Compositional Zero-Shot Learning on MIT-States Open-world (test)
50.7Seen AccuracyLOGICZSL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LOGICZSLleveraging LLM knowledge=true2026.03 | 50.7 | 21.4 | 22.4 | 8.7 | |
| Troika + FlowComposerbaseline=Troika2026.03 | 50.4 | 19 | 20.3 | 7.5 | |
| Troikaimplementation source=re-implemented2026.03 | 50.3 | 17.5 | 19 | 6.8 | |
| IMAX2026.03 | 50.2 | 18.6 | 21.4 | 7.6 | |
| PLOleveraging LLM knowledge=true2026.03 | 49.7 | 19.4 | 21.4 | 7.8 | |
| CDS-CZSL2026.03 | 49.4 | 21.8 | 22.1 | 8.5 | |
| PLIDleveraging LLM knowledge=true2026.03 | 49.1 | 18.7 | 20.4 | 7.3 | |
| CSP + FlowComposerbaseline=CSP2026.03 | 48.2 | 15.6 | 17.8 | 6.1 | |
| DFSP(t2i)2026.03 | 47.5 | 18.5 | 19.3 | 6.8 | |
| CSPimplementation source=re-implemented2026.03 | 47.4 | 15.3 | 17.1 | 5.6 | |
| DFSP(i2t)2026.03 | 47.2 | 18.2 | 19.1 | 6.7 | |
| DFSP(BiF)2026.03 | 47.1 | 18.1 | 19.2 | 6.7 | |
| CoOP2026.03 | 34.6 | 9.3 | 12.3 | 2.8 | |
| CLIP2026.03 | 30.1 | 14.3 | 12.8 | 3 |