tuning2025.03 | 92.51 | — | |
| LBP-WHTBackbone=ViT-B, Task=Fine-tuning2025.03 | 92.47 | — | |
| LUQBackbone=ViT-B, Task=Fine-tuning2025.03 | 91.76 | — | |
| DGMMC-SEmbedding=ImageBind, G (Number of Gaussians)=12024.10 | 91.2 | — | |
| MVT + FTArch=ViT-g2025.12 | 91.2 | — | |
| OTF-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 90.21 | — | |
| Vanilla FTArch=ViT-g2025.12 | 89.1 | — | |
| MVTArch=ViT-g2025.12 | 89 | — | |
| EVAArch=ViT-g2025.12 | 88.8 | — | |
| TC-JEPAViT=H/14, Protocol=linear probing2026.05 | 88.5 | — | |
| FPBackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 88.04 | — | |
| StoPViT=H/14, Protocol=linear probing2026.05 | 87.7 | — | |
| I-JEPAViT=H/14, Protocol=linear probing2026.05 | 87.5 | — | |
| HOTBackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 87.39 | — | |
| FPBackbone=EfficientNetV2-s, Task=Fine-tuning2025.03 | 87.36 | — | |
| LBP-WHTBackbone=EfficientNetV2-s, Task=Fine-tuning2025.03 | 87.1 | — | |
| FPBackbone=ViT-S, Task=Fine-tuning2025.03 | 86.81 | — | |
| Supervised-INArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 86.4 | — | |
| LUQBackbone=ViT-S, Task=Fine-tuning2025.03 | 86.37 | — | |
| HOTBackbone=EfficientNetV2-s, Task=Fine-tuning2025.03 | 86.29 | — | |
| BYOLArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 86.1 | — | |
| LBP-WHTBackbone=ViT-S, Task=Fine-tuning2025.03 | 85.96 | — | |
| SimCLRArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 85.9 | — | |
| DOT-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 85.83 | — | |
| HOTBackbone=ViT-S, Task=Fine-tuning2025.03 | 85.81 | — | |
| FPBackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 85.73 | — | |
| LoRABackbone=ViT-S, Task=Fine-tuning2025.03 | 85.67 | — | |
| iBOTViT=B/16, Protocol=linear probing2026.05 | 85.5 | — | |
| HOT + LoRABackbone=ViT-S, Task=Fine-tuning2025.03 | 85.49 | — | |
| SimCLR (repro)Architecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 85.3 | — | |
| LBP-WHTBackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 85.1 | — | |
| LoRABackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 85.09 | — | |
| HOT + LoRABackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 85.01 | — | |
| FPBackbone=EfficientNetV2-m, Task=Fine-tuning2025.03 | 84.89 | — | |
| LoRABackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 84.84 | — | |
| DINOViT=B/16, Protocol=linear probing2026.05 | 84.8 | — | |
| SparseCBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 84.75 | — | |
| CoopCBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 84.66 | — | |
| VQAArch=ViT-g2025.12 | 84.5 | — | |
| TC-JEPAViT=B/16, Protocol=linear probing2026.05 | 84.5 | — | |
| LaBobackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 84.1 | — | |
| HOTBackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 84.07 | — | |
| HOTBackbone=EfficientNetV2-m, Task=Fine-tuning2025.03 | 83.91 | — | |
| HOT + LoRABackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 83.42 | — | |
| LBP-WHTBackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 83.27 | — | |
| VQAArch=ViT-L2025.12 | 83.2 | — | |
| MVT + FTArch=ViT-L2025.12 | 82 | — | |
| data2vecViT=L/16, Protocol=linear probing2026.05 | 81.6 | — | |
| CEMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 81.23 | — | |
| Vanilla FTArch=ViT-L2025.12 | 80.3 | — | |
| Random initArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 80.2 | — | |
| Vanilla-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 80.04 | — | |
| MVTArch=ViT-L2025.12 | 78.9 | — | |
| LUQBackbone=EfficientFormer-L1, Task=Fine-tuning2025.03 | 78.7 | — | |
| SDGM-FEvaluation Protocol=end-to-end pipeline2024.10 | 78.68 | — | |
| BYOLArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 78.4 | — | |
| Supervised-INArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 78.3 | — | |
| CLIPArch=ViT-L2025.12 | 78.2 | — | |
| MAEViT=H/14, Protocol=linear probing2026.05 | 77.3 | — | |
| Weight RecyclingK (shots per class)=42026.05 | 76.42 | — | |
| LUQBackbone=EfficientFormer-L7, Task=Fine-tuning2025.03 | 76.33 | — | |
| Weight RecyclingK (shots per class)=22026.05 | 75.77 | — | |
| Weight RecyclingK (shots per class)=12026.05 | 75.14 | — | |
| SimCLR (repro)Architecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 74.4 | — | |
| Noun SubmanifoldBackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 73.9 | — | |
| POS PGABackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 72.8 | — | |
| CLIPBackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 72.7 | — | |
| POS PCABackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 72.7 | — | |
| SimCLRArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 71.6 | — | |
| NCCK (shots per class)=42026.05 | 71.35 | — | |
| CLIPArch=ViT-B2025.12 | 65 | — | |
| KNNK (shots per class)=42026.05 | 62.41 | — | |
| NCCK (shots per class)=22026.05 | 61.64 | — | |
| Partial G-CNNsBase Group=E(2), No. Elems=162021.10 | 61.46 | — | |
| Partial G-CNNsBase Group=SE(2), No. Elems=162021.10 | 57.31 | — | |
| Partial G-CNNsBase Group=SE(2), No. Elems=82021.10 | 57.26 | — | |
| Partial G-CNNsBase Group=E(2), No. Elems=82021.10 | 55.22 | — | |
| LBP-WHTBackbone=EfficientNetV2-m, Task=Fine-tuning2025.03 | 55.08 | — | |
| AugerinoBase Group=E(2), No. Elems=162021.10 | 54.76 | — | |
| AugerinoBase Group=SE(2), No. Elems=82021.10 | 52.98 | — | |
| AugerinoBase Group=E(2), No. Elems=82021.10 | 52.59 | — | |
| AugerinoBase Group=SE(2), No. Elems=162021.10 | 51.47 | — | |
| NCCK (shots per class)=12026.05 | 49.98 | — | |
| KNNK (shots per class)=12026.05 | 49.98 | — | |
| CLIPArch=RN1012025.12 | 48.8 | — | |
| KNNK (shots per class)=22026.05 | 48.5 | — | |
| CLIPArch=RN502025.12 | 41.9 | — | |
| brSF+BN+BMFeedback=Biased Random Sign-concordant, Batch Normalization=true, Batch Manhattan=true2015.10 | — | 50.74 | |
| frSF+BN+BMFeedback=Fixed Random Sign-concordant, Batch Normalization=true, Batch Manhattan=true2015.10 | — | 52.25 | |
| uSF+BN+BMFeedback=Uniform Sign-concordant, Batch Normalization=true, Batch Manhattan=true2015.10 | — | 53.12 |