TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Predicting Novel Views Using Generative Adversarial Query Network

Tutkimustuotosvertaisarvioitu

Yksityiskohdat

AlkuperäiskieliEnglanti
OtsikkoImage Analysis - 21st Scandinavian Conference, SCIA 2019, Proceedings
ToimittajatMichael Felsberg, Per-Erik Forssén, Jonas Unger, Ida-Maria Sintorn
KustantajaSpringer Verlag
Sivut16-27
Sivumäärä12
ISBN (painettu)9783030202040
DOI - pysyväislinkit
TilaJulkaistu - 2019
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisussa
TapahtumaScandinavian Conference on Image Analysis - Norrköping, Ruotsi
Kesto: 11 kesäkuuta 201913 kesäkuuta 2019

Julkaisusarja

NimiLecture Notes in Computer Science
Vuosikerta11482
ISSN (painettu)0302-9743
ISSN (elektroninen)1611-3349

Conference

ConferenceScandinavian Conference on Image Analysis
MaaRuotsi
KaupunkiNorrköping
Ajanjakso11/06/1913/06/19

Tiivistelmä

The problem of predicting a novel view of the scene using an arbitrary number of observations is a challenging problem for computers as well as for humans. This paper introduces the Generative Adversarial Query Network (GAQN), a general learning framework for novel view synthesis that combines Generative Query Network (GQN) and Generative Adversarial Networks (GANs). The conventional GQN encodes input views into a latent representation that is used to generate a new view through a recurrent variational decoder. The proposed GAQN builds on this work by adding two novel aspects: First, we extend the current GQN architecture with an adversarial loss function for improving the visual quality and convergence speed. Second, we introduce a feature-matching loss function for stabilizing the training procedure. The experiments demonstrate that GAQN is able to produce high-quality results and faster convergence compared to the conventional approach.

Tutkimusalat

Julkaisufoorumi-taso