GUI Action Step Prediction on AndroidControl (test)
50Step Accuracy (High)SeeAct-V + UGround-V1-2B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SeeAct-V + UGround-V1-2BInput=Image (SeeAct-V), Planner=GPT-4o, Grounding=UGround-V1-2B2024.10 | 50 | 65 | |
| SeeAct-V + UGround-V1-7BInput=Image (SeeAct-V), Planner=GPT-4o, Grounding=UGround-V1-7B2024.10 | 49.8 | 66.2 | |
| SeeAct-V + UGroundInput=Image (SeeAct-V), Planner=GPT-4, Grounding=UGround2024.10 | 48.4 | 62.4 | |
| GPT-4 + UGroundInput=Text, Planner=GPT-4, Grounding=UGround2024.10 | 46.2 | 58 | |
| GPT-4 + ChoiceInput=Text, Planner=GPT-4, Grounding=Choice2024.10 | 42.1 | 55 | |
| SeeAct-V + SeeClickInput=Image (SeeAct-V), Planner=GPT-4, Grounding=SeeClick2024.10 | 41.8 | 52.8 | |
| GPT-4 + SeeClickInput=Text, Planner=GPT-4, Grounding=SeeClick2024.10 | 39.4 | 47.2 |