Omnimodal Audio & Video Question Answering on AVUTBench
38.8EL ScoreDyCoke (V&A)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DyCoke (V&A)Backbone=Qwen2.5-Omni-7B, Retained Ratio=50%, FLOPs Ratio=44%2025.11 | 38.8 | 67.2 | 58.2 | 81.9 | 39 | 62.4 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio=45%, FLOPs Ratio=39%2025.11 | 38.4 | 67.2 | 56.9 | 85.3 | 42.4 | 66 | |
| Full TokensBackbone=Qwen2.5-Omni-7B, Retained Ratio=100%, FLOPs Ratio=100%2025.11 | 38.2 | 67.8 | 59.6 | 85.6 | 44.1 | 66.7 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio=55%, FLOPs Ratio=48%2025.11 | 38.2 | 64.9 | 55.6 | 80.1 | 34.7 | 65 | |
| FastVBackbone=Qwen2.5-Omni-7B, Retained Ratio=50%, FLOPs Ratio=54%2025.11 | 34.1 | 64.3 | 57.1 | 77.6 | 36.4 | 56.4 | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=29%2025.11 | 34.1 | 67.5 | 54.6 | 83.7 | 42.4 | 61.2 | |
| DyCoke (V&A)Backbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=29%2025.11 | 32.9 | 62.1 | 54.9 | 74.5 | 39 | 58.3 | |
| Full TokensBackbone=Qwen2.5-Omni-3B, Retained Ratio=100%, FLOPs Ratio=100%2025.11 | 32.9 | 65.3 | 58.4 | 85 | 44.1 | 62.6 | |
| DyCoke (V&A)Backbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=26%2025.11 | 32.9 | 62.1 | 54.9 | 74.5 | 38.9 | 58.3 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio=45%, FLOPs Ratio=36%2025.11 | 32.4 | 65 | 57.7 | 84.9 | 41.5 | 61.4 | |
| DyCoke (V&A)Backbone=Qwen2.5-Omni-3B, Retained Ratio=50%, FLOPs Ratio=40%2025.11 | 31.9 | 64.3 | 57.3 | 82.2 | 40.7 | 61.3 | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio=40%, FLOPs Ratio=34%2025.11 | 31.7 | 58.5 | 53.3 | 74.9 | 43.2 | 59 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio=55%, FLOPs Ratio=45%2025.11 | 31.7 | 59.2 | 55.4 | 77.3 | 44.9 | 62.1 | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=26%2025.11 | 28.8 | 63.1 | 58.2 | 84 | 42.4 | 60.4 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio=40%, FLOPs Ratio=31%2025.11 | 28.2 | 60.8 | 54.9 | 73.1 | 42.3 | 61.6 | |
| FastVBackbone=Qwen2.5-Omni-3B, Retained Ratio=50%, FLOPs Ratio=49%2025.11 | 27.1 | 57 | 56.3 | 80.5 | 42.3 | 60.1 | |
| FastVBackbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=37%2025.11 | 24.2 | 60.8 | 54.3 | 81.6 | 40.7 | 58.3 | |
| FastVBackbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=42%2025.11 | 24.1 | 60.7 | 54.3 | 81.6 | 40.7 | 58.3 |