ReGroup: Recursive Neural Networks for Hierarchical Grouping of Vector Graphic Primitives

TL;DR

This paper introduces ReGroup, a recursive neural network approach that learns hierarchical groupings of vector graphic primitives to improve selection functionality, reducing ambiguity and reliance on heuristics.

Contribution

It presents a novel data-centric method using recursive neural networks to learn hierarchical groupings of vector primitives for better selection in vector graphics.

Findings

Successfully trained on a custom hierarchy dataset

Demonstrated improved grouping accuracy over heuristic methods

Enabled a prototype selection tool based on learned hierarchies

Abstract

Selection functionality is as fundamental to vector graphics as it is for raster data. But vector selection is quite different: instead of pixel-level labeling, we make a binary decision to include or exclude each vector primitive. In the absence of intelligible metadata, this becomes a perceptual grouping problem. These have previously relied on heuristics derived from empirical principles like Gestalt Theory, but since these are ill-defined and subjective, they often result in ambiguity. Here we take a data-centric approach to the problem. By exploiting the recursive nature of perceptual grouping, we interpret the task as constructing a hierarchy over the primitives of a vector graphic, which is amenable to learning with recursive neural networks with few human annotations. We verify this by building a dataset of these hierarchies on which we train a hierarchical grouping network. We then demonstrate how this can underpin a prototype selection tool.