Look Over Here: Attention-Directing Composition of Manga Elements Supplementary Material

Annotation. We manually segmented and annotated all the pages in our dataset. Each page was segmented into a set of panels. For each panel, we first manually label its shot type (i.e., long, medium, close-up, or big close-up) and segmented the foreground subjects and their corresponding balloons. Given a set of segmented panels across all the pages, we then partitioned the panels into three groups with similar geometric features, which include aspect ratio (i.e., width/height) and size (i.e., area). The clustering was done using a Gaussian mixture model (GMM), initialized by k-means clustering. Grouping geometrically similar panels allows the probabilistic model to learn composition patterns that vary with the panel shape.