Compositional Zero-Shot Learning on C-GQA Closed World
34HMTroika + FlowComposer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Troika + FlowComposerbaseline=Troika2026.03 | 34 | 15.9 | — | — | 44.8 | 40.7 | |
| H2EMSetting=Closed-World2025.12 | 33.9 | 16 | — | — | 45 | 40.4 | |
| LOGICZSLSetting=Closed-World2025.12 | 33.3 | 15.3 | — | — | 44.4 | 39.4 | |
| LOGICZSLleveraging LLM knowledge=true2026.03 | 33.3 | 15.3 | — | — | 44.4 | 39.4 | |
| PLOSetting=Closed-World2025.12 | 33 | 14.9 | — | — | 44.7 | 38.1 | |
| PLOleveraging LLM knowledge=true2026.03 | 33 | 14.9 | — | — | 44.7 | 38.1 | |
| CAILACLIP-based=true, Backbone=ViT-L/142023.05 | 32.7 | 14.8 | 43.9 | 38.5 | — | — | |
| CAILASetting=Closed-World, Reproduction=Official Code2025.12 | 31.4 | 13.9 | — | — | 44.3 | 35.8 | |
| IMAX2026.03 | 29.8 | 12.8 | — | — | 39.7 | 35.8 | |
| Troikaimplementation source=re-implemented2026.03 | 29.7 | 12.5 | — | — | 41 | 35.7 | |
| TroikaSetting=Closed-World2025.12 | 29.4 | 12.4 | — | — | 41 | 35.7 | |
| CDS-CZSLadopting_CLIP=true2024.02 | 28.1 | 11.1 | 38.3 | 34.2 | — | — | |
| CDS-CZSLSetting=Closed-World2025.12 | 28.1 | 11.1 | — | — | 38.3 | 34.2 | |
| CDS-CZSL2026.03 | 28.1 | 11.1 | — | — | 38.3 | 34.2 | |
| PLIDSetting=Closed-World2025.12 | 27.9 | 11 | — | — | 38.8 | 33 | |
| PLIDleveraging LLM knowledge=true2026.03 | 27.9 | 11 | — | — | 38.8 | 33 | |
| DFSPadopting_CLIP=true2024.02 | 27.1 | 10.5 | 38.2 | 32 | — | — | |
| DFSP(t2i)Setting=Closed-World2025.12 | 27.1 | 10.5 | — | — | 38.2 | 32 | |
| DFSPCLIP-based=true, Backbone=ViT-L/142023.05 | 27.1 | 10.5 | 38.2 | 32.9 | — | — | |
| DFSP(t2i)2026.03 | 27.1 | 10.5 | — | — | 38.2 | 32 | |
| DFSP(BiF)Setting=Closed-World2025.12 | 26.2 | 9.9 | — | — | 36.5 | 32 | |
| DFSP(BiF)2026.03 | 26.2 | 9.9 | — | — | 36.5 | 32 | |
| DFSP(i2t)Setting=Closed-World2025.12 | 24.3 | 8.7 | — | — | 35.6 | 29.3 | |
| DFSP(i2t)2026.03 | 24.3 | 8.7 | — | — | 35.6 | 29.3 | |
| CSP + FlowComposerbaseline=CSP2026.03 | 22.9 | 7.7 | — | — | 29 | 30.9 | |
| GIPCOLSetting=Closed-World2025.12 | 22.5 | 7.1 | — | — | 31.9 | 28.4 | |
| HPLadopting_CLIP=true2024.02 | 22.4 | 7.2 | 30.8 | 28.4 | — | — | |
| DLM2026.03 | 21.9 | 7.3 | — | — | 32.4 | 28.5 | |
| CSPadopting_CLIP=true2024.02 | 20.5 | 6.2 | 28.8 | 26.8 | — | — | |
| CSPSetting=Closed-World2025.12 | 20.5 | 6.2 | — | — | 28.8 | 26.8 | |
| CSPCLIP-based=true, Backbone=ViT-L/142023.05 | 20.5 | 6.2 | 28.8 | 26.8 | — | — | |
| CSPimplementation source=re-implemented2026.03 | 19.3 | 5.8 | — | — | 24.6 | 29.6 | |
| CLIP-based Co-CGEadopting_CLIP=true2024.02 | 18.9 | 5.7 | 34.1 | 21.2 | — | — | |
| Co-CGECLIP-based=true, Backbone=ViT-L/142023.05 | 18.9 | 5.7 | 34.1 | 21.2 | — | — | |
| ADEadopting_CLIP=false2024.02 | 18 | 5.2 | 35 | 17.7 | — | — | |
| SCENadopting_CLIP=false2024.02 | 17.5 | 5.5 | 28.9 | 25.4 | — | — | |
| CoTadopting_CLIP=false2024.02 | 17.5 | 5.1 | 34 | 18.8 | — | — | |
| CoOPadopting_CLIP=true2024.02 | 17.1 | 4.4 | 20.5 | 26.8 | — | — | |
| CoOpSetting=Closed-World2025.12 | 17.1 | 4.4 | — | — | 20.5 | 26.8 | |
| CoOpCLIP-based=true, Backbone=ViT-L/142023.05 | 17.1 | 4.4 | 26.8 | 20.5 | — | — | |
| CoOP2026.03 | 17.1 | 4.4 | — | — | 20.5 | 26.8 | |
| CAPECLIP-based=false2023.05 | 16.3 | 4.6 | 33 | 16.4 | — | — | |
| CGECLIP-based=false2023.05 | 15.5 | 4.2 | 33.5 | 16 | — | — | |
| ProtoPropCLIP-based=false2023.05 | 15.1 | 3.7 | 26.4 | 18.1 | — | — | |
| CGEadopting_CLIP=false2024.02 | 14.5 | 3.6 | 31.4 | 14 | — | — | |
| CANetadopting_CLIP=false2024.02 | 14.5 | 3.3 | 30 | 13.2 | — | — | |
| Co-CGEadopting_CLIP=false2024.02 | 14.4 | 4.1 | 33.3 | 14.9 | — | — | |
| Co-CGECLIP-based=false2023.05 | 14.4 | 4.1 | 33.3 | 14.9 | — | — | |
| CompCosadopting_CLIP=false2024.02 | 12.4 | 2.6 | 28.1 | 11.2 | — | — | |
| CompCosCLIP-based=false2023.05 | 12.4 | 2.6 | 28.1 | 11.2 | — | — | |
| SCENCLIP-based=false2023.05 | 12.4 | 2.9 | 28.9 | 12.1 | — | — | |
| CLIPadopting_CLIP=true2024.02 | 8.6 | 1.4 | 7.5 | 25 | — | — | |
| CLIPSetting=Closed-World2025.12 | 8.6 | 1.4 | — | — | 7.5 | 25 | |
| CLIP-ZSCLIP-based=true, Backbone=ViT-L/142023.05 | 8.6 | 1.4 | 7.5 | 25 | — | — | |
| CLIP2026.03 | 8.6 | 1.4 | — | — | 7.5 | 25 | |
| Co-CGESetting=Closed-World2025.12 | 5.3 | 0.9 | — | — | 33.2 | 3.9 |