Multimodal Understanding on SEED-Bench-I Image Subset
71AccuracyMutual
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MutualAverage Token Budget=192, Compression Level=↓ 66.7%2026.05 | 71 | 99.4 | |
| VisionZip ‡Average Token Budget=64, Compression Level=↓ 88.9%2026.05 | 71 | 95.1 | |
| CLSAverage Token Budget=192, Compression Level=↓ 66.7%2026.05 | 70.8 | 99.3 | |
| VanillaAverage Token Budget=576, Compression Level=100%2026.05 | 70.7 | 100 | |
| VisionZipAverage Token Budget=192, Compression Level=↓ 66.7%2026.05 | 70.7 | 98.2 | |
| VisionZipAverage Token Budget=64, Compression Level=↓ 88.9%2026.05 | 70.7 | 91.8 | |
| MutualAverage Token Budget=128, Compression Level=↓ 77.8%2026.05 | 70.6 | 98.4 | |
| MutualAverage Token Budget=64, Compression Level=↓ 88.9%2026.05 | 70.6 | 96.8 | |
| CLSAverage Token Budget=64, Compression Level=↓ 88.9%2026.05 | 70.5 | 96.6 | |
| CLSAverage Token Budget=128, Compression Level=↓ 77.8%2026.05 | 70.3 | 98.6 | |
| VisionZip ‡Average Token Budget=192, Compression Level=↓ 66.7%2026.05 | 70.2 | 98.4 | |
| VisionZip ‡Average Token Budget=128, Compression Level=↓ 77.8%2026.05 | 70.1 | 97 | |
| VisionZipAverage Token Budget=128, Compression Level=↓ 77.8%2026.05 | 70 | 96.1 |