Multiple-Choice Question Answering on EgoLifeQA
57.5Average ScoreEGAgent Gemini 2.5 Pro (EG + F + T)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| EGAgent Gemini 2.5 Pro (EG + F + T)Category=Ours, # Frames=1FPS→50, Modality=F, C, T2026.01 | 57.5 | 54.4 | 57.1 | 60.3 | 62.4 | 74.6 | 20.6 | — | |
| EGAgent GPT-4.1 (EG + F + T)Category=Ours, # Frames=1FPS→50, Modality=F, C, T2026.01 | 50.7 | 44 | 49.2 | 55.7 | 53.6 | 66.7 | 13.8 | — | |
| EGAgent GPT-4.1 (F + T)Category=Ours, # Frames=1FPS→50, Modality=F, T2026.01 | 48.6 | 48 | 48.4 | 55.7 | 40 | 61.9 | 11.7 | — | |
| Gemini 2.5 ProCategory=MLLMs (Uniform Sampling), # Frames=3000, Modality=F, T2026.01 | 46.8 | 45.6 | 48.4 | 51.7 | 41.6 | 52.4 | 9.9 | — | |
| EGAgent GPT-4o (EG + F + T)Category=Ours, # Frames=1FPS→50, Modality=F, C, T2026.01 | 44.6 | 44.8 | 54.8 | 59 | 44 | 61.9 | 7.7 | — | |
| EgoButler Gemini 1.5 ProCategory=Agentic Baselines, # Frames=0, Modality=C, T2026.01 | 36.9 | 36 | 37.3 | 45.9 | 30.4 | 34.9 | 0 | — | |
| LLaVA-Video-7BCategory=MLLMs (Uniform Sampling), # Frames=64, Modality=F2026.01 | 36.4 | — | — | — | — | — | — | — | |
| EgoButler GPT-4oCategory=Agentic Baselines, # Frames=0, Modality=C, T2026.01 | 36.2 | 34.4 | 42.1 | 29.5 | 30.4 | 44.4 | — | — | |
| GPT-4.1Category=MLLMs (Uniform Sampling), # Frames=1FPS, Modality=C2026.01 | 36 | 32 | 39.7 | 39.3 | 32.8 | 39.7 | — | — | |
| Ego-R1 Qwen-2.5-3B-InstructCategory=Agentic Baselines, # Frames=1FPS, Modality=F, C, T2026.01 | 36 | — | — | — | — | — | — | — | |
| LLaVA-OneVision-7B + T*Category=Agentic Baselines, # Frames=1FPS→8, Modality=F, T2026.01 | 35.4 | — | — | — | — | — | — | — | |
| LLaVA-Video-7B + Video-RAGCategory=RAG, # Frames=64, Modality=F2026.01 | 30 | — | — | — | — | — | — | — | |
| VideoAgentCategory=Agentic Baselines, # Frames=1FPS→8, Modality=F2026.01 | 29.2 | — | — | — | — | — | — | — |