CSC-GAN: Cycle and Semantic Consistency for Dataset Augmentation

Renato Barros Arantes*, George Vogiatzis, Diego R. Faria

*Corresponding author for this work

Research output: Chapter in Book/Published conference outputConference publication

Abstract

Image-to-image translation is a computer vision problem where a task learns a mapping from a source domain A to a target domain B using a training set. However, this translation is not always accurate, and during the translation process, relevant semantic information can deteriorate. To handle this problem, we propose a new cycle-consistent, adversarially trained image-to-image translation with a loss function that is constrained by semantic segmentation. This formulation encourages the model to preserve semantic information during the translation process. For this purpose, our loss function evaluates the accuracy of the synthetically generated image against a semantic segmentation model, previously trained. Reported results show that our proposed method can significantly increase the level of details in the synthetic images. We further demonstrate our method’s effectiveness by applying it as a dataset augmentation technique, for a minimal dataset, showing that it can improve the semantic segmentation accuracy.

Original languageEnglish
Title of host publicationAdvances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings
EditorsGeorge Bebis, Zhaozheng Yin, Edward Kim, Jan Bender, Kartic Subr, Bum Chul Kwon, Jian Zhao, Denis Kalkofen, George Baciu
PublisherSpringer
Pages170-181
Number of pages12
Volume12509
ISBN (Electronic)9783030645564
ISBN (Print)9783030645557
DOIs
Publication statusPublished - 7 Dec 2020
Event15th International Symposium on Visual Computing, ISVC 2020 - San Diego, United States
Duration: 5 Oct 20207 Oct 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12509 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th International Symposium on Visual Computing, ISVC 2020
Country/TerritoryUnited States
CitySan Diego
Period5/10/207/10/20

Bibliographical note

© Springer Nature B.V. 2020. The final publication is available at Springer via http://dx.doi.org/10.1007/978-3-030-64556-4_14

Keywords

  • Dataset augmentation
  • GAN
  • Semantic segmentation

Fingerprint

Dive into the research topics of 'CSC-GAN: Cycle and Semantic Consistency for Dataset Augmentation'. Together they form a unique fingerprint.

Cite this