Motion-level Perception on MotionBench (test)
58AccuracyQwen2VL-72B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2VL-72BLLM=Qwen2, # Frames=1fps2025.01 | 58 | 58 | 61 | 63 | 72 | 47 | 31 | |
| TE FusionLLM=GLM4-9B, # Frames=162025.01 | 58 | 64 | 59 | 51 | 69 | 41 | 39 | |
| GLM-4V-plusLLM=GLM4, # Frames=302025.01 | 55 | 57 | 57 | 54 | 69 | 40 | 37 | |
| InternVL-40BLLM=NH-2-Yi-34B, # Frames=82025.01 | 54 | 54 | 58 | 49 | 76 | 41 | 30 | |
| MiniCPM-V2.6LLM=Qwen2, # Frames=642025.01 | 53 | 56 | 49 | 45 | 72 | 39 | 33 | |
| Qwen2VL-7BLLM=Qwen2, # Frames=1fps2025.01 | 52 | 52 | 55 | 49 | 68 | 39 | 32 | |
| PLLaVA-34BLLM=Yi-34B, # Frames=162025.01 | 51 | 55 | 51 | 47 | 66 | 38 | 31 | |
| Gemini 1.5 Pro# Frames=1fps2025.01 | 50 | 51 | 52 | 54 | 67 | 40 | 22 | |
| Oryx-34BLLM=Yi-34B, # Frames=642025.01 | 49 | 48 | 52 | 44 | 65 | 42 | 32 | |
| Qwen2VL-2BLLM=Qwen2, # Frames=1fps2025.01 | 47 | 49 | 49 | 42 | 62 | 32 | 28 | |
| CogVLM2-VideoLLM=LLaMA3-8B, # Frames=242025.01 | 44 | 43 | 39 | 38 | 64 | 37 | 33 | |
| LLaVA-NeXTLLM=Yi-34B, # Frames=322025.01 | 40 | 53 | 45 | 36 | 66 | 39 | 23 | |
| GPT-4o2025.01 | 33 | 31 | 34 | 36 | 37 | 42 | 23 | |
| Random2025.01 | 25 | 25 | 25 | 25 | 25 | 25 | 25 |