Goal reconstruction on OneStop (New Participant & Item)
54.8Question Word AccuracyDalEye-Llama
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DalEye-LlamaInput Modality=Eye-tracking + Text2025.05 | 54.8 | 42.8 | 24.4 | |
| DalEye-LLaVAInput Modality=Eye-tracking + Text2025.05 | 54.8 | 46 | 23.1 | |
| Text-only Llama 3.1Input Modality=Text-only2025.05 | 54 | 45.3 | 23.8 | |
| Text-only LLaVA OneVisionInput Modality=Text-only2025.05 | 49.8 | 41.2 | 20.7 | |
| DalEye-GPTInput Modality=Eye-tracking + Text2025.05 | 49.6 | 45.4 | 25.6 | |
| Text-only GPT-4o-miniInput Modality=Text-only2025.05 | 47.9 | 42.9 | 23.7 | |
| Gemini few-shotPrompting Strategy=few-shot2025.05 | 41 | 43.9 | 26.8 | |
| Incorrect Human (different critical span)Input Modality=Human Performance2025.05 | 40.3 | 42.3 | 21.9 | |
| Incorrect Human (same critical span)Input Modality=Human Performance2025.05 | 40.1 | 35.7 | 30 | |
| Gemini zero-shotPrompting Strategy=zero-shot2025.05 | 39.3 | 32.3 | 24.2 | |
| Arbitrary Gemini 3Prompting Strategy=Arbitrary2025.05 | 32.8 | 35.5 | 18.4 |