The success of various applications including robotics, digital content
creation, and visualization demand a structured and abstract representation of
the 3D world from limited sensor data. Inspired by the nature of human
perception of 3D shapes as a collection of simple parts, we explore such an
abstract shape representation based on primitives. Given a single depth image
of an object, we present 3D-PRNN, a generative recurrent neural network that
synthesizes multiple plausible shapes composed of a set of primitives. Our
generative model encodes symmetry characteristics of common man-made objects,
preserves long-range structural coherence, and describes objects of varying
complexity with a compact representation. We also propose a method based on
Gaussian Fields to generate a large scale dataset of primitive-based shape
representations to train our network. We evaluate our approach on a wide range
of examples and show that it outperforms nearest-neighbor based shape retrieval
methods and is on-par with voxel-based generative models while using a
significantly reduced parameter space.
Description
[1708.01648] 3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks
%0 Generic
%1 zou20173dprnn
%A Zou, Chuhang
%A Yumer, Ersin
%A Yang, Jimei
%A Ceylan, Duygu
%A Hoiem, Derek
%D 2017
%K 2017 3D arxiv computer-vision deep-learning iccv paper rnn shape
%T 3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks
%U http://arxiv.org/abs/1708.01648
%X The success of various applications including robotics, digital content
creation, and visualization demand a structured and abstract representation of
the 3D world from limited sensor data. Inspired by the nature of human
perception of 3D shapes as a collection of simple parts, we explore such an
abstract shape representation based on primitives. Given a single depth image
of an object, we present 3D-PRNN, a generative recurrent neural network that
synthesizes multiple plausible shapes composed of a set of primitives. Our
generative model encodes symmetry characteristics of common man-made objects,
preserves long-range structural coherence, and describes objects of varying
complexity with a compact representation. We also propose a method based on
Gaussian Fields to generate a large scale dataset of primitive-based shape
representations to train our network. We evaluate our approach on a wide range
of examples and show that it outperforms nearest-neighbor based shape retrieval
methods and is on-par with voxel-based generative models while using a
significantly reduced parameter space.
@misc{zou20173dprnn,
abstract = {The success of various applications including robotics, digital content
creation, and visualization demand a structured and abstract representation of
the 3D world from limited sensor data. Inspired by the nature of human
perception of 3D shapes as a collection of simple parts, we explore such an
abstract shape representation based on primitives. Given a single depth image
of an object, we present 3D-PRNN, a generative recurrent neural network that
synthesizes multiple plausible shapes composed of a set of primitives. Our
generative model encodes symmetry characteristics of common man-made objects,
preserves long-range structural coherence, and describes objects of varying
complexity with a compact representation. We also propose a method based on
Gaussian Fields to generate a large scale dataset of primitive-based shape
representations to train our network. We evaluate our approach on a wide range
of examples and show that it outperforms nearest-neighbor based shape retrieval
methods and is on-par with voxel-based generative models while using a
significantly reduced parameter space.},
added-at = {2018-04-05T13:47:43.000+0200},
author = {Zou, Chuhang and Yumer, Ersin and Yang, Jimei and Ceylan, Duygu and Hoiem, Derek},
biburl = {https://www.bibsonomy.org/bibtex/279bd123c63bb4d6def2179108207906d/achakraborty},
description = {[1708.01648] 3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks},
interhash = {c6e35e4a7de1d8a7ac96a4e357f681bb},
intrahash = {79bd123c63bb4d6def2179108207906d},
keywords = {2017 3D arxiv computer-vision deep-learning iccv paper rnn shape},
note = {cite arxiv:1708.01648Comment: ICCV 2017},
timestamp = {2018-04-05T13:48:13.000+0200},
title = {3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks},
url = {http://arxiv.org/abs/1708.01648},
year = 2017
}