Text to 3D Scene Generation with Rich Lexical Grounding

TL;DR

This paper presents a novel approach for generating 3D scenes from natural language descriptions by learning lexical grounding from a new dataset, improving scene accuracy and plausibility over previous rule-based methods.

Contribution

Introduces a dataset of 3D scenes with natural language annotations and a learning method for lexical grounding, advancing text-to-3D scene generation.

Findings

Grounds a variety of lexical terms to concrete objects

Improves 3D scene generation accuracy over rule-based methods

Develops an automated metric correlating with human judgments

Abstract

The ability to map descriptions of scenes to 3D geometric representations has many applications in areas such as art, education, and robotics. However, prior work on the text to 3D scene generation task has used manually specified object categories and language that identifies them. We introduce a dataset of 3D scenes annotated with natural language descriptions and learn from this data how to ground textual descriptions to physical objects. Our method successfully grounds a variety of lexical terms to concrete referents, and we show quantitatively that our method improves 3D scene generation over previous work using purely rule-based methods. We evaluate the fidelity and plausibility of 3D scenes generated with our grounding approach through human judgments. To ease evaluation on this task, we also introduce an automated metric that strongly correlates with human judgments.