Utilising Domain Transformations in Multi-Camera Re-Identification Scenarios beyond Data Augmentation

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

GANs and CycleGANs, such as CamStyle, have proved to be very effective when augmenting multi-camera datasets in re-identification scenarios. They achieve this by transforming the training images into the domain of each camera. However, this learned domain adaptation is not exploited at the re-identification stage. In this work we propose an extension to CamStyle where the domain transformation is not only used in training for data augmentation, but also further integrated intesting for improving the re-identification performance when different cameras in the scenario are indistinct domains.
Original languageEnglish
Title of host publicationProceedings of the IMVIP 2020 conference 2020
PublisherIrish Pattern Recognition & Classification Society
Pages133-136
Number of pages4
ISBN (Electronic)978-0-9934207-5-7
Publication statusEarly online date - Aug 2020
EventIris Machine Vision and Image Processing Conference - Sligo, Ireland
Duration: 31 Aug 202002 Sep 2020
https://imvipconference.github.io/

Conference

ConferenceIris Machine Vision and Image Processing Conference
Abbreviated titleIMVIP 2020
CountryIreland
CitySligo
Period31/08/202002/09/2020
Internet address

Fingerprint Dive into the research topics of 'Utilising Domain Transformations in Multi-Camera Re-Identification Scenarios beyond Data Augmentation'. Together they form a unique fingerprint.

Cite this