Multimodal Evaluation on SEED-IMG
70.2AccuracyVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaRetained Tokens=2880, Fine-tuning=false2024.12 | 70.2 | 100 | |
| VisionZip ‡Retained Tokens=640, Fine-tuning=true2024.12 | 67.8 | 96.6 | |
| VisionZipRetained Tokens=640, Fine-tuning=false2024.12 | 66.7 | 95 | |
| VisionZip ‡Retained Tokens=320, Fine-tuning=true2024.12 | 65.9 | 93.9 | |
| VisionZipRetained Tokens=320, Fine-tuning=false2024.12 | 63.4 | 90.3 | |
| VisionZip ‡Retained Tokens=160, Fine-tuning=true2024.12 | 62.9 | 89.6 | |
| VisionZipRetained Tokens=160, Fine-tuning=false2024.12 | 58.3 | 83 |