Visual Question Answering on GQA (GQA Score, Average, and Relative Performance %)
64.2GQA ScoreLLaVA-Next (Full)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LLaVA-Next (Full)Remain Tokens=28802026.05 | 64.2 | 76.8 | 100 | |
| ASAP (Ours)Remain Tokens=6402026.05 | 62.3 | 75.9 | 98.8 | |
| LLaVA-1.5-7BVision Token=5762026.06 | 62 | 100 | — | |
| MS-ResamplerVision Token=1442026.06 | 62 | 100.43 | — | |
| PruneSIDRemain Tokens=6402026.05 | 61.6 | 74.8 | 97.4 | |
| VisionZipRemain Tokens=6402026.05 | 61.3 | 75.1 | 97.8 | |
| ASAP (Ours)Remain Tokens=3202026.05 | 61 | 74 | 96.3 | |
| PruneSIDRemain Tokens=3202026.05 | 60.5 | 73 | 95.1 | |
| FitPruneVision Token=1922026.06 | 60.4 | — | — | |
| DARTVision Token=1922026.06 | 60 | 97.98 | — | |
| VisionZipRemain Tokens=3202026.05 | 59.3 | 72.2 | 94 | |
| PruneSIDRemain Tokens=1602026.05 | 58.9 | 70.5 | 91.7 | |
| HiREDVision Token=1922026.06 | 58.7 | 89.06 | — | |
| MustDropVision Token=1922026.06 | 58.2 | 96.26 | — | |
| SparseVLMVision Token=1922026.06 | 57.6 | 95.79 | — | |
| PDropVision Token=1922026.06 | 57.1 | 96.19 | — | |
| ASAP (Ours)Remain Tokens=1602026.05 | 57 | 71.6 | 93.2 | |
| VisionZipRemain Tokens=1602026.05 | 55.5 | 68.6 | 89.3 | |
| LLaVA-PruMergeVision Token=1922026.06 | 54.3 | 88.92 | — | |
| FastVVision Token=1922026.06 | 52.7 | 89.58 | — |