Generalized Zero-Shot Learning on CUB
344H ScoreALE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ALE2019.10 | 344 | 237 | 62.8 | — | — | — | |
| DEVISE2019.10 | 328 | 238 | 53 | — | — | — | |
| CJME2019.10 | 301 | 356 | 26.1 | — | — | — | |
| ESZSL2019.10 | 210 | 126 | 63.8 | — | — | — | |
| SAE2019.10 | 136 | 78 | 54 | — | — | — | |
| ZeroDiff++Venue=Ours, Backbone=Res101, Fine-tune features=our2026.02 | 85.4 | 84.4 | 86.4 | — | — | — | |
| ZeroDiffVenue=ICLR25, Backbone=Res101, Fine-tune features=our2026.02 | 81.6 | 80 | 83.2 | — | — | — | |
| RLVCBackbone=ViT2026.03 | 81.2 | 80.9 | 81.4 | — | — | — | |
| TFVAEGANVenue=ECCV20, Backbone=Res101, Fine-tune features=our2026.02 | 81.1 | 79 | 83.3 | — | — | — | |
| TF-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 80.9 | 78.4 | 83.5 | — | — | — | |
| f-VAEGANVenue=CVPR19, Backbone=Res101, Fine-tune features=our2026.02 | 80.3 | 77.4 | 83.5 | — | — | — | |
| DFCAFlowVenue=TCSVT23, Backbone=Res101, Fine-tune features=our2026.02 | 80 | 77.3 | 82.9 | — | — | — | |
| CEGANVenue=CVPR21, Backbone=Res101, Fine-tune features=our2026.02 | 79.9 | 78.9 | 80.9 | — | — | — | |
| f-CLSWGANVenue=CVPR18, Backbone=Res101, Fine-tune features=our2026.02 | 79.7 | 76.4 | 83.3 | — | — | — | |
| SDVAEVenue=ICCV21, Backbone=Res101, Fine-tune features=our2026.02 | 77.7 | 81.6 | 74.2 | — | — | — | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=true2019.03 | 77.3 | 73.8 | 81.4 | — | — | — | |
| f-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 77.3 | 73.8 | 81.4 | — | — | — | |
| VSPCNBackbone=ViT, Method Category=Embedding2026.03 | 75.7 | 72.8 | 78.9 | — | — | — | |
| AENetBackbone=ViT, Method Category=Embedding2026.03 | 74.7 | 73.1 | 76.4 | — | — | — | |
| PSVMA++Venue=TPAMI25, Backbone=ViT2026.02 | 74.6 | 71.8 | 77.8 | — | — | — | |
| PSVMA+Backbone=ViT, Method Category=Embedding2026.03 | 74.6 | 71.8 | 77.8 | — | — | — | |
| VADSVenue=CVPR24, Backbone=ViT, Fine-tune features=other2026.02 | 74.3 | 74.1 | 74.6 | — | — | — | |
| VADSBackbone=ViT, Method Category=Generative2026.03 | 74.3 | 74.1 | 74.6 | — | — | — | |
| ZeroMambaBackbone=VMamba, Method Category=Embedding2026.03 | 74.2 | 72.1 | 76.4 | — | — | — | |
| PSVMAAttack=Clean2025.12 | 74 | 72.8 | 75.2 | — | 65.8 | 1.6 | |
| ReZSLVenue=TIP23, Backbone=Res1012026.02 | 73.8 | 72.8 | 74.8 | — | — | — | |
| PSVMAVenue=CVPR23, Backbone=ViT2026.02 | 73.8 | 70.1 | 77.8 | — | — | — | |
| PSVMAAttack=clsA-1, Protocol=Best gamma2025.12 | 73.7 | 72.3 | 75.1 | — | 65 | 1.6 | |
| PSVMAAttack=CBEA-1, Protocol=Best gamma2025.12 | 73.7 | 72 | 75.5 | — | 65.2 | 1.6 | |
| ZSLViTFeature Type=ViT2024.04 | 73.6 | 69.4 | 78.2 | — | — | — | |
| ZSLViTVenue=CVPR24, Backbone=ViT2026.02 | 73.6 | 69.4 | 78.2 | — | — | — | |
| ZSLViTBackbone=ViT, Method Category=Embedding2026.03 | 73.6 | 69.4 | 78.2 | — | — | — | |
| PSVMAAttack=clsA-1, Protocol=Original gamma2025.12 | 73.5 | 72.6 | 74.5 | — | 65 | 1.6 | |
| UE-finetuneLearning Setting=Transductive, Fine-tuned=true2019.03 | 73.2 | 74.9 | 71.5 | — | — | — | |
| ReZSLAttack=Clean2025.12 | 72.9 | 71.9 | 73.8 | — | 64.3 | 1.2 | |
| UE-finetuneSetting=Fine-tuned Transductive (FT-TR)2020.03 | 72.3 | 71.5 | 73.2 | — | — | — | |
| ReZSLAttack=clsA-1, Protocol=Original gamma2025.12 | 71.9 | 70.8 | 73.9 | — | 63.2 | 1.2 | |
| ReZSLAttack=clsA-1, Protocol=Best gamma2025.12 | 71.9 | 69.1 | 75 | — | 63.2 | 1.1 | |
| ReZSLAttack=CBEA-12025.12 | 71.9 | 69.1 | 75 | — | 63.2 | 1.1 | |
| HASBackbone=ResNet, Method Category=Embedding2026.03 | 71.8 | 69.6 | 74.1 | — | — | — | |
| TF-VAEGANSetting=Transductive (TR)2020.03 | 71 | 69.9 | 72.1 | — | — | — | |
| TF-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 70.7 | 63.8 | 79.3 | — | — | — | |
| APN+TF-VAEGANLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 70.6 | 65.6 | 76.3 | — | — | — | |
| MSDN++Category=Embedding-based Methods, Attention-based=true2026.03 | 70.6 | 70.8 | 70.3 | — | — | — | |
| GEM-ZSLVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 70.4 | 64.8 | 77.1 | — | — | — | |
| TransZero++Venue=TPAMI22, Backbone=Res1012026.02 | 70.4 | 67.5 | 73.6 | — | — | — | |
| TransZero++Backbone=ResNet, Method Category=Embedding2026.03 | 70.4 | 67.5 | 73.6 | — | — | — | |
| SBAR-TSetting=Fine-tuned Transductive (FT-TR)2020.03 | 70.3 | 67.2 | 73.7 | — | — | — | |
| APN+f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=APN features2022.04 | 70.2 | 65.5 | 75.6 | — | — | — | |
| TF-VAEGAN*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 70 | 65.5 | 75.1 | — | — | — | |
| APN+ABPLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 69.8 | 65.5 | 74.6 | — | — | — | |
| APNLearning Paradigm=Non-generative, ZoomInMod=true, Calibrated Stacking=true2022.04 | 69.4 | 67.4 | 71.6 | — | — | — | |
| GNDANCategory=Embedding-based Methods, Attention-based=true2026.03 | 69.4 | 69.2 | 69.6 | — | — | — | |
| APNCategory=Embedding-based Methods, Attention-based=true2026.03 | 69.4 | 67.4 | 71.6 | — | — | — | |
| f-VAEGAN-D2Learning Setting=Inductive, Fine-tuned=true2019.03 | 68.9 | 63.2 | 75.6 | — | — | — | |
| f-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 68.9 | 63.2 | 75.6 | — | — | — | |
| f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 68.9 | 63.2 | 75.6 | — | — | — | |
| TransZeroVenue=AAAI'22, Feature Type=CNN, Attention=true2024.04 | 68.8 | 69.3 | 68.3 | — | — | — | |
| TransZeroCategory=Embedding-based Methods, Attention-based=true2026.03 | 68.8 | 69.3 | 68.3 | — | — | — | |
| SC-EGGBackbone=ResNet, Method Category=Generative2026.03 | 68.5 | 64.1 | 73.6 | — | — | — | |
| VS-BoostVenue=IJCAI'23, Feature Type=CNN2024.04 | 68.4 | 68 | 68.7 | — | — | — | |
| VS-BoostCategory=Generative Methods2026.03 | 68.4 | 68 | 68.7 | — | — | — | |
| MSDNVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 68.1 | 68.7 | 67.5 | — | — | — | |
| MSDNCategory=Embedding-based Methods, Attention-based=true2026.03 | 68.1 | 68.7 | 67.5 | — | — | — | |
| APN+CVCLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 67.7 | 62 | 74.5 | — | — | — | |
| ViFRBackbone=ResNet, Method Category=Generative2026.03 | 67.6 | 63.9 | 72 | — | — | — | |
| DUETVenue=AAAI'23, Feature Type=ViT2024.04 | 67.5 | 62.9 | 72.8 | — | — | — | |
| DUETBackbone=ViT, Method Category=Embedding2026.03 | 67.5 | 62.9 | 72.8 | — | — | — | |
| APN+GDANLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 67.3 | 67.9 | 66.7 | — | — | — | |
| APN+Learning Paradigm=Non-generative, ZoomInMod=false2022.04 | 67.2 | 65.3 | 69.3 | — | — | — | |
| APNVenue=NeurIPS'20, Feature Type=CNN, Attention=true2024.04 | 67.2 | 65.3 | 69.3 | — | — | — | |
| APNVenue=NeurIPS20, Backbone=Res1012026.02 | 67.2 | 65.3 | 69.3 | — | — | — | |
| CVC*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 67 | 61.1 | 74.2 | — | — | — | |
| ABP*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 66.8 | 61.6 | 73 | — | — | — | |
| ICCEBackbone=ResNet, Method Category=Generative2026.03 | 66.4 | 67.3 | 65.5 | — | — | — | |
| STHS-GZSLmode=Transductive, sampling=Hardness Sampling2021.06 | 66.2 | 65.1 | 67.4 | — | — | — | |
| GDAN*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 66.2 | 65.7 | 66.7 | — | — | — | |
| ARENLearning Paradigm=Non-generative, Calibrated Stacking=true2022.04 | 66 | 63.2 | 69 | — | — | — | |
| ARENBackbone=ResNet-101, Split setting=PS, Post-processing=Calibrated Stacking2022.03 | 66 | 63.2 | 69 | — | — | — | |
| EGVenue=TPAMI24, Backbone=ViT2026.02 | 66 | 65.3 | 66.7 | — | — | — | |
| EG-part-netCategory=Embedding-based Methods2026.03 | 65.4 | 64.8 | 66.1 | — | — | — | |
| CE-GZSL2021.03 | 65.3 | 63.9 | 66.8 | — | — | — | |
| CE-GZSLBackbone=ResNet, Method Category=Generative2026.03 | 65.3 | 63.9 | 66.8 | — | — | — | |
| ZeroMambaAttack=Clean2025.12 | 65.2 | 64.3 | 66 | — | 55.5 | 0 | |
| EGGCategory=Generative Methods2026.03 | 64.7 | 58.6 | 72.3 | — | — | — | |
| ZeroMambaAttack=clsA-1, Protocol=Original gamma2025.12 | 64 | 62.9 | 65.2 | — | 54 | 0 | |
| ZeroMambaAttack=clsA-1, Protocol=Best gamma2025.12 | 64 | 62.9 | 65.2 | — | 54 | 0 | |
| ZeroMambaAttack=CBEA-1, Protocol=Best gamma2025.12 | 64 | 65.9 | 62.3 | — | 54.2 | 0 | |
| ST-WGAN-RSmode=Transductive, sampling=Random Sampling2021.06 | 63.9 | 59.7 | 68.8 | — | — | — | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=false2019.03 | 63.2 | 61.4 | 65.1 | — | — | — | |
| f-VAEGANSetting=Transductive (TR)2020.03 | 63.2 | 61.4 | 65.1 | — | — | — | |
| SDGZSLNumber of seen class samples=All2022.03 | 63.2 | 61.2 | 65.3 | — | — | — | |
| GCM-CFVenue=CVPR'21, Feature Type=CNN2024.04 | 60.3 | 61 | 59.7 | — | — | — | |
| GCM-CFCategory=Generative Methods2026.03 | 60.3 | 61 | 59.7 | — | — | — | |
| ViFRCategory=Generative Methods2026.03 | 60.1 | 57.8 | 62.7 | — | — | — | |
| ComposerVenue=NeurIPS'20, Feature Type=CNN, Attention=true2024.04 | 59.9 | 56.4 | 63.8 | — | — | — | |
| ComposerCategory=Generative Methods, Attention-based=true2026.03 | 59.9 | 56.4 | 63.8 | — | — | — | |
| ZeroNASBackbone=ResNet, Method Category=Generative2026.03 | 59.6 | 56 | 63.8 | — | — | — | |
| IZF2021.03 | 59.4 | 52.7 | 68 | — | — | — | |
| TF-VAEGANSetting=Inductive (IN)2020.03 | 58.1 | 52.8 | 64.7 | — | — | — |