Visual Question Answering on SQA-Image
70.2AccuracyDUET-VLM (C)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DUET-VLM (C)Average Tokens=128, Token Reduction (%)=77.8%2026.02 | 70.2 | 101 | |
| DUET-VLM (C)Average Tokens=192, Token Reduction (%)=66.7%2026.02 | 69.6 | 100.1 | |
| LLaVA-1.5-7BAverage Tokens=576, Token Reduction (%)=0%2026.02 | 69.5 | 100 | |
| PyramidDropAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 69.2 | 99.6 | |
| SparseVLMAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 69.1 | 99.4 | |
| PyramidDropAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 69 | 99.3 | |
| VisionZipAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 69 | 99.3 | |
| VisionZipAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 68.9 | 99.1 | |
| VisionZipAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 68.9 | 99.1 | |
| HiREDAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 68.4 | 98.4 | |
| SparseVLMAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 68.4 | 98.4 | |
| PyramidDropAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 68.4 | 98.4 | |
| DUET-VLM (C)Average Tokens=64, Token Reduction (%)=88.9%2026.02 | 68.3 | 98.3 | |
| HiREDAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 68.2 | 98.1 | |
| HiREDAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 68.1 | 98 | |
| PruMergeAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 68.1 | 98 | |
| FitPruneAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 68 | 97.8 | |
| FitPruneAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 68 | 97.8 | |
| PruMergeAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 67.9 | 97.7 | |
| FitPruneAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 67.8 | 97.6 | |
| FastVAverage Tokens=192, Token Reduction (%)=66.7%2026.02 | 67.3 | 96.8 | |
| PruMergeAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 67.1 | 96.5 | |
| SparseVLMAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 62.2 | 89.5 | |
| FastVAverage Tokens=128, Token Reduction (%)=77.8%2026.02 | 60.2 | 86.6 | |
| FastVAverage Tokens=64, Token Reduction (%)=88.9%2026.02 | 51.1 | 73.5 |