Visual Question Answering on VizWiz (C@20, AURC)
68.2AccBaseline
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaselineBackbone=Qwen2-VL 7B, Token Retention Ratio=100%2026.05 | 68.2 | — | — | — | |
| PACT CVPR-25Backbone=Qwen2-VL 7B, Token Retention Ratio=22.2%2026.05 | 67.4 | — | — | — | |
| OccamTokenBackbone=Qwen2-VL 7B, Token Retention Ratio=22.2%2026.05 | 67.1 | — | — | — | |
| OccamTokenBackbone=Qwen2-VL 7B, Token Retention Ratio=11.1%2026.05 | 64.8 | — | — | — | |
| PACT CVPR-25Backbone=Qwen2-VL 7B, Token Retention Ratio=11.1%2026.05 | 63.1 | — | — | — | |
| Self-Filter2025.02 | 53.2 | — | — | — | |
| PRISM2025.02 | 50.1 | — | — | — | |
| Full-Finetune2025.02 | 50 | — | — | — | |
| DataTailor2025.02 | 49.5 | — | — | — | |
| Random2025.02 | 48.1 | — | — | — | |
| Length2025.02 | 47 | — | — | — | |
| Zero-shot verb.Reasoner=o3, Localization setting=w/ localization2026.04 | 46.9 | 0 | 0 | 41.6 | |
| Zero-shot logitReasoner=o3, Localization setting=w/ localization2026.04 | 46.9 | 0 | 0 | 39 | |
| SelectorReasoner=o3, Localization setting=w/ localization2026.04 | 46.9 | 0 | 3.7 | 35.2 | |
| SIEVESReasoner=o3, Localization setting=w/ localization2026.04 | 46.9 | 11.7 | 12.4 | 33.3 | |
| COINCIDE2025.02 | 46.8 | — | — | — | |
| SelectorReasoner=o3, Localization setting=w/o localization2026.04 | 45.8 | 6.3 | 3.8 | 40.9 | |
| EL2N2025.02 | 44.4 | — | — | — | |
| SelectorReasoner=Gemini-3-Pro, Localization setting=w/o localization2026.04 | 44.1 | 3.8 | 4.2 | 41.9 | |
| Zero-shot verb.Reasoner=Gemini-3-Pro, Localization setting=w/ localization2026.04 | 43.6 | 0 | 0 | 46.2 | |
| Zero-shot logitReasoner=Gemini-3-Pro, Localization setting=w/ localization2026.04 | 43.6 | 0 | 0 | 47.7 | |
| SelectorReasoner=Gemini-3-Pro, Localization setting=w/ localization2026.04 | 43.6 | 0 | 2.7 | 39.5 | |
| SIEVESReasoner=Gemini-3-Pro, Localization setting=w/ localization2026.04 | 43.6 | 8.1 | 8.7 | 37.2 | |
| LogprobsReasoner=Pixel-Reasoner, Localization setting=w/o localization2026.04 | 38.1 | 0 | 0.1 | 48.8 | |
| SelectorReasoner=Pixel-Reasoner, Localization setting=w/o localization2026.04 | 38.1 | 1.6 | 3.5 | 45.6 | |
| LogprobsReasoner=Pixel-Reasoner, Localization setting=w/ localization2026.04 | 38 | 0.1 | 0.2 | 51.3 | |
| Zero-shot verb.Reasoner=Pixel-Reasoner, Localization setting=w/ localization2026.04 | 38 | 0 | 0 | 50.8 | |
| Zero-shot logitReasoner=Pixel-Reasoner, Localization setting=w/ localization2026.04 | 38 | 0 | 0 | 53.1 | |
| SelectorReasoner=Pixel-Reasoner, Localization setting=w/ localization2026.04 | 38 | 0.1 | 1.5 | 44.2 | |
| SIEVESReasoner=Pixel-Reasoner, Localization setting=w/ localization2026.04 | 38 | 7.6 | 6.6 | 42.3 | |
| GraNd2025.02 | 37.8 | — | — | — |