ExploreTree: Interactive tree modeling in semantic trait space with online intent learning

Abstract

Perceptually modeling realistic trees is important for many graphics applications. However, existing methods are mainly rule-based. Few have directly associated control parameters with user modeling intent and semantic tree shape descriptions. In this paper, we propose a new interactive tree modeling system, ExploreTree, that automatically deduces user modeling intent and supports iteratively design of 3D tree models. It consists of two major phases. The first phase is an off-line learning process, where semantic tree traits perceived by humans are learned. Crowdsourced data on example tree models are collected and analyzed to construct the semantic trait space as well as the embedding of trees into this space. Built upon it, the second phase is an interactive exploration of tree models via a few user clicks, where a user intent evaluation model is learned online to guide the modeling process. Modeled trees and user studies demonstrate the efficiency and capability of ExploreTree.