Compositional Zero-Shot Learning on VAW CZSL (test)
34.6HMCAILA
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CAILACLIP usage=With CLIP, Evaluation protocol=Closed World2023.05 | 34.6 | — | — | 41.6 | 49.2 | — | — | 17.2 | — | — | — | — | |
| DFSPCLIP usage=With CLIP, Evaluation protocol=Closed World2023.05 | 31.1 | — | — | 40.1 | 40.9 | — | — | 14.1 | — | — | — | — | |
| CSPCLIP usage=With CLIP, Evaluation protocol=Closed World2023.05 | 23.3 | — | — | 31.9 | 33.6 | — | — | 8.5 | — | — | — | — | |
| DFSP+SPAParadigm=VLM Prompting Paradigm2026.03 | 21.91 | — | — | — | — | — | — | 8.14 | 26.39 | 8.38 | 3.61 | 5.29 | |
| DFSPParadigm=VLM Prompting Paradigm2026.03 | 20.87 | — | — | — | — | — | — | 7.64 | 26.57 | 7.26 | 3.24 | 4.78 | |
| Troika+SPAParadigm=VLM Prompting Paradigm2026.03 | 20.34 | — | — | — | — | — | — | 5.26 | 21.79 | 9.16 | 5.07 | 3.24 | |
| TroikaParadigm=VLM Prompting Paradigm2026.03 | 18.25 | — | — | — | — | — | — | 5.17 | 24.94 | 7.42 | 2.09 | 0.72 | |
| Baseline Avg+SPAParadigm=VLM Prompting Paradigm2026.03 | 17.3 | — | — | — | — | — | — | 4.56 | 17.33 | 7.35 | 3.95 | 5.08 | |
| Baseline AvgParadigm=VLM Prompting Paradigm2026.03 | 16 | — | — | — | — | — | — | 4.41 | 18.48 | 5.82 | 2.74 | 3.82 | |
| OADisBackbone=Frozen ResNet, Embeddings=GloVe, Feature Pooling=Convolutional2022.05 | 15.2 | 3.6 | 6.1 | 24.9 | 18.7 | 17.5 | 43.3 | — | — | — | — | — | |
| OADisCLIP usage=Without CLIP, Evaluation protocol=Closed World2023.05 | 15.2 | — | — | 24.9 | 18.7 | — | — | 6.1 | — | — | — | — | |
| CompCosBackbone=Frozen ResNet, Embeddings=GloVe2022.05 | 14.2 | 3.2 | 5.6 | 23.9 | 18 | 16.9 | 41.9 | — | — | — | — | — | |
| CompCosCLIP usage=Without CLIP, Evaluation protocol=Closed World2023.05 | 14.2 | — | — | 23.9 | 18 | — | — | 5.6 | — | — | — | — | |
| CSP+SPAParadigm=VLM Prompting Paradigm2026.03 | 13.92 | — | — | — | — | — | — | 2.42 | 9.75 | 6.91 | 4.36 | 6.7 | |
| CSPParadigm=VLM Prompting Paradigm2026.03 | 13.26 | — | — | — | — | — | — | 2.71 | 12.95 | 4.43 | 3.21 | 5.59 | |
| HPL+SPAParadigm=VLM Prompting Paradigm2026.03 | 13.03 | — | — | — | — | — | — | 2.45 | 11.38 | 4.94 | 2.74 | 5.07 | |
| GraphEmbBackbone=Frozen ResNet, Embeddings=GloVe2022.05 | 13 | 2.9 | 5.1 | 23.4 | 16.8 | 16.9 | 40.8 | — | — | — | — | — | |
| CGECLIP usage=Without CLIP, Evaluation protocol=Closed World2023.05 | 13 | — | — | 23.4 | 16.8 | — | — | 5.1 | — | — | — | — | |
| SymnetBackbone=Frozen ResNet, Embeddings=GloVe2022.05 | 12.2 | 2.3 | 3.9 | 19.1 | 15.8 | 18.6 | 40.9 | — | — | — | — | — | |
| TMNBackbone=Frozen ResNet, Embeddings=GloVe2022.05 | 11.9 | 2.3 | 4 | 19.9 | 15.4 | 15.9 | 38.3 | — | — | — | — | — | |
| CLIP-ZSCLIP usage=With CLIP, Evaluation protocol=Closed World2023.05 | 11.9 | — | — | 12.8 | 27.8 | — | — | 2.6 | — | — | — | — | |
| HPLParadigm=VLM Prompting Paradigm2026.03 | 11.6 | — | — | — | — | — | — | 2.12 | 9.44 | 4.18 | 2.42 | 4.18 | |
| OADis+NELParadigm=Traditional Paradigm (Non-VLM)2026.03 | 11.35 | — | — | — | — | — | — | 2.91 | 23.02 | 16.18 | 7.86 | 1.36 | |
| CANetParadigm=Traditional Paradigm (Non-VLM)2026.03 | 11.21 | — | — | — | — | — | — | 2.89 | 24.56 | 18.42 | 5.74 | 0.95 | |
| OADisParadigm=Traditional Paradigm (Non-VLM)2026.03 | 10.91 | — | — | — | — | — | — | 2.68 | 21.19 | 15.65 | 6.75 | 0.76 | |
| CompCosParadigm=Traditional Paradigm (Non-VLM)2026.03 | 10.68 | — | — | — | — | — | — | 2.69 | 20.21 | 20.58 | 5.04 | 0.5 | |
| SCENParadigm=Traditional Paradigm (Non-VLM)2026.03 | 10.64 | — | — | — | — | — | — | 2.53 | 19.06 | 20.76 | 4.52 | 0.42 | |
| LabelEmbed+Backbone=Frozen ResNet, Embeddings=GloVe2022.05 | 9.8 | 1.6 | 2.8 | 16.2 | 13.2 | 13.4 | 35.1 | — | — | — | — | — | |
| AttrOprBackbone=Frozen ResNet, Embeddings=GloVe2022.05 | 9.1 | 1.4 | 2.6 | 16.4 | 11.7 | 13.7 | 34.9 | — | — | — | — | — | |
| LEParadigm=Traditional Paradigm (Non-VLM)2026.03 | 8.27 | — | — | — | — | — | — | 1.49 | 15.62 | 10.48 | 5.79 | 0.98 |