Learning shape placements by example

Abstract

KAUST propagate propagateFigure 1: New York-style skyscrapers generated by our method: Individual shape placement operations are learned during an example modeling session. Two of these operations are shown in the top row. New skyscraper variations like the ones in the bottom row can be generated from the learned model without additional user input. AbstractWe present a method to learn and propagate shape placements in 2D polygonal scenes from a few examples provided by a user. The placement of a shape is modeled as an oriented bounding box. Simple geometric relationships between this bounding box and nearby scene polygons define a feature set for the placement. The feature sets of all example placements are then used to learn a probabilistic model over all possible placements and scenes. With this model, we can generate a new set of placements with similar geometric relationships in any given scene. We introduce extensions that enable propagation and generation of shapes in 3D scenes, as well as the application of a learned modeling session to large scenes without additional user interaction. These concepts allow us to generate complex scenes with thousands of objects with relatively little user interaction.

A probabilistic framework for component‐based vector graphics

“…Guerro et al propose a probabilistic model for propagating shape placements in 2D polygonal scenes [GJWW15]. Guerro et al propose a probabilistic model for propagating shape placements in 2D polygonal scenes [GJWW15].…”

“…While our method in general produces shapes that are deemed plausible (Figure 12), we did notice a slight preference of the original images over our method (but not significant). Guerrero et al solve this problem by kernel regression over shape placements, assuming shapes to be defined as simple polygons with known orientations [GJWW15]. Most noticeable are cases where the placement of components are not aligned properly with either parent components (car case in the figure) or with adjacent components (chair case).…”

Learning 3D Scene Synthesis from Annotated RGB‐D Images

“…Edit and shape placement propagation. Employing geometric relations as the building blocks, both works in [GJWW14, GJWW15] intend to propagate edit operations in the 2D domain. In [GJWW14], the edit propagation is applied to similar parts based on a set of geometric relationships.…”

“…In [GJWW14], the edit propagation is applied to similar parts based on a set of geometric relationships. By incorporating example 2D scenes for guiding the shape placement, a more recent work [GJWW15] learns a probabilistic model based on the feature sets of geometric relations in example placements, which further enables generating novel similar placements. Although the second method can be extended to 3D scenes, both approaches are originally developed for 2D polygons.…”

Retargeting 3D Objects and Scenes with a General Framework

“…These methods take advantage of exemplar shapes to generate large amount of shape variations by using procedural modeling [PM01, WWSR03, MWH * 06, YYT * 11, TLL * 11, VGDA * 12, BSW13], inverse procedural modeling [BWS10,TYK * 12,KWS16], and probabilistic model and inference [CKGK11,KCKK12,FRS * 12,MSK10,YYW * 12,PYW14, LVW * 15, GJWW15]. These methods take advantage of exemplar shapes to generate large amount of shape variations by using procedural modeling [PM01, WWSR03, MWH * 06, YYT * 11, TLL * 11, VGDA * 12, BSW13], inverse procedural modeling [BWS10,TYK * 12,KWS16], and probabilistic model and inference [CKGK11,KCKK12,FRS * 12,MSK10,YYW * 12,PYW14, LVW * 15, GJWW15].…”