Generalized Zero-Shot Learning on AWA2
80.4H ScoreRLVC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLVCBackbone=ViT2026.03 | 80.4 | 82.4 | 78.4 | — | — | |
| PSVMA+Backbone=ViT, Method Category=Embedding2026.03 | 79.8 | 86.4 | 74.2 | — | — | |
| VADSBackbone=ViT, Method Category=Generative2026.03 | 79.3 | 83.6 | 75.4 | — | — | |
| VSPCNBackbone=ViT, Method Category=Embedding2026.03 | 77.6 | 84.3 | 71.8 | — | — | |
| ZeroMambaBackbone=VMamba, Method Category=Embedding2026.03 | 76.5 | 87.6 | 67.9 | — | — | |
| I2DFormer+Backbone=ViT, Method Category=Embedding2026.03 | 75.9 | 83.2 | 69.8 | — | — | |
| Unified feature-generative frameworkEvaluation Setting=Online-CGZSL2022.03 | 75.44 | 80.06 | 72.62 | — | — | |
| AENetBackbone=ViT, Method Category=Embedding2026.03 | 74.9 | 80.1 | 70.3 | — | — | |
| ZSLViTBackbone=ViT, Method Category=Embedding2026.03 | 74.2 | 84.6 | 66.1 | — | — | |
| HASBackbone=ResNet, Method Category=Embedding2026.03 | 73.3 | 87.3 | 63.1 | — | — | |
| ViFRBackbone=ResNet, Method Category=Generative2026.03 | 73.2 | 78.9 | 68.2 | — | — | |
| ICCEBackbone=ResNet, Method Category=Generative2026.03 | 72.8 | 82.3 | 65.3 | — | — | |
| DUETBackbone=ViT, Method Category=Embedding2026.03 | 72.7 | 84.7 | 63.7 | — | — | |
| TransZero++Backbone=ResNet, Method Category=Embedding2026.03 | 72.5 | 82.7 | 64.6 | — | — | |
| SC-EGGBackbone=ResNet, Method Category=Generative2026.03 | 72.4 | 89.3 | 60.9 | — | — | |
| DSPBackbone=ResNet, Method Category=Generative2026.03 | 70.7 | 86 | 60 | — | — | |
| CE-GZSLBackbone=ResNet, Method Category=Generative2026.03 | 70 | 78.6 | 63.1 | — | — | |
| ZeroNASBackbone=ResNet, Method Category=Generative2026.03 | 67.6 | 75.3 | 61.4 | — | — | |
| HSVABackbone=ResNet, Method Category=Generative2026.03 | 66.3 | 79.8 | 56.7 | — | — | |
| TDCSSBackbone=ResNet, Method Category=Generative2026.03 | 66.1 | 74.9 | 59.2 | — | — | |
| EGANSBackbone=ResNet, Method Category=Generative2026.03 | 65 | 81.8 | 53.9 | — | — | |
| CADA-VAEFeature Size=642018.12 | 63.9 | 75 | 5,580 | — | — | |
| VGSE-APNBackbone=ResNet, Method Category=Generative2026.03 | 63 | 81.8 | 51.2 | — | — | |
| SE2018.12 | 62.8 | 68.1 | 5,830 | — | — | |
| Tf-GCZSLEvaluation Setting=Online-CGZSL2022.03 | 62.76 | 65.59 | 60.17 | — | — | |
| NM-ZSLEvaluation Setting=Online-CGZSL2022.03 | 61.91 | 84.25 | 48.94 | — | — | |
| FREE + ESZSLBackbone=ResNet, Method Category=Generative2026.03 | 61.8 | 78 | 51.3 | — | — | |
| TF-VAEGAN + SHIPBackbone=ViT, Method Category=Generative, Semantic Prototypes Source=Class Name2026.03 | 60.1 | 96.3 | 43.7 | — | — | |
| f-CLSWGAN2018.12 | 59.4 | 68.9 | 5,210 | — | — | |
| CDL + OSCOBackbone=ResNet, Method Category=Generative2026.03 | 57.1 | 71 | 48 | — | — | |
| CA +memEvaluation Setting=Online-CGZSL2022.03 | 54.12 | 79.23 | 41.09 | — | — | |
| DSECNBackbone=ResNet, Method Category=Embedding2026.03 | 53.7 | — | — | — | — | |
| DVGR-CZSLEvaluation Setting=Online-CGZSL2022.03 | 53.62 | 78.36 | 40.75 | — | — | |
| ICISBackbone=ResNet, Method Category=Embedding2026.03 | 51.6 | 93.3 | 35.6 | — | — | |
| A-CZSLEvaluation Setting=Online-CGZSL2022.03 | 51.57 | 79.75 | 38.11 | — | — | |
| CVAEFeature Size=10242018.12 | 51.2 | — | — | — | — | |
| CV +memEvaluation Setting=Online-CGZSL2022.03 | 49.27 | 85.29 | 34.64 | — | — | |
| Seq-CADAEvaluation Setting=Online-CGZSL2022.03 | 49.2 | 78.59 | 35.81 | — | — | |
| CJME2019.10 | 43.1 | 36.8 | 5,190 | — | — | |
| ReViSEFeature Size=75/1002018.12 | 42.8 | 39.7 | 4,640 | — | — | |
| Seq-CVAEEvaluation Setting=Online-CGZSL2022.03 | 37.45 | 71.84 | 25.33 | — | — | |
| DeViSEFeature Size=20482018.12 | 27.8 | 74.7 | 1,710 | — | — | |
| DEVISE2019.10 | 27.8 | 74.7 | 1,710 | — | — | |
| ALEFeature Size=20482018.12 | 23.9 | 81.8 | 1,400 | — | — | |
| ALE2019.10 | 23.9 | 81.8 | 1,400 | — | — | |
| LATEMFeature Size=20482018.12 | 20 | 77.3 | 1,150 | — | — | |
| Seq-fCLSWGANEvaluation Setting=Online-CGZSL2022.03 | 19.98 | 24.56 | 16.84 | — | — | |
| SYNCFeature Size=20482018.12 | 18 | 90.5 | 1,000 | — | — | |
| SJEFeature Size=20482018.12 | 14.4 | 73.9 | 800 | — | — | |
| EZSLFeature Size=20482018.12 | 11 | 77.8 | 590 | — | — | |
| ESZSL2019.10 | 11 | 77.8 | 590 | — | — | |
| SAE2019.10 | 2.2 | 82.2 | 110 | — | — | |
| CMTFeature Size=20482018.12 | 1 | 90 | 50 | — | — | |
| CONSE2019.10 | 1 | 90.6 | 50 | — | — | |
| ZeroDiff++Venue=Ours, Backbone=Res101, Fine-tune features=our2026.02 | 0.923 | 93.9 | 90.7 | — | — | |
| STHS-GZSLmode=Transductive, sampling=Hardness Sampling2021.06 | 0.832 | 81 | 85.4 | — | — | |
| TPR*Venue=NeurIPS24, Backbone=ViT2026.02 | 0.816 | 87.1 | 76.8 | — | — | |
| ZeroDiffVenue=ICLR25, Backbone=Res101, Fine-tune features=our2026.02 | 0.814 | 89.3 | 74.7 | — | — | |
| PSVMA++Venue=TPAMI25, Backbone=ViT2026.02 | 0.798 | 86.4 | 74.2 | — | — | |
| CLIP*Venue=ICML21, Backbone=ViT2026.02 | 0.796 | 81.6 | 77.6 | — | — | |
| VADSVenue=CVPR24, Backbone=ViT, Fine-tune features=other2026.02 | 0.793 | 83.6 | 75.4 | — | — | |
| ST-WGAN-RSmode=Transductive, sampling=Random Sampling2021.06 | 0.787 | 82.7 | 75.1 | — | — | |
| PSVMAAttack=Clean2025.12 | 0.765 | 84.4 | 69.9 | 71.5 | 2.8 | |
| PSVMAAttack=CBEA-1, Protocol=Best gamma after attack2025.12 | 0.762 | 83.1 | 70.4 | 71 | 2.9 | |
| PSVMAAttack=clsA-1, Protocol=Best gamma after attack2025.12 | 0.76 | 86.5 | 67.8 | 70.7 | 2.5 | |
| PSVMAAttack=clsA-1, Protocol=Original gamma (Eq. 11)2025.12 | 0.759 | 84.3 | 69.1 | 70.7 | 2.8 | |
| PSVMAVenue=CVPR23, Backbone=ViT2026.02 | 0.754 | 77.3 | 73.6 | — | — | |
| CoOp*Venue=IJCV22, Backbone=ViT2026.02 | 0.749 | 81.3 | 69.4 | — | — | |
| ZSLViTFeature Type=ViT2024.04 | 0.742 | 84.6 | 66.1 | — | — | |
| ZSLViTVenue=CVPR24, Backbone=ViT2026.02 | 0.742 | 84.6 | 66.1 | — | — | |
| I2MVFormer-WikiVenue=CVPR'23, Feature Type=ViT2024.04 | 0.738 | 82.9 | 66.6 | — | — | |
| DFCAFlowVenue=TCSVT23, Backbone=Res101, Fine-tune features=our2026.02 | 0.737 | 81 | 67.6 | — | — | |
| ReZSLVenue=TIP23, Backbone=Res1012026.02 | 0.731 | 85.6 | 63.8 | — | — | |
| ZeroMambaAttack=Clean2025.12 | 0.73 | 85.4 | 63.8 | 63.4 | 0.1 | |
| WGAN+ZLAPReference=Proposed, Uses generator=true2022.04 | 0.728 | 82.2 | 65.4 | — | — | |
| DUETVenue=AAAI'23, Feature Type=ViT2024.04 | 0.727 | 84.7 | 63.7 | — | — | |
| TransZero++Venue=TPAMI22, Backbone=Res1012026.02 | 0.725 | 82.7 | 64.6 | — | — | |
| MSDN++Category=Embedding-based Methods, Attention-based=true2026.03 | 0.725 | 79.7 | 66.5 | — | — | |
| ZeroMambaAttack=clsA-1, Protocol=Best gamma after attack2025.12 | 0.721 | 84.8 | 62.7 | 62.9 | 0 | |
| ZeroMambaAttack=clsA-1, Protocol=Original gamma (Eq. 11)2025.12 | 0.72 | 84.7 | 62.6 | 62.9 | 0.1 | |
| ZeroMambaAttack=CBEA-1, Protocol=Best gamma after attack2025.12 | 0.72 | 84.7 | 62.6 | 62.9 | 0.1 | |
| CE-GZSL+ZLAPReference=Proposed, Uses generator=true2022.04 | 0.72 | 80.9 | 64.8 | — | — | |
| EGVenue=TPAMI24, Backbone=ViT2026.02 | 0.717 | 79.6 | 65.2 | — | — | |
| I2DFormer-WikiVenue=NeurIPS'22, Feature Type=ViT2024.04 | 0.715 | 76.8 | 66.8 | — | — | |
| RGENReference=[Xie et al., 2020], Uses generator=false2022.04 | 0.715 | 76.5 | 67.1 | — | — | |
| AGZSLReference=[Chou et al., 2021], Uses generator=true2022.04 | 0.713 | 78.9 | 65.1 | — | — | |
| APN+CVCLearning Paradigm=Generative, Backbone Features=APN features2022.04 | 0.71 | 81 | 63.2 | — | — | |
| ReZSLAttack=Clean2025.12 | 0.71 | 84.8 | 61.1 | 62.1 | 3.8 | |
| EG-part-netCategory=Embedding-based Methods2026.03 | 0.71 | 78.7 | 64.7 | — | — | |
| DSPVenue=ICML23, Backbone=Res101, Fine-tune features=other2026.02 | 0.707 | 86 | 60 | — | — | |
| GEM-ZSLVenue=CVPR'22, Feature Type=CNN, Attention=true2024.04 | 0.706 | 77.5 | 64.8 | — | — | |
| ReZSLAttack=clsA-1, Protocol=Best gamma after attack2025.12 | 0.705 | 84.3 | 60.6 | 61.7 | 3 | |
| ReZSLAttack=CBEA-12025.12 | 0.705 | 84.4 | 60.6 | 61.7 | 3 | |
| TransZeroVenue=AAAI'22, Feature Type=CNN, Attention=true2024.04 | 0.702 | 82.3 | 61.3 | — | — | |
| TransZeroCategory=Embedding-based Methods, Attention-based=true2026.03 | 0.702 | 82.3 | 61.3 | — | — | |
| CE-GZSL2021.03 | 0.7 | 78.6 | 63.1 | — | — | |
| CE-GZSLReference=[Han et al., 2021], Uses generator=true2022.04 | 0.699 | 75 | 65.3 | — | — | |
| APNLearning Paradigm=Non-generative, ZoomInMod=true, Calibrated Stacking=true2022.04 | 0.696 | 79.4 | 61.9 | — | — | |
| CEGANVenue=CVPR21, Backbone=Res101, Fine-tune features=our2026.02 | 0.696 | 89 | 57.1 | — | — | |
| APNCategory=Embedding-based Methods, Attention-based=true2026.03 | 0.696 | 79.4 | 61.9 | — | — |