Video Understanding on MSRVTT
57.7AccCDPruner
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CDPrunerRetained Tokens=256, Reduction Ratio=87.5%2026.04 | 57.7 | 3.52 | — | — | |
| Video-LLaVARetained Tokens=2048, Reduction Ratio=0%2026.04 | 57.5 | 3.5 | — | — | |
| Video-LLaVA-7BBase Model=Video-LLaVA-7B2026.04 | 57.3 | 3.5 | — | — | |
| CDPrunerRetained Tokens=128, Reduction Ratio=93.75%2026.04 | 57.2 | 3.48 | — | — | |
| VanillaBackbone=Video-LLaVA-7B, Retained Tokens=20482026.02 | 57.1 | 3.49 | — | — | |
| ID-SelectionRetained Tokens=256, Reduction Ratio=87.5%, Importance Scoring=unified score (Eq. 6)2026.04 | 57.1 | 3.49 | — | — | |
| DivPruneRetained Tokens=128, Reduction Ratio=93.75%2026.04 | 57.1 | 3.51 | — | — | |
| VanillaBackbone=Video-LLaVA-7B, Original visual tokens=2048, Retained visual tokens=256, Reduction ratio=87.5%2026.05 | 57.1 | 3.49 | 100 | 100 | |
| Video-LLaVARetained Tokens=20482026.06 | 57.1 | 3.49 | — | — | |
| ID-SelectionRetained Tokens=128, Reduction Ratio=93.75%, Importance Scoring=unified score (Eq. 6)2026.04 | 57 | 3.49 | — | — | |
| ApETBackbone=Video-LLaVA-7B, Retained Tokens=2562026.02 | 56.9 | 3.47 | — | — | |
| DivPruneRetained Tokens=256, Reduction Ratio=87.5%2026.04 | 56.9 | 3.52 | — | — | |
| Video-LLaVA-7B + HalfVBase Model=Video-LLaVA-7B, Technique=HalfV2026.04 | 56.9 | 3.6 | — | — | |
| Video-LLaVA-7B + DARTBase Model=Video-LLaVA-7B, Technique=DART2026.04 | 56.7 | 3.6 | — | — | |
| Video-LLaVA2026.05 | 56.7 | — | — | — | |
| Video-LLaVA-7B + HoloVBase Model=Video-LLaVA-7B, Technique=HoloV2026.04 | 56.5 | 3.6 | — | — | |
| ID-SelectionRetained Tokens=256, Reduction Ratio=87.5%, Importance Scoring=visual encoder [CLS] attention2026.04 | 56.2 | 3.48 | — | — | |
| ID-SelectionRetained Tokens=128, Reduction Ratio=93.75%, Importance Scoring=visual encoder [CLS] attention2026.04 | 56.1 | 3.45 | — | — | |
| OccamToken2026.05 | 56.1 | — | — | — | |
| VisPrunerRetained Tokens=256, Reduction Ratio=87.5%2026.04 | 56 | 3.44 | — | — | |
| EvoCutRetained Tokens=2562026.06 | 55.8 | 3.46 | — | — | |
| LRCPBackbone=Video-LLaVA-7B, Original visual tokens=2048, Retained visual tokens=256, Reduction ratio=87.5%2026.05 | 55.6 | 3.45 | 97.4 | 98.9 | |
| FasterVLMRetained Tokens=256, Reduction Ratio=87.5%2026.04 | 55.3 | 3.42 | — | — | |
| Video-LLaVA-7B + FastVBase Model=Video-LLaVA-7B, Technique=FastV2026.04 | 55 | 3.5 | — | — | |
| VisPrunerRetained Tokens=128, Reduction Ratio=93.75%2026.04 | 54.7 | 3.45 | — | — | |
| VisionZipBackbone=Video-LLaVA-7B, Retained Tokens=2562026.02 | 54.5 | 3.43 | — | — | |
| VisionZipBackbone=Video-LLaVA-7B, Original visual tokens=2048, Retained visual tokens=256, Reduction ratio=87.5%2026.05 | 54.5 | 3.43 | 95.4 | 98.3 | |
| VisionZipRetained Tokens=2562026.06 | 54.5 | 3.43 | — | — | |
| FastVRetained Tokens=256, Reduction Ratio=87.5%2026.04 | 52.9 | 3.38 | — | — | |
| VisionZip2026.05 | 52.1 | — | — | — | |
| FasterVLMRetained Tokens=128, Reduction Ratio=93.75%2026.04 | 52 | 3.34 | — | — | |
| FastVRetained Tokens=128, Reduction Ratio=93.75%2026.04 | 50.9 | 3.29 | — | — | |
| CLSE-MEvaluator=Gemini-2.5-Flash, Pruning Strategy=w/ Token Merging, Retained Tokens=1942026.06 | 49.5 | 2.35 | — | — | |
| Video-ChatGPT2026.04 | 49.3 | 2.8 | — | — | |
| SparseVLMEvaluator=Gemini-2.5-Flash, Pruning Strategy=w/ Token Merging, Retained Tokens=1942026.06 | 48 | 2.38 | — | — | |
| CLSEEvaluator=Gemini-2.5-Flash, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 47.7 | 2.27 | — | — | |
| Video-LLaVA-7BEvaluator=Gemini-2.5-Flash, Pruning Strategy=None, Retained Tokens=20482026.06 | 47.5 | 2.32 | — | — | |
| VideoChat-7B2026.04 | 45 | 2.5 | — | — | |
| FiCoCo-VEvaluator=Gemini-2.5-Flash, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 44.3 | 2.1 | — | — | |
| LLaMA-Adapter-7B2026.04 | 43.8 | 2.7 | — | — | |
| Video-LLaVA-7BEvaluator=GPT-4o-mini, Pruning Strategy=None, Retained Tokens=20482026.06 | 42.7 | 2.69 | — | — | |
| CLSE-MEvaluator=GPT-4o-mini, Pruning Strategy=w/ Token Merging, Retained Tokens=1942026.06 | 42.2 | 2.61 | — | — | |
| SparseVLMEvaluator=GPT-4o-mini, Pruning Strategy=w/ Token Merging, Retained Tokens=1942026.06 | 41.9 | 2.63 | — | — | |
| PDropBackbone=Video-LLaVA-7B, Retained Tokens=2562026.02 | 41.8 | 3.19 | — | — | |
| PDropBackbone=Video-LLaVA-7B, Original visual tokens=2048, Retained visual tokens=256, Reduction ratio=87.5%2026.05 | 41.8 | 3.19 | 73.2 | 91.4 | |
| PDropRetained Tokens=2562026.06 | 41.8 | 3.19 | — | — | |
| FastVBackbone=Video-LLaVA-7B, Retained Tokens=2562026.02 | 40.6 | 3.18 | — | — | |
| FastVRetained Tokens=2562026.06 | 40.6 | 3.18 | — | — | |
| CLSEEvaluator=GPT-4o-mini, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 39.3 | 2.56 | — | — | |
| FiCoCo-VEvaluator=GPT-4o-mini, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 37.6 | 2.47 | — | — | |
| PDropEvaluator=Gemini-2.5-Flash, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 34.6 | 1.65 | — | — | |
| FastVEvaluator=Gemini-2.5-Flash, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 34.3 | 1.69 | — | — | |
| SparseVLMBackbone=Video-LLaVA-7B, Retained Tokens=2562026.02 | 32.9 | 3.02 | — | — | |
| SparseVLMRetained Tokens=2562026.06 | 32.9 | 3.02 | — | — | |
| FastVEvaluator=GPT-4o-mini, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 31.4 | 2.19 | — | — | |
| SparseVLM2026.05 | 31 | — | — | — | |
| PDropEvaluator=GPT-4o-mini, Pruning Strategy=Hard Pruning, Retained Tokens=1942026.06 | 27.9 | 1.95 | — | — |