Generalized Zero-Shot Learning on SUN
69.6HZeroDiff++
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ZeroDiff++Venue=Ours, Backbone=Res101, Fine-tune features=our2026.02 | 69.6 | 73.4 | 66.2 | — | — | — | — | — | |
| DFCAFlowVenue=TCSVT23, Backbone=Res101, Fine-tune features=our2026.02 | 61.2 | 63 | 59.6 | — | — | — | — | — | |
| ZeroDiffVenue=ICLR25, Backbone=Res101, Fine-tune features=our2026.02 | 59.8 | 63 | 56.9 | — | — | — | — | — | |
| f-CLSWGANVenue=CVPR18, Backbone=Res101, Fine-tune features=our2026.02 | 59.5 | 63.8 | 55.7 | — | — | — | — | — | |
| SDVAEVenue=ICCV21, Backbone=Res101, Fine-tune features=our2026.02 | 59.5 | 62.3 | 56.9 | — | — | — | — | — | |
| f-VAEGANVenue=CVPR19, Backbone=Res101, Fine-tune features=our2026.02 | 58.4 | 63.6 | 54.1 | — | — | — | — | — | |
| CEGANVenue=CVPR21, Backbone=Res101, Fine-tune features=our2026.02 | 57.8 | 58.1 | 57.4 | — | — | — | — | — | |
| RLVCBackbone=ViT2026.03 | 57.6 | 59.6 | 55.6 | — | — | — | — | — | |
| EGVenue=TPAMI24, Backbone=ViT2026.02 | 57 | 54.7 | 59.4 | — | — | — | — | — | |
| APN+GDANLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 56.7 | 41.4 | 89.9 | — | — | — | — | — | |
| VADSVenue=CVPR24, Backbone=ViT, Fine-tune features=other2026.02 | 55.7 | 64.6 | 49 | — | — | — | — | — | |
| VADSBackbone=ViT, Method Category=Generative2026.03 | 55.7 | 64.6 | 49 | — | — | — | — | — | |
| IZF2021.03 | 54.8 | 52.7 | 57 | — | — | — | — | — | |
| PSVMA++Venue=TPAMI25, Backbone=ViT2026.02 | 54.8 | 61.5 | 49.4 | — | — | — | — | — | |
| PSVMA+Backbone=ViT, Method Category=Embedding2026.03 | 54.8 | 61.5 | 49.4 | — | — | — | — | — | |
| VSPCNBackbone=ViT, Method Category=Embedding2026.03 | 53.8 | 59.4 | 49.1 | — | — | — | — | — | |
| TF-VAEGANSetting=Transductive (TR)2020.03 | 53.7 | 62.4 | 47.1 | — | — | — | — | — | |
| GDAN*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 53.4 | 38.1 | 89.9 | — | — | — | — | — | |
| TF-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 53.3 | 44.3 | 66.9 | — | — | — | — | — | |
| TFVAEGANVenue=ECCV20, Backbone=Res101, Fine-tune features=our2026.02 | 52.8 | 51.8 | 53.8 | — | — | — | — | — | |
| PSVMAAttack=Clean2025.12 | 52.4 | 62.5 | 45.1 | — | — | — | 36 | 1.6 | |
| PSVMAVenue=CVPR23, Backbone=ViT2026.02 | 52.3 | 61.7 | 45.3 | — | — | — | — | — | |
| PSVMAAttack=clsA-1, Protocol=Best gamma2025.12 | 51.7 | 60.6 | 45 | — | — | — | 35.2 | 1.5 | |
| PSVMAAttack=clsA-1, Protocol=Original gamma2025.12 | 51.4 | 61.4 | 44.2 | — | — | — | 35.2 | 1.6 | |
| PSVMAAttack=CBEA-1, Protocol=Best gamma2025.12 | 51.4 | 59.3 | 45.3 | — | — | — | 34.7 | 1.4 | |
| AENetBackbone=ViT, Method Category=Embedding2026.03 | 51 | 58.6 | 45.2 | — | — | — | — | — | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=false2019.03 | 49.6 | 60.6 | 41.9 | — | — | — | — | — | |
| f-VAEGANSetting=Transductive (TR)2020.03 | 49.6 | 60.6 | 41.9 | — | — | — | — | — | |
| CoOp*Venue=IJCV22, Backbone=ViT2026.02 | 49.5 | 49.3 | 49.8 | — | — | — | — | — | |
| SBAR-TSetting=Fine-tuned Transductive (FT-TR)2020.03 | 48.6 | 58.8 | 41.5 | — | — | — | — | — | |
| CLIP*Venue=ICML21, Backbone=ViT2026.02 | 47.8 | 49.5 | 46.2 | — | — | — | — | — | |
| TPR*Venue=NeurIPS24, Backbone=ViT2026.02 | 47.8 | 45.4 | 50.4 | — | — | — | — | — | |
| ZeroMambaBackbone=VMamba, Method Category=Embedding2026.03 | 47.7 | 56.5 | 41.4 | — | — | — | — | — | |
| ZSLViTFeature Type=ViT2024.04 | 47.3 | 45.9 | 48.4 | — | — | — | — | — | |
| ZSLViTVenue=CVPR24, Backbone=ViT2026.02 | 47.3 | 45.9 | 48.4 | — | — | — | — | — | |
| ZSLViTBackbone=ViT, Method Category=Embedding2026.03 | 47.3 | 45.9 | 48.4 | — | — | — | — | — | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=true2019.03 | 47.2 | 54.2 | 41.8 | — | — | — | — | — | |
| f-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 47.2 | 54.2 | 41.8 | — | — | — | — | — | |
| TF-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 46.3 | 41.8 | 51.9 | — | — | — | — | — | |
| DUETVenue=AAAI'23, Feature Type=ViT2024.04 | 45.8 | 45.7 | 45.8 | — | — | — | — | — | |
| DUETBackbone=ViT, Method Category=Embedding2026.03 | 45.8 | 45.7 | 45.8 | — | — | — | — | — | |
| DSPVenue=ICML23, Backbone=Res101, Fine-tune features=other2026.02 | 45.5 | 48.3 | 43 | — | — | — | — | — | |
| DSPBackbone=ResNet, Method Category=Generative2026.03 | 45.5 | 48.3 | 43 | — | — | — | — | — | |
| ViFRBackbone=ResNet, Method Category=Generative2026.03 | 44.7 | 51.3 | 40 | — | — | — | — | — | |
| SC-EGGBackbone=ResNet, Method Category=Generative2026.03 | 44.3 | 45.1 | 43.6 | — | — | — | — | — | |
| ZeroNASBackbone=ResNet, Method Category=Generative2026.03 | 44.3 | 47.1 | 41.8 | — | — | — | — | — | |
| APN+f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=APN features2022.04 | 43.8 | 49.9 | 39.3 | — | — | — | — | — | |
| APN+TF-VAEGANLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 43.7 | 52.6 | 37.3 | — | — | — | — | — | |
| HSVAVenue=NeurIPS'21, Feature Type=CNN2024.04 | 43.3 | 48.6 | 39 | — | — | — | — | — | |
| HSVACategory=Common Space Learning2026.03 | 43.3 | 48.6 | 39 | — | — | — | — | — | |
| HSVABackbone=ResNet, Method Category=Generative2026.03 | 43.3 | 48.6 | 39 | — | — | — | — | — | |
| f-VAEGAN-D2Learning Setting=Inductive, Fine-tuned=true2019.03 | 43.1 | 50.1 | 37.8 | — | — | — | — | — | |
| f-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 43.1 | 50.1 | 37.8 | — | — | — | — | — | |
| CE-GZSL2021.03 | 43.1 | 48.8 | 38.6 | — | — | — | — | — | |
| f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 43.1 | 50.1 | 37.8 | — | — | — | — | — | |
| SE-GZSLVenue=AAAI'22, Feature Type=CNN2024.04 | 43.1 | 45.8 | 40.7 | — | — | — | — | — | |
| CE-GZSLBackbone=ResNet, Method Category=Generative2026.03 | 43.1 | 48.8 | 38.6 | — | — | — | — | — | |
| TF-VAEGANSetting=Inductive (IN)2020.03 | 43 | 45.6 | 40.7 | — | — | — | — | — | |
| TF-VAEGAN2021.03 | 43 | 45.6 | 40.7 | — | — | — | — | — | |
| TF-VAEGANVenue=ECCV'20, Feature Type=CNN2024.04 | 43 | 45.6 | 40.7 | — | — | — | — | — | |
| CDL + OSCOBackbone=ResNet, Method Category=Generative2026.03 | 42.9 | 32 | 65 | — | — | — | — | — | |
| TF-VAEGAN*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 42.8 | 45.3 | 40.7 | — | — | — | — | — | |
| VS-BoostVenue=IJCAI'23, Feature Type=CNN2024.04 | 42.5 | 49.2 | 37.4 | — | — | — | — | — | |
| TransZero++Venue=TPAMI22, Backbone=Res1012026.02 | 42.5 | 48.6 | 37.8 | — | — | — | — | — | |
| VS-BoostCategory=Generative Methods2026.03 | 42.5 | 49.2 | 37.4 | — | — | — | — | — | |
| TransZero++Backbone=ResNet, Method Category=Embedding2026.03 | 42.5 | 48.6 | 37.8 | — | — | — | — | — | |
| ZeroMambaAttack=Clean2025.12 | 42.4 | 47 | 38.6 | — | — | — | 25.9 | 0 | |
| GCM-CFVenue=CVPR'21, Feature Type=CNN2024.04 | 42.2 | 47.9 | 37.8 | — | — | — | — | — | |
| GCM-CFCategory=Generative Methods2026.03 | 42.2 | 47.9 | 37.8 | — | — | — | — | — | |
| MSDN++Category=Embedding-based Methods, Attention-based=true2026.03 | 42.1 | 51.9 | 35.4 | — | — | — | — | — | |
| RFF-GZSL2021.03 | 41.9 | 45.7 | 38.6 | — | — | — | — | — | |
| ZeroMambaAttack=clsA-1, Protocol=Best gamma2025.12 | 41.8 | 50.4 | 35.7 | — | — | — | 25.3 | 0 | |
| UE-finetuneLearning Setting=Transductive, Fine-tuned=true2019.03 | 41.7 | 33.6 | 54.8 | — | — | — | — | — | |
| UE-finetuneSetting=Fine-tuned Transductive (FT-TR)2020.03 | 41.7 | 33.6 | 54.8 | — | — | — | — | — | |
| FREELearning Paradigm=Generative2022.04 | 41.7 | 47.4 | 37.2 | — | — | — | — | — | |
| FREEVenue=ICCV'21, Feature Type=CNN2024.04 | 41.7 | 47.4 | 37.2 | — | — | — | — | — | |
| ZeroMambaAttack=CBEA-1, Protocol=Best gamma2025.12 | 41.7 | 50.9 | 35.4 | — | — | — | 25.3 | 0 | |
| FREECategory=Generative Methods2026.03 | 41.7 | 47.4 | 37.2 | — | — | — | — | — | |
| APN+ABPLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 41.6 | 46.2 | 37.8 | — | — | — | — | — | |
| ZeroMambaAttack=clsA-1, Protocol=Original gamma2025.12 | 41.6 | 46.2 | 37.8 | — | — | — | 25.3 | 0 | |
| SBAR-ISetting=Fine-tuned Inductive (FT-IN)2020.03 | 41.5 | 50.7 | 35.1 | — | — | — | — | — | |
| FREE+ESZSLCategory=Generative Methods2026.03 | 41.5 | 48.2 | 36.5 | — | — | — | — | — | |
| FREE + ESZSLBackbone=ResNet, Method Category=Generative2026.03 | 41.5 | 48.2 | 36.5 | — | — | — | — | — | |
| f-VAEGAN-D2Learning Setting=Inductive, Fine-tuned=false2019.03 | 41.3 | 45.1 | 38 | — | — | — | — | — | |
| f-VAEGANSetting=Inductive (IN)2020.03 | 41.3 | 45.1 | 38 | — | — | — | — | — | |
| f-VAEGAN-D22021.03 | 41.3 | 45.1 | 38 | — | — | — | — | — | |
| f-VAEGANVenue=CVPR'19, Feature Type=CNN2024.04 | 41.3 | 45.1 | 38 | — | — | — | — | — | |
| MSDNVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 41.3 | 52.2 | 34.2 | — | — | — | — | — | |
| f-VAEGAN-D2Category=Generative Methods2026.03 | 41.3 | 45.1 | 38 | — | — | — | — | — | |
| MSDNCategory=Embedding-based Methods, Attention-based=true2026.03 | 41.3 | 52.2 | 34.2 | — | — | — | — | — | |
| ABP*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 41.2 | 43.3 | 39.3 | — | — | — | — | — | |
| APN+CVCLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 41.2 | 37.9 | 45.2 | — | — | — | — | — | |
| GNDANCategory=Embedding-based Methods, Attention-based=true2026.03 | 41 | 50 | 34.7 | — | — | — | — | — | |
| LsrGANVenue=ECCV'20, Feature Type=CNN2024.04 | 40.9 | 44.8 | 37.7 | — | — | — | — | — | |
| ViFRCategory=Generative Methods2026.03 | 40.9 | 48.8 | 35.2 | — | — | — | — | — | |
| TransZeroVenue=AAAI'22, Feature Type=CNN, Attention=true2024.04 | 40.8 | 52.6 | 33.4 | — | — | — | — | — | |
| TransZeroCategory=Embedding-based Methods, Attention-based=true2026.03 | 40.8 | 52.6 | 33.4 | — | — | — | — | — | |
| HASBackbone=ResNet, Method Category=Embedding2026.03 | 40.8 | 42.8 | 38.9 | — | — | — | — | — | |
| CADA-VAEFeature Size=642018.12 | 40.6 | 47.2 | 35.7 | — | — | — | — | — | |
| Zhu et al.2021.03 | 40.6 | 45.3 | 36.8 | — | — | — | — | — |