Visual Question Answering on GQA v1.2 (test)
61.9GQA ScoreVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaBackbone=LLaVA-1.5-7B, Token Budget=576, Method Type=Baseline2026.02 | 61.9 | 100 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE, Venue=Ours2026.02 | 61.1 | 98.9 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/o VE, Venue=Ours2026.02 | 61 | 98.8 | |
| SCOPEBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE, Venue=NeurIPS’252026.02 | 60.1 | 98.3 | |
| DARTBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/o VE, Venue=EMNLP’252026.02 | 60 | 98.3 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/o VE, Venue=Ours2026.02 | 60 | 98.1 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/ VE, Venue=Ours2026.02 | 60 | 98.1 | |
| SCOPEBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/ VE, Venue=NeurIPS’252026.02 | 59.7 | 97.5 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/o VE, Venue=ICML’252026.02 | 59.5 | 97.2 | |
| VisionZipBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE, Venue=CVPR’252026.02 | 59.3 | 97.8 | |
| HoloVBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE, Venue=NeurIPS’252026.02 | 59 | 98.7 | |
| DARTBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/o VE, Venue=EMNLP’252026.02 | 58.7 | 97 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/o VE, Venue=ICML’252026.02 | 58.4 | 95.6 | |
| SCOPEBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/ VE, Venue=NeurIPS’252026.02 | 58.3 | 95.4 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/ VE, Venue=Ours2026.02 | 58.3 | 96.6 | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/o VE, Venue=Ours2026.02 | 58 | 94.7 | |
| HoloVBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/ VE, Venue=NeurIPS’252026.02 | 57.7 | 97.2 | |
| VisionZipBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/ VE, Venue=CVPR’252026.02 | 57.6 | 96.4 | |
| PDropBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/o VE, Venue=CVPR’252026.02 | 57.1 | 96.2 | |
| PDropBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/o VE, Venue=CVPR’252026.02 | 56 | 94 | |
| DARTBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/o VE, Venue=EMNLP’252026.02 | 55.9 | 92.8 | |
| HoloVBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/ VE, Venue=NeurIPS’252026.02 | 55.3 | 94.4 | |
| VisionZipBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/ VE, Venue=CVPR’252026.02 | 55.1 | 93 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/o VE, Venue=ICML’252026.02 | 53.8 | 90.6 | |
| FastVBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/o VE, Venue=ECCV’242026.02 | 52.7 | 89 | |
| FastVBackbone=LLaVA-1.5-7B, Token Budget=128, VE=w/o VE, Venue=ECCV’242026.02 | 49.6 | 83.2 | |
| FastVBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/o VE, Venue=ECCV’242026.02 | 46.1 | 74 | |
| PDropBackbone=LLaVA-1.5-7B, Token Budget=64, VE=w/o VE, Venue=CVPR’252026.02 | 41.9 | 74.4 |