Open-Ended Abstract Retrieval (OAR) on Polyvore
76.8DiversityR4T-FOLM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R4T-FOLMBackbone=Gemma3-4B2026.03 | 76.8 | 30.8 | 39.8 | 49.1 | |
| R4T-DiffusionBackbone=Gemma3-4B2026.03 | 74.3 | — | 37.6 | — | |
| R4T-DiffusionBackbone=Qwen3-4B2026.03 | 65 | — | 27.4 | — | |
| R4T-FOLMBackbone=Qwen3-4B2026.03 | 62.8 | 37 | 28 | 42.6 | |
| Gemma3-4BStrategy=Best-of-N2026.03 | 61 | 28.9 | 32.7 | 40.9 | |
| Gemma3-4BStrategy=Zero-shot2026.03 | 56 | 28.4 | 31.2 | 38.5 | |
| Gemini-2.5-FlashStrategy=Best-of-N2026.03 | 52.2 | 26.1 | 25.2 | 34.5 | |
| Gemini-2.5-FlashStrategy=Zero-shot2026.03 | 47 | 24 | 23.6 | 31.5 | |
| Qwen3-4BStrategy=Best-of-N2026.03 | 40.3 | 27 | 24 | 30.4 | |
| Qwen3-4BStrategy=Zero-shot2026.03 | 37 | 23.8 | 23.4 | 28.1 | |
| No Fan-out2026.03 | 34.4 | 22.4 | 21.4 | 26.1 |