Sequential Gallery for Interactive Visual Design Optimization

摘要

Visual design tasks often involve tuning many design parameters. For example,color grading of a photograph involves many parameters, some of whichnon-expert users might be unfamiliar with. We propose a novel user-in-theloopoptimization method that allows users to efficiently find an appropriateparameter set by exploring such a high-dimensional design space throughmuch easier two-dimensional search subtasks. This method, called sequentialplane search, is based on Bayesian optimization to keep necessary queriesto users as few as possible. To help users respond to plane-search queries,we also propose using a gallery-based interface that provides options inthe two-dimensional subspace arranged in an adaptive grid view. We callthis interactive framework Sequential Gallery since users sequentially selectthe best option from the options provided by the interface. Our experimentwith synthetic functions shows that our sequential plane search can findsatisfactory solutions in fewer iterations than baselines.We also conducted apreliminary user study, results of which suggest that novices can effectivelycomplete search tasks with Sequential Gallery in a photo-enhancementscenario.