Visual Question Answering on GQA OOD (test)
57.5AccuracyCgRAG
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CgRAG2026.05 | 57.5 | — | — | — | — | — | |
| VCIN2026.05 | 54.2 | — | — | — | — | — | |
| REX2026.05 | 52.1 | — | — | — | — | — | |
| GenB2022.08 | 49.43 | 45.63 | 51.76 | — | — | — | |
| EXP2026.05 | 47.6 | — | — | — | — | — | |
| UpDn2022.08 | 46.87 | 42.13 | 49.16 | — | — | — | |
| VQA-E2026.05 | 46.2 | — | — | — | — | — | |
| RUBi2022.08 | 45.85 | 43.37 | 47.37 | — | — | — | |
| CSS2022.08 | 44.24 | 41.2 | 46.11 | — | — | — | |
| LMH2022.08 | 43.96 | 40.73 | 45.93 | — | — | — | |
| VISFISModel Architecture=UpDn2022.06 | 34.88 | — | — | — | — | — | |
| Singla et al.Model Architecture=UpDn2022.06 | 32.58 | — | — | — | — | — | |
| Selvaraju et al.Model Architecture=UpDn2022.06 | 31.99 | — | — | — | — | — | |
| Simpson et al.Model Architecture=UpDn2022.06 | 31.99 | — | — | — | — | — | |
| BaselineModel Architecture=UpDn2022.06 | 31.8 | — | — | — | — | — | |
| Suff-RandomModel Architecture=UpDn2022.06 | 31.65 | — | — | — | — | — | |
| Wu and MooneyModel Architecture=UpDn2022.06 | 31.61 | — | — | — | — | — | |
| Chang et al.Model Architecture=UpDn2022.06 | 30.4 | — | — | — | — | — | |
| w/ Rand. Supervis.Model Architecture=UpDn2022.06 | 29.93 | — | — | — | — | — | |
| MRVQA2026.05 | 26.5 | — | — | — | — | — | |
| BAN4Visual features=RCNN [2], Additional supervision=None, Training images (M)=≈ 0.1, Training sentences (M)=≈ 12021.06 | — | 0.472 | 0.519 | — | — | — | |
| HONESBackbone=LLaVA-1.5 7B, Baseline Comparison=Base, OOD tuning protocol=re-learn scaling on 20% subset2026.04 | — | — | — | 2.15 | 1 | — | |
| HONESBackbone=LLaVA-1.5 7B, Baseline Comparison=Fixed-Amp, OOD tuning protocol=re-learn scaling on 20% subset2026.04 | — | — | — | 1.25 | 0.96 | — | |
| HONESBackbone=LLaVA-1.5 7B, Baseline Comparison=RandNeuron, OOD tuning protocol=re-learn scaling on 20% subset2026.04 | — | — | — | 2.05 | 1.04 | — | |
| LXMERTVisual features=RCNN [2], Additional supervision=None, Training images (M)=~0.18, Training sentences (M)=~ 92021.06 | — | 0.498 | 0.577 | — | — | — | |
| MCANVisual features=RCNN [2], Additional supervision=None, Training images (M)=≈ 0.1, Training sentences (M)=≈ 12021.06 | — | 0.465 | 0.534 | — | — | — | |
| MMNVisual features=RCNN [2], Additional supervision=Program, Training images (M)=≈ 0.1, Training sentences (M)=≈ 152021.06 | — | 0.48 | 0.555 | — | — | — | |
| Oracle transferVisual features=RCNN [2], Additional supervision=None, Training images (M)=~0.18, Training sentences (M)=≈ 12021.06 | — | 0.483 | 0.555 | — | — | — | |
| OursVisual features=VinVL [45], Additional supervision=Program, Training images (M)=≈ 0.1, Training sentences (M)=≈ 152021.06 | — | 0.491 | 0.597 | — | — | — |