| Popis: |
Transforming lo-fi UI sketches to higher-fidelities is an expensive, time-consuming process that requires significant rework. In this paper, we systematically research utilizing AI to assist the transformation of lo-fi sketches to higher fidelities. To provide this assistance, we introduce MetaMorph, an AI tool to detect the constituent UI elements of lo-fi sketches. To train MetaMorph, we collected the UISketch dataset that contains 6,785 hand-drawn sketches of 21 UI elements, 201 hand-drawn lo-fi sketches, and 125,000 synthetically generated lo-fi sketches. MetaMorph provides 63.5% mAP for hand-drawn lo-fi sketches and 82.9% mAP for synthetic lo-fi sketches. Results from ASQ indicate that designers experience an above-average satisfaction level towards ease of task completion (4.9), time taken (5.3), and supporting information (5.3) upon utilizing AI assistance for transforming lo-fi sketches. Their qualitative feedback indicates that they perceive utilizing AI as a novel and useful approach to transform lo-fi sketches into higher fidelities. |
