Classification on CUB
91.9AccuracyBlack-box
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Black-box2026.02 | 91.9 | — | — | — | — | — | |
| OTF-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 89.92 | — | — | — | — | — | |
| MMAL-NetTrain Anno.=false2020.03 | 89.6 | — | — | — | — | — | |
| SupervisedBackbone=ResNet-50, Pre-trained=ImageNet, Shot=5-shot, Way=5-way, Number of tasks=6002023.03 | 89.13 | — | — | — | — | — | |
| TASNTrain Anno.=false2020.03 | 87.9 | — | — | — | — | — | |
| DCLTrain Anno.=false2020.03 | 87.8 | — | — | — | — | — | |
| MetaFGNetTrain Anno.=false2020.03 | 87.6 | — | — | — | — | — | |
| HSnetTrain Anno.=true2020.03 | 87.5 | — | — | — | — | — | |
| NTS-NetTrain Anno.=false2020.03 | 87.5 | — | — | — | — | — | |
| DFL-CNNTrain Anno.=false2020.03 | 87.4 | — | — | — | — | — | |
| Mask-CNNTrain Anno.=true2020.03 | 87.3 | — | — | — | — | — | |
| HBPTrain Anno.=false2020.03 | 87.1 | — | — | — | — | — | |
| PCTrain Anno.=false2020.03 | 86.9 | — | — | — | — | — | |
| MA-CNNTrain Anno.=false2020.03 | 86.5 | — | — | — | — | — | |
| OSME+MAMCTrain Anno.=false2020.03 | 86.5 | — | — | — | — | — | |
| KPTrain Anno.=false2020.03 | 86.2 | — | — | — | — | — | |
| DOT-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 85.39 | — | — | — | — | — | |
| RA-CNNTrain Anno.=false2020.03 | 85.3 | — | — | — | — | — | |
| SPDA-CNNTrain Anno.=true2020.03 | 85.1 | — | — | — | — | — | |
| ResNet-50Train Anno.=false2020.03 | 84.5 | — | — | — | — | — | |
| Bilinear-CNNTrain Anno.=false2020.03 | 84.1 | — | — | — | — | — | |
| BYOL + PsCoBackbone=ResNet-50, Pre-trained=ImageNet, Shot=5-shot, Way=5-way, Number of tasks=6002023.03 | 82.13 | — | — | — | — | — | |
| CoopCBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 82.1 | — | — | — | — | — | |
| SparseCBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 82.07 | — | — | — | — | — | |
| LaBobackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 81.3 | — | — | — | — | — | |
| CEMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 80.47 | — | — | — | — | — | |
| Black boxinterpretable=false2026.03 | 80 | — | — | — | — | — | |
| LF-CBM2026.03 | 80 | — | — | — | — | — | |
| EQ-CBM2024.09 | 79.31 | — | — | — | — | — | |
| Coop-CBM2024.09 | 79.154 | — | — | — | — | — | |
| CEM2024.09 | 79.029 | — | — | — | — | — | |
| Vanilla-CBMbackbone=ViT, concept_embeddings=Label-Free CBM, training_strategy=independent2026.06 | 78.32 | — | — | — | — | — | |
| IntCEMRegularizer=IBB2026.02 | 77.8 | — | 0.957 | — | — | — | |
| AR-CBMRegularizer=IBB2026.02 | 77.7 | — | 0.961 | — | — | — | |
| AR-CBMRegularizer=IBE2026.02 | 77.4 | — | 0.95 | — | — | — | |
| IntCEMRegularizer=IBE2026.02 | 77.3 | — | 0.959 | — | — | — | |
| ECBM2024.09 | 77.148 | — | — | — | — | — | |
| CEM2022.09 | 77.11 | — | — | — | — | — | |
| MoCo v2 + PsCoBackbone=ResNet-50, Pre-trained=ImageNet, Shot=5-shot, Way=5-way, Number of tasks=6002023.03 | 76.63 | — | — | — | — | — | |
| Prob-CBM2024.09 | 76.265 | — | — | — | — | — | |
| AR-CBMRegularizer=None2026.02 | 76.1 | — | 0.956 | — | — | — | |
| CEM2026.03 | 76 | — | — | — | — | — | |
| IntCEMRegularizer=None2026.02 | 75.9 | — | 0.954 | — | — | — | |
| Fuzzy-CBM2024.09 | 74.228 | — | — | — | — | — | |
| ProbCBMRegularizer=IBB2026.02 | 74.1 | — | 0.96 | — | — | — | |
| HiCEM + Concept Splitting2026.03 | 74 | — | — | — | — | — | |
| No concepts2022.09 | 73.41 | — | — | — | — | — | |
| ProbCBMRegularizer=IBE2026.02 | 73.4 | — | 0.962 | — | — | — | |
| PCBM-h2026.03 | 73 | — | — | — | — | — | |
| Deep-HiCEM + MLCS2026.03 | 73 | — | — | — | — | — | |
| Fuzzy-CBM2022.09 | 72.98 | — | — | — | — | — | |
| CBM (SJ)Regularizer=IBB2026.02 | 72.7 | — | 0.961 | — | — | — | |
| CBM (SJ)Regularizer=IBE2026.02 | 72.6 | — | 0.958 | — | — | — | |
| Bool-CBM2024.09 | 72.512 | — | — | — | — | — | |
| ProbCBMRegularizer=None2026.02 | 71.8 | — | 0.956 | — | — | — | |
| LAION-CLIP ViT-LN=400M, Zero-shot=true2024.09 | 71.4 | — | — | — | — | — | |
| CBM (SJ)Regularizer=None2026.02 | 70.8 | — | 0.956 | — | — | — | |
| Hybrid-CBM2022.09 | 70.7 | — | — | — | — | — | |
| CBM (HI)Regularizer=IBE2026.02 | 69.1 | — | 0.952 | — | — | — | |
| CBM (HI)Regularizer=IBB2026.02 | 68.9 | — | 0.954 | — | — | — | |
| BYOLBackbone=ResNet-50, Pre-trained=ImageNet, Shot=5-shot, Way=5-way, Number of tasks=6002023.03 | 67.45 | — | — | — | — | — | |
| Boolean-CBM2022.09 | 67.11 | — | — | — | — | — | |
| DINOv2-ARLN=20M, Zero-shot=true2024.09 | 66.1 | — | — | — | — | — | |
| CBM (HJ)Regularizer=IBE2026.02 | 65.5 | — | 0.951 | — | — | — | |
| CBM (HJ)Regularizer=IBB2026.02 | 65.4 | — | 0.951 | — | — | — | |
| CBM (HJ)Regularizer=None2026.02 | 65 | — | 0.956 | — | — | — | |
| PCBM2026.03 | 65 | — | — | — | — | — | |
| CBM2026.03 | 65 | — | — | — | — | — | |
| CBM (SI)Regularizer=IBE2026.02 | 64.6 | — | 0.958 | — | — | — | |
| CBM (HI)Regularizer=None2026.02 | 64.4 | — | 0.956 | — | — | — | |
| CBM (SI)Regularizer=IBB2026.02 | 64.3 | — | 0.95 | — | — | — | |
| MoCo v2Backbone=ResNet-50, Pre-trained=ImageNet, Shot=5-shot, Way=5-way, Number of tasks=6002023.03 | 64.16 | — | — | — | — | — | |
| CBM (SI)Regularizer=None2026.02 | 63.9 | — | 0.956 | — | — | — | |
| DINOv2-MpNetN=20M, Zero-shot=true2024.09 | 63.2 | — | — | — | — | — | |
| OpenAI-CLIP ViT-LN=400M, Zero-shot=true2024.09 | 61.4 | — | — | — | — | — | |
| LiT L16LN=112M, Zero-shot=true2024.09 | 58.7 | — | — | — | — | — | |
| BiFTABackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 58.24 | — | — | — | — | — | |
| WCABackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 56.74 | — | — | — | — | — | |
| BiFTAZero-shot=true, Backbone=Average over five CLIP variants2026.01 | 56.05 | — | — | — | — | — | |
| WCAZero-shot=true, Backbone=Average over five CLIP variants2026.01 | 55.31 | — | — | — | — | — | |
| CLIP-EBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 52.74 | — | — | — | — | — | |
| CLIP-DBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 52.69 | — | — | — | — | — | |
| WaffleBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 52.04 | — | — | — | — | — | |
| CLIPBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 51.21 | — | — | — | — | — | |
| BiFTABackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 50.43 | — | — | — | — | — | |
| WCABackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 50.16 | — | — | — | — | — | |
| CuPLBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.01 | 49.76 | — | — | — | — | — | |
| CuPLBackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 47.91 | — | — | — | — | — | |
| CLIP-DBackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 47.76 | — | — | — | — | — | |
| WaffleBackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 46.76 | — | — | — | — | — | |
| CLIP-EBackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 46.58 | — | — | — | — | — | |
| CLIPBackbone=RN-50, Zero-shot evaluation protocol=true2026.01 | 45.67 | — | — | — | — | — | |
| MetaMAEPretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 6.3 | — | — | — | — | — | |
| e-MixPretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 3.9 | — | — | — | — | — | |
| CapriPretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 3.7 | — | — | — | — | — | |
| MAEPretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 3.1 | — | — | — | — | — | |
| ShEDPretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 3 | — | — | — | — | — | |
| LiTDA-baseN=0.5M, Zero-shot=true2024.09 | 2.3 | — | — | — | — | — | |
| BaselinePretrain data=ImageNet32, Protocol=Linear evaluation2023.10 | 1.6 | — | — | — | — | — | |
| LiTLwA-baseN=0.5M, Zero-shot=true2024.09 | 1.4 | — | — | — | — | — |