Convex decomposition of indoor scenes

We describe a method to parse a complex, cluttered indoor scene into primitives which offer a parsimonious abstraction of scene structure. Our primitives are simple convexes. Our method uses a learned regression procedure to predict a parse into a fixed number of convexes from a depth map, but does not require labeled data. The result is then polished with a descent method which adjusts the convexes to produce a very good fit, and greedily removes superfluous primitives. Because the entire scene is parsed, we can evaluate using traditional depth and normal error metrics. Our evaluation procedure demonstrates that the error in our primitive representation is comparable to that of predicting depth from a single image. We show that our primitives segment the scene well, without ever having seen segmentation labels. Finally, we show that our primitive representation is stable under change of viewpoint.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Vavilala, Vaibhav

Contributors dc:contributor

• Forsyth, David A

• Deep Learning
• Tensorflow
• Scene Parsing

Statement dc:rights

• Copyright 2022 Vaibhav Vavilala

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/117609

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F117609