Compositional Zero-Shot Learning on C-GQA (open world)
24.3HM ScoreDFSP(i2t)
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DFSP(i2t)Setting=Open-World2025.12 | 24.3 | 8.7 | — | — | 35.6 | 29.3 | — | — | — | — | — | |
| DFSP + SPAParadigm=VLM Prompting Paradigm, SPA=true, Backbone=CLIP ViT-B/322026.03 | 22.81 | 0.0851 | — | — | 48.68 | 20.24 | 30.63 | 29.03 | 19.5 | 12.21 | 10.75 | |
| DFSPParadigm=VLM Prompting Paradigm, SPA=false, Backbone=CLIP ViT-B/322026.03 | 22.57 | 0.0835 | — | — | 48.51 | 19.86 | 30.63 | 28.06 | 19.6 | 12.02 | 9.11 | |
| Troika + SPAParadigm=VLM Prompting Paradigm, SPA=true, Backbone=CLIP ViT-B/322026.03 | 21.43 | 0.0703 | — | — | 40.89 | 19.49 | 27.81 | 18.45 | 21.91 | 11.74 | 11.54 | |
| TroikaParadigm=VLM Prompting Paradigm, SPA=false, Backbone=CLIP ViT-B/322026.03 | 20.52 | 0.069 | — | — | 43.87 | 18.08 | 27.98 | 22.91 | 19.4 | 11.74 | 1.73 | |
| SPA AvgParadigm=VLM Prompting Paradigm, SPA=true, Backbone=CLIP ViT-B/322026.03 | 19.29 | 0.0592 | — | — | 37.79 | 17.49 | 25.37 | 18.11 | 19.47 | 9.32 | 10.97 | |
| Baseline AvgParadigm=VLM Prompting Paradigm, SPA=false, Backbone=CLIP ViT-B/322026.03 | 18.15 | 0.0549 | — | — | 37.21 | 16.35 | 24.22 | 18.73 | 17.43 | 9.72 | 7.07 | |
| CSP + SPAParadigm=VLM Prompting Paradigm, SPA=true, Backbone=CLIP ViT-B/322026.03 | 16.7 | 0.0407 | — | — | 29.47 | 15.77 | 20.86 | 13.11 | 18.98 | 7.12 | 10.67 | |
| HPL + SPAParadigm=VLM Prompting Paradigm, SPA=true, Backbone=CLIP ViT-B/322026.03 | 16.21 | 0.0407 | — | — | 32.12 | 14.46 | 22.19 | 11.84 | 17.47 | 6.22 | 10.93 | |
| H2EMSetting=Open-World2025.12 | 15.2 | 4.4 | — | — | 45.1 | 10.8 | — | — | — | — | — | |
| CSPParadigm=VLM Prompting Paradigm, SPA=false, Backbone=CLIP ViT-B/322026.03 | 15.19 | 0.0354 | — | — | 29.47 | 14.03 | 19.54 | 10.87 | 16.49 | 7.65 | 9.28 | |
| HPLParadigm=VLM Prompting Paradigm, SPA=false, Backbone=CLIP ViT-B/322026.03 | 14.3 | 0.0316 | — | — | 26.99 | 13.42 | 18.71 | 13.06 | 14.24 | 7.45 | 8.15 | |
| PLOSetting=Open-World2025.12 | 13.9 | 3.9 | — | — | 43.9 | 10.4 | — | — | — | — | — | |
| PLOleveraging LLM knowledge=true2026.03 | 13.9 | 1 | — | — | 43.9 | 10.4 | — | — | — | — | — | |
| LOGICZSLSetting=Open-World2025.12 | 12.6 | 3.4 | — | — | 43.7 | 9.3 | — | — | — | — | — | |
| LOGICZSLleveraging LLM knowledge=true2026.03 | 12.6 | 0.8718 | — | — | 43.7 | 9.3 | — | — | — | — | — | |
| Troika + FlowComposerbaseline=Troika2026.03 | 12.6 | 0.8974 | — | — | 43.5 | 10.2 | — | — | — | — | — | |
| OADis+NELParadigm=Traditional (Non-VLM)2026.03 | 12.11 | 0.0318 | — | — | 42.38 | 9.77 | 19.78 | 16.07 | 12.86 | 2.87 | 3.04 | |
| CANetParadigm=Traditional (Non-VLM)2026.03 | 11.96 | 0.0304 | — | — | 40.52 | 9.21 | 22.43 | 20.87 | 4.95 | 2.03 | 0.64 | |
| Troikaimplementation source=re-implemented2026.03 | 11.7 | 0.7436 | — | — | 40.8 | 8.6 | — | — | — | — | — | |
| CDS-CZSLSetting=Open-World2025.12 | 11.6 | 2.7 | — | — | 37.6 | 8.2 | — | — | — | — | — | |
| CDS-CZSL2026.03 | 11.6 | 0.6923 | — | — | 37.6 | 8.2 | — | — | — | — | — | |
| CAILACLIP-based=true, Visual Backbone=ViT-L/142023.05 | 11.5 | 3.08 | 43.9 | 8 | — | — | — | — | — | — | — | |
| IMAX2026.03 | 11.2 | 0.641 | — | — | 38.7 | 7.9 | — | — | — | — | — | |
| TroikaSetting=Open-World2025.12 | 10.9 | 2.7 | — | — | 40.8 | 7.9 | — | — | — | — | — | |
| DFSP(BiF)Setting=Open-World2025.12 | 10.6 | 2.4 | — | — | 36.4 | 7.6 | — | — | — | — | — | |
| PLIDSetting=Open-World2025.12 | 10.6 | 2.5 | — | — | 39.1 | 7.5 | — | — | — | — | — | |
| DFSP(BiF)2026.03 | 10.6 | 0.6128 | — | — | 36.4 | 7.6 | — | — | — | — | — | |
| PLIDleveraging LLM knowledge=true2026.03 | 10.6 | 0.641 | — | — | 39.1 | 7.5 | — | — | — | — | — | |
| DFSP(t2i)Setting=Open-World2025.12 | 10.4 | 2.4 | — | — | 38.3 | 7.2 | — | — | — | — | — | |
| DFSPCLIP-based=true, Visual Backbone=ViT-L/142023.05 | 10.4 | 2.4 | 38.3 | 7.2 | — | — | — | — | — | — | — | |
| DFSP(t2i)2026.03 | 10.4 | 0.6154 | — | — | 38.3 | 7.2 | — | — | — | — | — | |
| OADisParadigm=Traditional (Non-VLM)2026.03 | 9.74 | 0.0233 | — | — | 42.88 | 7.12 | 20.86 | 15.19 | 6.17 | 3.47 | 0.61 | |
| CompCosParadigm=Traditional (Non-VLM)2026.03 | 9.64 | 0.0235 | — | — | 40.19 | 7.25 | 20.19 | 20.24 | 4.47 | 1.95 | 0.26 | |
| SCENParadigm=Traditional (Non-VLM)2026.03 | 9.03 | 0.0197 | — | — | 41.65 | 7.83 | 20.65 | 21.42 | 3.61 | 1.08 | 0.05 | |
| DFSP(i2t)2026.03 | 9 | 0.5 | — | — | 35.6 | 6.5 | — | — | — | — | — | |
| LEParadigm=Traditional (Non-VLM)2026.03 | 8.39 | 0.0117 | — | — | 19.37 | 8.36 | 10.76 | 6.51 | 9.53 | 2.67 | 1.08 | |
| CSP + FlowComposerbaseline=CSP2026.03 | 8 | 0.4615 | — | — | 31.2 | 5.8 | — | — | — | — | — | |
| CAILASetting=Open-World, Reproduction=Official Code2025.12 | 7.9 | 3.3 | — | — | 43.8 | 11.4 | — | — | — | — | — | |
| GIPCOLSetting=Open-World2025.12 | 7.3 | 1.3 | — | — | 31.6 | 5.5 | — | — | — | — | — | |
| CSPimplementation source=re-implemented2026.03 | 7.3 | 0.3333 | — | — | 29.6 | 5.4 | — | — | — | — | — | |
| CSPSetting=Open-World2025.12 | 6.9 | 1.2 | — | — | 28.7 | 5.2 | — | — | — | — | — | |
| CSPCLIP-based=true, Visual Backbone=ViT-L/142023.05 | 6.9 | 1.2 | 28.7 | 5.2 | — | — | — | — | — | — | — | |
| CoOp (a)CLIP-based=true, Visual Backbone=ViT-L/142023.05 | 5.7 | 0.73 | 20.9 | 4.5 | — | — | — | — | — | — | — | |
| CoOpSetting=Open-World2025.12 | 5.5 | 0.7 | — | — | 21 | 4.6 | — | — | — | — | — | |
| CoOp (b)CLIP-based=true, Visual Backbone=ViT-L/142023.05 | 5.5 | 0.7 | 21 | 4.6 | — | — | — | — | — | — | — | |
| CoOP2026.03 | 5.5 | 0.1795 | — | — | 21 | 4.6 | — | — | — | — | — | |
| Co-CGESetting=Open-World2025.12 | 5.3 | 0.9 | — | — | 33.2 | 3.9 | — | — | — | — | — | |
| Co-CGECLIP-based=true, Visual Backbone=ViT-L/14, mode=CLIP features (best reported)2023.05 | 5.3 | 0.91 | 33.2 | 3.9 | — | — | — | — | — | — | — | |
| SASOW2025.12 | 5 | 0.85 | 32.7 | 3.3 | — | — | — | — | — | — | — | |
| Co-CGEopenCLIP-based=false2023.05 | 4.8 | 0.78 | 32.1 | 3 | — | — | — | — | — | — | — | |
| KG-SPCLIP-based=false2023.05 | 4.7 | 0.78 | 31.5 | 2.9 | — | — | — | — | — | — | — | |
| KG-SPsource=paper2025.12 | 4.7 | 0.78 | 31.5 | 2.9 | — | — | — | — | — | — | — | |
| KG-SPsource=Implementation2025.12 | 4.3 | 0.69 | 31.6 | 2.7 | — | — | — | — | — | — | — | |
| CLIPSetting=Open-World2025.12 | 4 | 0.3 | — | — | 7.5 | 4.6 | — | — | — | — | — | |
| CLIP-ZSCLIP-based=true, Visual Backbone=ViT-L/142023.05 | 4 | 0.27 | 7.5 | 4.6 | — | — | — | — | — | — | — | |
| CLIP2026.03 | 4 | 0.0692 | — | — | 7.5 | 4.6 | — | — | — | — | — | |
| Co-CGECWCLIP-based=false2023.05 | 3.4 | 0.53 | 32.1 | 2 | — | — | — | — | — | — | — | |
| CompCosCLIP-based=false2023.05 | 3.3 | 0.43 | 26.7 | 2.2 | — | — | — | — | — | — | — | |
| SymNet2025.12 | 3.3 | 0.43 | 26.7 | 2.2 | — | — | — | — | — | — | — | |
| CGECLIP-based=false2023.05 | 2.9 | 0.47 | 32.7 | 1.8 | — | — | — | — | — | — | — | |
| CGE2025.12 | 2.9 | 0.47 | 32.7 | 1.8 | — | — | — | — | — | — | — | |
| VisProd2025.12 | 2.8 | 0.33 | 24.8 | 1.7 | — | — | — | — | — | — | — | |
| CompCos2025.12 | 2.8 | 0.39 | 28.4 | 1.8 | — | — | — | — | — | — | — | |
| LE+2025.12 | 1 | 0.08 | 19.2 | 0.9 | — | — | — | — | — | — | — |