Multimodal Understanding on LLaVA High-IC tasks (MMB, POPE, MME, SEED, GQA) v1.5
94.7Performance RatioMMTok
Evaluation Results
| Method | Links | |
|---|---|---|
| MMTokToken Budget=642025.08 | 94.7 | |
| DivPruneToken Budget=642025.08 | 93.1 | |
| MMTokToken Budget=322025.08 | 91 | |
| VisionZipToken Budget=642025.08 | 90 | |
| DivPruneToken Budget=322025.08 | 89.6 | |
| MMTokToken Budget=162025.08 | 86.4 | |
| VisionZipToken Budget=322025.08 | 83.5 | |
| DivPruneToken Budget=162025.08 | 81.4 | |
| MMTokToken Budget=82025.08 | 79.8 | |
| MMTokToken Budget=42025.08 | 71.4 | |
| VisionZipToken Budget=162025.08 | 69.7 | |
| DivPruneToken Budget=82025.08 | 68.4 | |
| MMTokToken Budget=22025.08 | 62.1 | |
| DivPruneToken Budget=42025.08 | 54.3 | |
| DivPruneToken Budget=22025.08 | 50.1 | |
| VisionZipToken Budget=82025.08 | 48.9 | |
| VisionZipToken Budget=42025.08 | 43.8 | |
| VisionZipToken Budget=22025.08 | 43 |