Vision-Language Evaluation on MMBench
69.3MMB ScoreVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaRetained Tokens=576, Token Reduction Ratio=100%2026.03 | 69.3 | 100 | |
| VILA-7BSIMA=true2024.05 | 69.2 | — | |
| VILA-7BSIMA=false2024.05 | 68.9 | — | |
| VisionZipRetained Tokens=192, Token Reduction Ratio=↓ 66.7%2026.03 | 68.9 | 98.3 | |
| VisionZipRetained Tokens=128, Token Reduction Ratio=↓ 77.8%2026.03 | 68.1 | 96.2 | |
| PRUNESIDRetained Tokens=192, Token Reduction Ratio=↓ 66.7%2026.03 | 67.2 | 98.7 | |
| PRUNESIDRetained Tokens=128, Token Reduction Ratio=↓ 77.8%2026.03 | 66.6 | 97.4 | |
| VisionZipRetained Tokens=64, Token Reduction Ratio=↓ 88.9%2026.03 | 65.9 | 92.4 | |
| LLaVA-1.5-7BSIMA=true2024.05 | 64.9 | — | |
| mPLUG-Owl22024.05 | 64.5 | — | |
| LLaVA-1.5-7BSIMA=false2024.05 | 64.3 | — | |
| PRUNESIDRetained Tokens=64, Token Reduction Ratio=↓ 88.9%2026.03 | 63.1 | 94.4 | |
| BaselineRetain Tokens=Full2026.02 | 62.2 | — | |
| Qwen-VL-Chat2024.05 | 60.6 | — | |
| IDPrunerRetain Tokens=1282026.02 | 60.4 | — | |
| IDPrunerRetain Tokens=642026.02 | 57.65 | — | |
| IDPrunerRetain Tokens=322026.02 | 55.5 | — | |
| InstructBLIP2024.05 | 36 | — | |
| IDEFICS2024.05 | 30 | — |