Audio-Visual Perception on WorldSense
47.4ScoreEchoingPixels
Evaluation Results
| Method | Links | |
|---|---|---|
| EchoingPixelsModel Scale=7B, Token Budget=10%2025.12 | 47.4 | |
| Full Model (Qwen2.5-Omni-7B)Model Scale=7B, Token Budget=100%2025.12 | 46.1 | |
| Full TokensRetained Ratio (%)=100, Model Backbone=Qwen2.5-Omni-7B2026.07 | 45.81 | |
| Full Model (Qwen2.5-Omni-3B)Model Scale=3B, Token Budget=100%2025.12 | 45.4 | |
| Full TokensRetained Ratio (%)=100, Model Backbone=Qwen2.5-Omni-3B2026.07 | 45.21 | |
| EchoingPixelsModel Scale=3B, Token Budget=20%2025.12 | 45 | |
| OmniFocusRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-7B2026.07 | 44.77 | |
| OmniZipRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-7B2026.07 | 44.55 | |
| OmniFocusRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-3B2026.07 | 44.51 | |
| OmniFocusRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-7B2026.07 | 44.23 | |
| RandomRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-7B2026.07 | 44.14 | |
| OmniFocusRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-3B2026.07 | 44.1 | |
| OmniZipRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-3B2026.07 | 43.88 | |
| RandomRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-7B2026.07 | 43.82 | |
| OmniZipRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-7B2026.07 | 43.69 | |
| EchoingPixelsModel Scale=3B, Token Budget=10%2025.12 | 43.5 | |
| RandomRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-3B2026.07 | 43.44 | |
| DyCokeRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-3B2026.07 | 43.32 | |
| DyCokeRetained Ratio (%)=35, Model Backbone=Qwen2.5-Omni-7B2026.07 | 43.28 | |
| DyCokeRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-7B2026.07 | 43.25 | |
| DyCokeRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-3B2026.07 | 43.25 | |
| OmniZipRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-3B2026.07 | 43.06 | |
| RandomRetained Ratio (%)=25, Model Backbone=Qwen2.5-Omni-3B2026.07 | 42.78 | |
| EchoingPixelsModel Scale=3B, Token Budget=5%2025.12 | 40.9 | |
| IntraModalModel Scale=7B, Token Budget=25%2025.12 | 40.6 | |
| IntraModalModel Scale=3B, Token Budget=25%2025.12 | 37.4 |