Multi-modal Evaluation on MME (P+C Metric)
1,818P+C ScoreLLaVA-1.5 Vanilla 13B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaVA-1.5 Vanilla 13BTokens Retained=576, Compression Ratio=100%2025.08 | 1,818 | 100 | |
| VisionZip 🔥Tokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 1,790 | 98.5 | |
| MMTokTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 1,784.16 | 98.1 | |
| DivPruneTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 1,781.5 | 98 | |
| DivPruneTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 1,777.93 | 97.8 | |
| MMTokTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 1,763.39 | 97 | |
| MMTokTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 1,756.2 | 96.6 | |
| VisionZipTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 1,754 | 96.5 | |
| DivPruneTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 1,748.56 | 96.2 | |
| VisionZipTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 1,743 | 95.9 | |
| VisionZip 🔥Tokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 1,736 | 95.5 | |
| VisionZipTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 1,676 | 92.2 | |
| VisionZip 🔥Tokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 1,671 | 91.9 |