Audio-visual Question Answering on WorldSense
50AccuracyOmniSIFT
Evaluation Results
| Method | Links | |
|---|---|---|
| OmniSIFTBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 50 | |
| OmniSIFTBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 49.9 | |
| Full TokensBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=1002026.02 | 49.7 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 48.9 | |
| DyCokeBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 48.6 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 48.3 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 48.1 | |
| DyCokeBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 48.1 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 47.1 | |
| Full TokensBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=1002026.02 | 45.8 | |
| OmniSIFTBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 45.8 | |
| OmniSIFTBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 45.7 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 45.5 | |
| DyCokeBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 45.3 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 44.1 | |
| DyCokeBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 44.1 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 43.8 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 43.3 |