Long-form Egocentric Video Understanding on EgoSchema
78.2AccuracyVideoARM
Evaluation Results
| Method | Links | |
|---|---|---|
| VideoARMBackbone Configuration=o3+gpt-4.12025.12 | 78.2 | |
| DVDMethod Type=Other Training-Free Approaches2025.12 | 76.6 | |
| Qwen2.5-VL-72BMethod Type=Open-Source VLMs2025.12 | 76.2 | |
| VideoARMBackbone Configuration=o3+gpt-4o2025.12 | 76.2 | |
| EgoGPT#Param=7B, #Frames=32, training_dataset=EgoIT+EgoLifeD12025.03 | 75.4 | |
| VCAMethod Type=Other Training-Free Approaches2025.12 | 73.6 | |
| EgoGPT#Param=7B, #Frames=32, training_dataset=EgoIT2025.03 | 73.2 | |
| MR. VideoMethod Type=Other Training-Free Approaches2025.12 | 73 | |
| Gemini-1.5-Pro#Frames=322025.03 | 72.2 | |
| GPT-4o#Frames=322025.03 | 72.2 | |
| GPT-4oMethod Type=Commercial VLMs2025.12 | 72.2 | |
| Gemini-2.0-FlashMethod Type=Commercial VLMs2025.12 | 71.2 | |
| Gemini-1.5-ProMethod Type=Commercial VLMs2025.12 | 71.1 | |
| VideoTreeMethod Type=Other Training-Free Approaches2025.12 | 67 | |
| Qwen2-VL#Param=7B, #Frames=322025.03 | 66.7 | |
| OpenAI o3Method Type=Commercial VLMs2025.12 | 63.2 | |
| VideoAgentMethod Type=Other Training-Free Approaches2025.12 | 63.2 | |
| LLaVA-OV#Param=7B, #Frames=322025.03 | 60.1 | |
| LLaVA-Videos#Param=7B, #Frames=322025.03 | 57.3 | |
| GPT-4v#Frames=322025.03 | 56.6 | |
| Oryx#Param=7B, #Frames=322025.03 | 56 | |
| InternVideo2#Param=8B, #Frames=322025.03 | 55.2 | |
| IXC-2.5#Param=7B, #Frames=322025.03 | 54.6 | |
| LLaVA-Next-Video#Param=7B, #Frames=322025.03 | 49.7 | |
| LongVA#Param=7B, #Frames=322025.03 | 44.1 |