Compositional Zero-Shot Learning on MIT-States Closed World (test)
23.5AUCTroika + FlowComposer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Troika + FlowComposerbaseline=Troika2026.03 | 23.5 | 40.2 | — | — | 51.5 | 53.2 | |
| CAILACLIP-based=true, Backbone=ViT-L/142023.05 | 23.4 | 39.9 | 51 | 53.9 | — | — | |
| PLOleveraging LLM knowledge=true2026.03 | 23.4 | 40.2 | — | — | 51.6 | 53.7 | |
| LOGICZSLleveraging LLM knowledge=true2026.03 | 23.4 | 40.5 | — | — | 50.8 | 53.9 | |
| CDS-CZSL2026.03 | 22.4 | 39.2 | — | — | 50.3 | 52.9 | |
| PLIDleveraging LLM knowledge=true2026.03 | 22.1 | 39 | — | — | 49.7 | 52.4 | |
| Troikaimplementation source=re-implemented2026.03 | 22.1 | 39.2 | — | — | 49.3 | 52.5 | |
| IMAX2026.03 | 21.9 | 39.1 | — | — | 48.7 | 53.8 | |
| DFSP(BiF)2026.03 | 20.8 | 37.7 | — | — | 47.1 | 52.8 | |
| DFSP(i2t)2026.03 | 20.7 | 37.2 | — | — | 47.4 | 52.4 | |
| CSP + FlowComposerbaseline=CSP2026.03 | 20.7 | 37.6 | — | — | 48.3 | 50.4 | |
| DFSPCLIP-based=true, Backbone=ViT-L/142023.05 | 20.6 | 37.3 | 46.9 | 52 | — | — | |
| DFSP(t2i)2026.03 | 20.6 | 37.3 | — | — | 46.9 | 52 | |
| DLM2026.03 | 20 | 37.4 | — | — | 46.3 | 49.8 | |
| CSPimplementation source=re-implemented2026.03 | 19.6 | 36.6 | — | — | 47 | 49.6 | |
| CSPCLIP-based=true, Backbone=ViT-L/142023.05 | 19.4 | 36.3 | 46.6 | 49.9 | — | — | |
| Co-CGECLIP-based=true, Backbone=ViT-L/142023.05 | 17 | 33.1 | 46.7 | 45.9 | — | — | |
| CoOpCLIP-based=true, Backbone=ViT-L/142023.05 | 13.5 | 29.8 | 34.4 | 47.6 | — | — | |
| CoOP2026.03 | 13.5 | 29.8 | — | — | 34.4 | 47.6 | |
| CLIP-ZSCLIP-based=true, Backbone=ViT-L/142023.05 | 11 | 26.1 | 30.2 | 46 | — | — | |
| CLIP2026.03 | 11 | 26.1 | — | — | 30.2 | 46 | |
| CAPECLIP-based=false2023.05 | 6.7 | 20.4 | 32.1 | 28 | — | — | |
| Co-CGECLIP-based=false2023.05 | 6.6 | 20 | 32.1 | 28.3 | — | — | |
| CGECLIP-based=false2023.05 | 6.5 | 21.4 | 32.8 | 28 | — | — | |
| OADisCLIP-based=false2023.05 | 5.9 | 18.9 | 31.1 | 25.6 | — | — | |
| SCENCLIP-based=false2023.05 | 5.3 | 18.4 | 29.9 | 25.2 | — | — | |
| CompCosCLIP-based=false2023.05 | 4.5 | 16.4 | 25.3 | 24.6 | — | — |