Audio-visual Question Answering on video-SALMONN 2 (test)
29.1Miss RateFull Tokens
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full TokensBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=1002026.02 | 29.1 | 19 | 48.1 | |
| OmniSIFTBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 30.7 | 19.8 | 50.5 | |
| OmniSIFTBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 30.9 | 20.3 | 51.2 | |
| DyCokeBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 32.6 | 20.1 | 52.7 | |
| Full TokensBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=1002026.02 | 32.8 | 20.8 | 53.6 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 33.2 | 19.9 | 53.1 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=352026.02 | 34.1 | 20 | 54.1 | |
| OmniSIFTBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 35.2 | 21.8 | 56.9 | |
| DyCokeBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 35.3 | 20 | 56.3 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 35.8 | 21.4 | 57.2 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=252026.02 | 36.2 | 20.7 | 56.9 | |
| OmniSIFTBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 36.4 | 21.9 | 58.3 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 36.9 | 22.2 | 59.1 | |
| DyCokeBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 36.9 | 21.6 | 58.5 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=352026.02 | 37 | 21.7 | 58.7 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 39.3 | 22.6 | 62 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 39.5 | 22.6 | 62.1 | |
| DyCokeBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=252026.02 | 40.2 | 21.7 | 61.9 |