Dense Prompt Alignment on DPG-Bench
85.08Overall ScoreBAGEL Image-Only
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BAGEL Image-OnlyModality Ablation=Image-Only, Base model baseline=true2026.01 | 85.08 | 90.73 | 90.05 | 89.72 | 92.23 | 90.8 | |
| Multimodal RWRModality Ablation=Multimodal2026.01 | 84.37 | 89.2 | 90.63 | 89.73 | 90.03 | 86.81 | |
| Janus-Pro-7BModality Ablation=Multimodal2026.01 | 84.19 | 86.9 | 88.9 | 89.4 | 89.32 | 89.48 | |
| SD3-MediumModality Ablation=Image-Only2026.01 | 84.08 | 87.9 | 91.01 | 88.83 | 80.7 | 88.68 | |
| BAGEL MultimodalModality Ablation=Multimodal, Base model baseline=true2026.01 | 84.01 | 88.75 | 90.84 | 89.01 | 89.44 | 89.7 | |
| FLUX.1[dev]Modality Ablation=Image-Only2026.01 | 83.84 | 74.35 | 90 | 88.96 | 90.87 | 88.33 | |
| Text RWRModality Ablation=Multimodal2026.01 | 83.84 | 89.94 | 89.1 | 89.24 | 89.41 | 90.58 | |
| Image RWRModality Ablation=Multimodal2026.01 | 83.6 | 90.39 | 90.87 | 88.74 | 87.67 | 86.97 | |
| Multimodal SFT*Modality Ablation=Multimodal, SFT baseline=true2026.01 | 83.57 | 88.73 | 88.92 | 88.89 | 90.04 | 87.02 | |
| Image-Only SFT*Modality Ablation=Image-Only, SFT baseline=true2026.01 | 83.56 | 91.59 | 89.93 | 89.33 | 89.97 | 84.15 | |
| BLIP3-oModality Ablation=Multimodal2026.01 | 81.6 | — | — | — | — | — |