Temporal Verification on EgoExo-Con 1.0 (test)
92.1Accuracy (Exo)Human
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HumanInput frames=-, Trained on ego data=-2025.10 | 92.1 | 91.3 | 89.4 | |
| Gemini-2.5 FlashInput frames=1 fps, Trained on ego data=-2025.10 | 70.4 | 70.1 | 52.3 | |
| TimeChat-VTInput frames=96, Trained on ego data=✗2025.10 | 62.1 | 61.4 | 42.1 | |
| GPT-5Input frames=32, Trained on ego data=-2025.10 | 60.5 | 61.3 | 52.5 | |
| Video-LLaMA3Input frames=1 fps, Trained on ego data=✓2025.10 | 56.7 | 54.6 | 36.6 | |
| Qwen2.5-VLInput frames=1 fps, Trained on ego data=✗2025.10 | 54.3 | 56.3 | 33 | |
| Video-LLaMA2Input frames=8, Trained on ego data=✓2025.10 | 53.3 | 52.1 | 27.9 | |
| RandomInput frames=-, Trained on ego data=-2025.10 | 50 | 50 | 50 | |
| VTimeLLMInput frames=100, Trained on ego data=✗2025.10 | 48.9 | 48.5 | 23.5 | |
| TimeChatInput frames=96, Trained on ego data=✗2025.10 | 48.9 | 48.4 | 25.1 | |
| TimeSuiteInput frames=128, Trained on ego data=✓2025.10 | 47.4 | 48.5 | 25.6 | |
| VideoChat2Input frames=16, Trained on ego data=✓2025.10 | 46 | 45.1 | 23.4 |