Long Video Understanding on Long Video Bench (Temporal Segments)
66.67Score (15s)-
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| -Ratio=100%, Tokens=21632, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 66.67 | 68.61 | 58.98 | 51.77 | 58.27 | |
| FLoCRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 61.91 | 69.19 | 60.19 | 51.6 | 57.97 | |
| K-meansRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 60.38 | 69.93 | 60.41 | 50.89 | 57.19 | |
| SCRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 60.32 | 68.02 | 59.22 | 52.13 | 57.52 | |
| FLoCRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 60.32 | 68.02 | 58.74 | 48.76 | 55.95 | |
| K-meansRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 59.26 | 66.28 | 58.74 | 49.11 | 55.72 | |
| DivPruneRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 58.2 | 65.12 | 56.55 | 51.42 | 55.72 | |
| TS-LLaVARatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 57.67 | 68.02 | 58.98 | 51.6 | 56.84 | |
| K-meansRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.67 | 67.02 | 57.5 | 46.87 | 54.12 | |
| PoolingRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.61 | 68.61 | 56.31 | 48.05 | 54.45 | |
| LongVURatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.61 | 65.12 | 55.83 | 52.31 | 55.65 | |
| K-medoidsRatio=2^-3, Tokens=2704, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.61 | 68.02 | 57.77 | 50.71 | 55.95 | |
| DivPruneRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.61 | 64.54 | 54.84 | 48.23 | 53.55 | |
| SCRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.61 | 68.02 | 56.07 | 50 | 55.12 | |
| TS-LLaVARatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 56.09 | 68.02 | — | 47.52 | 54.68 | |
| FLoCRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 55.03 | 67.44 | 55.34 | 48.76 | 54.07 | |
| Frame UniformRatio=2^-3, Tokens=2704, Frames=16, Backbone=LLaVA-Video-7B-Qwen22025.10 | 54.5 | 66.38 | 54.37 | 50 | 54.08 | |
| PoolingRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 54.5 | 59.3 | 50.73 | 44.86 | 49.89 | |
| DivPruneRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 53.97 | 55.81 | 51.94 | 46.1 | 50.26 | |
| K-medoidsRatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 53.44 | 66.28 | 54.85 | 47.34 | 52.95 | |
| TS-LLaVARatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 52.38 | 65.12 | 53.64 | 46.45 | 51.91 | |
| LongVURatio=2^-4, Tokens=1352, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 51.85 | 62.21 | 51.21 | 49.65 | 52.06 | |
| K-medoidsRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 51.32 | 63.37 | 55.1 | 46.28 | 51.91 | |
| PoolingRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 50.79 | 56.4 | 49.27 | 43.97 | 48.17 | |
| SCRatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 49.74 | 63.95 | 53.88 | 48.05 | 52.13 | |
| LongVURatio=2^-5, Tokens=676, Frames=128, Backbone=LLaVA-Video-7B-Qwen22025.10 | 49.21 | 56.4 | 48.3 | 48.23 | 49.44 | |
| Frame UniformRatio=2^-4, Tokens=1352, Frames=8, Backbone=LLaVA-Video-7B-Qwen22025.10 | 43.92 | 56.4 | 53.16 | 49.82 | 50.86 | |
| Frame UniformRatio=2^-5, Tokens=676, Frames=4, Backbone=LLaVA-Video-7B-Qwen22025.10 | 43.92 | 51.16 | 51.46 | 46.99 | 48.47 |