Zero-Shot Learning on SUN
80.2Top-1 AccuracyZeroDiff++
Evaluation Results
| Method | Links | |
|---|---|---|
| ZeroDiff++Venue=Ours, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 80.2 | |
| UE-finetuneSetting=Fine-tuned Transductive (FT-TR)2020.03 | 79.7 | |
| ZeroDiffVenue=ICLR25, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 77.3 | |
| DFCAFlowVenue=TCSVT23, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 77.2 | |
| SDVAEVenue=ICCV21, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 77 | |
| VADSVenue=CVPR24, Backbone=ViT, Fine-tune features=Other fine-tune features2026.02 | 76.3 | |
| f-CLSWGANVenue=CVPR18, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 75.5 | |
| f-VAEGANVenue=CVPR19, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 75.4 | |
| PSVMA++Venue=TPAMI25, Backbone=ViT, Fine-tune features=None2026.02 | 74.5 | |
| TFVAEGANVenue=ECCV20, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 74.1 | |
| CEGANVenue=CVPR21, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 74.1 | |
| TF-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 73.8 | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=true2019.03 | 72.6 | |
| f-VAEGANSetting=Fine-tuned Transductive (FT-TR)2020.03 | 72.6 | |
| TF-VAEGANSetting=Transductive (TR)2020.03 | 70.9 | |
| f-VAEGAN-D2Learning Setting=Transductive, Fine-tuned=false2019.03 | 70.1 | |
| f-VAEGANSetting=Transductive (TR)2020.03 | 70.1 | |
| ZSLViTFeature Type=ViT2024.04 | 68.3 | |
| ZSLViTVenue=CVPR24, Backbone=ViT, Fine-tune features=None2026.02 | 68.3 | |
| TransZero++Venue=TPAMI22, Backbone=Res101, Fine-tune features=None2026.02 | 67.6 | |
| SBAR-TSetting=Fine-tuned Transductive (FT-TR)2020.03 | 67.5 | |
| DiffusionZSLVenue=ICCV23, Backbone=Res101, Fine-tune features=Author's fine-tune features2026.02 | 67.5 | |
| MSDN++Category=Embedding-based Methods, Attention-based=true2026.03 | 67.5 | |
| TF-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 66.7 | |
| TF-VAEGANSetting=Inductive (IN)2020.03 | 66 | |
| TF-VAEGAN2021.03 | 66 | |
| TF-VAEGANVenue=ECCV'20, Feature Type=CNN2024.04 | 66 | |
| APN+f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=APN features2022.04 | 65.9 | |
| MSDNVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 65.8 | |
| MSDNCategory=Embedding-based Methods, Attention-based=true2026.03 | 65.8 | |
| f-VAEGAN-D2Learning Setting=Inductive, Fine-tuned=true2019.03 | 65.6 | |
| f-VAEGANSetting=Fine-tuned Inductive (FT-IN)2020.03 | 65.6 | |
| f-VAEGAN-D2*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 65.6 | |
| TransZeroVenue=AAAI'22, Feature Type=CNN, Attention=true2024.04 | 65.6 | |
| ViFRCategory=Generative Methods2026.03 | 65.6 | |
| TransZeroCategory=Embedding-based Methods, Attention-based=true2026.03 | 65.6 | |
| APN+TF-VAEGANLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 65.5 | |
| TF-VAEGAN*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 65.4 | |
| GNDANCategory=Embedding-based Methods, Attention-based=true2026.03 | 65.3 | |
| SBAR-ISetting=Fine-tuned Inductive (FT-IN)2020.03 | 65.2 | |
| f-VAEGAN-D2Learning Setting=Inductive, Fine-tuned=false2019.03 | 64.7 | |
| f-VAEGANSetting=Inductive (IN)2020.03 | 64.7 | |
| f-VAEGAN-D22021.03 | 64.7 | |
| f-VAEGANVenue=CVPR'19, Feature Type=CNN2024.04 | 64.7 | |
| f-VAEGAN-D2Category=Generative Methods2026.03 | 64.7 | |
| DUETVenue=AAAI'23, Feature Type=ViT2024.04 | 64.4 | |
| GFZSLLearning Setting=Transductive, Fine-tuned=false2019.03 | 64 | |
| GFZSLSetting=Transductive (TR)2020.03 | 64 | |
| HSVAVenue=NeurIPS'21, Feature Type=CNN2024.04 | 63.8 | |
| HSVACategory=Common Space Learning2026.03 | 63.8 | |
| SE-GZSLLearning Setting=Inductive, Fine-tuned=false2019.03 | 63.4 | |
| APN+ABPLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 63.4 | |
| SEAdditional synthetic data=true2019.04 | 63.4 | |
| CE-GZSL2021.03 | 63.3 | |
| ReZSLVenue=TIP23, Backbone=Res101, Fine-tune features=None2026.02 | 63 | |
| GEM-ZSLVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 62.8 | |
| ABP*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 62.6 | |
| ComposerVenue=NeurIPS'20, Feature Type=CNN, Attention=true2024.04 | 62.6 | |
| ComposerCategory=Generative Methods, Attention-based=true2026.03 | 62.6 | |
| ALEFeature Generator=f-CLSWGAN2017.12 | 62.1 | |
| DEMAdditional synthetic data=false2019.04 | 61.9 | |
| DCN2021.03 | 61.8 | |
| DCNCategory=Common Space Learning2026.03 | 61.8 | |
| LisGANSetting=Inductive (IN)2020.03 | 61.7 | |
| LisGANLearning Paradigm=Generative2022.04 | 61.7 | |
| CADA-VAECategory=Common Space Learning2026.03 | 61.7 | |
| APN+Learning Paradigm=Non-generative, ZoomInMod=false2022.04 | 61.6 | |
| APNVenue=NeurIPS'20, Feature Type=CNN, Attention=true2024.04 | 61.6 | |
| APNVenue=NeurIPS20, Backbone=Res101, Fine-tune features=None2026.02 | 61.6 | |
| TCNSetting=Inductive (IN)2020.03 | 61.5 | |
| LFGAA2021.03 | 61.5 | |
| Zhu et al.2021.03 | 61.5 | |
| TCN2021.03 | 61.5 | |
| LFGAA+HybridLearning Paradigm=Non-generative2022.04 | 61.5 | |
| APNLearning Paradigm=Non-generative, ZoomInMod=true2022.04 | 61.5 | |
| LFGAACategory=Embedding-based Methods, Attention-based=true2026.03 | 61.5 | |
| APNCategory=Embedding-based Methods, Attention-based=true2026.03 | 61.5 | |
| LATEMFeature Generator=f-CLSWGAN2017.12 | 61.3 | |
| DLFZRL2021.03 | 61.3 | |
| CVC*Learning Paradigm=Generative, Backbone Features=ResNet101 (finetuned)2022.04 | 61 | |
| DEVISEFeature Generator=f-CLSWGAN2017.12 | 60.9 | |
| TAFE-NetAdditional synthetic data=false2019.04 | 60.9 | |
| SoftmaxFeature Generator=f-CLSWGAN2017.12 | 60.8 | |
| CLSWGANLearning Setting=Inductive, Fine-tuned=false2019.03 | 60.8 | |
| f-CLSWGANSetting=Inductive (IN)2020.03 | 60.8 | |
| f-CLSWGAN2021.03 | 60.8 | |
| CLSWGANLearning Paradigm=Generative2022.04 | 60.8 | |
| f-CLSWGANAdditional synthetic data=true2019.04 | 60.8 | |
| f-CLSWGANCategory=Generative Methods2026.03 | 60.8 | |
| ARENLearning Paradigm=Non-generative2022.04 | 60.6 | |
| APN+CVCLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 60.6 | |
| ARENVenue=CVPR'19, Feature Type=CNN, Attention=true2024.04 | 60.6 | |
| ARENCategory=Embedding-based Methods, Attention-based=true2026.03 | 60.6 | |
| cycle-CLSWGAN2021.03 | 60 | |
| Cycle-CLSWGANLearning Setting=Inductive, Fine-tuned=false2019.03 | 59.9 | |
| Cycle-WGANSetting=Inductive (IN)2020.03 | 59.9 | |
| DAZLEVenue=CVPR'20, Feature Type=CNN, Attention=true2024.04 | 59.4 | |
| DAZLECategory=Embedding-based Methods, Attention-based=true2026.03 | 59.4 | |
| SP-AEN2021.03 | 59.2 | |
| SP-AENAdditional synthetic data=true2019.04 | 59.2 |