Deep Perceptual Similarity is Adaptable to Ambiguous Contexts
This work examines the adaptability of Deep Perceptual Similarity (DPS) metrics to context beyond those that align with average human perception and contexts in which the standard metrics have been shown to perform well. Prior works have shown that DPS metrics are good at estimating human perception...
Main Authors: | , , |
---|---|
Format: | Conference Object |
Language: | English |
Published: |
Luleå tekniska universitet, EISLAB
2024
|
Subjects: | |
Online Access: | http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-105093 |
id |
ftluleatu:oai:DiVA.org:ltu-105093 |
---|---|
record_format |
openpolar |
spelling |
ftluleatu:oai:DiVA.org:ltu-105093 2024-05-12T07:57:22+00:00 Deep Perceptual Similarity is Adaptable to Ambiguous Contexts Pihlgren, Gustav Grund Sandin, Fredrik Liwicki, Marcus 2024 application/pdf http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-105093 eng eng LuleÃ¥ tekniska universitet, EISLAB Dept. of Computing Science, UmeÃ¥ University Proceedings of Machine Learning Research Proceedings of Machine Learning Research 233 Proceedings of Machine Learning Research, PMLR : Volume 233: Northern Lights Deep Learning Conference, 9-11 January 2024, UiT The Arctic University, Tromsø, Norway, p. 212-219 orcid:0000-0001-5662-825X orcid:0000-0003-4029-6574 http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-105093 Scopus 2-s2.0-85189301791 info:eu-repo/semantics/openAccess Computer and Information Sciences Data- och informationsvetenskap Electrical Engineering Electronic Engineering Information Engineering Elektroteknik och elektronik Conference paper info:eu-repo/semantics/conferenceObject text 2024 ftluleatu 2024-04-17T14:01:42Z This work examines the adaptability of Deep Perceptual Similarity (DPS) metrics to context beyond those that align with average human perception and contexts in which the standard metrics have been shown to perform well. Prior works have shown that DPS metrics are good at estimating human perception of similarity, so-called perceptual similarity. However, it remains unknown whether such metrics can be adapted to other contexts. In this work, DPS metrics are evaluated for their adaptability to different contradictory similarity contexts. Such contexts are created by randomly ranking six image distortions. Metrics are adapted to consider distortions more or less disruptive to similarity depending on their place in the random rankings. This is done by training pretrained CNNs to measure similarity according to given contexts. The adapted metrics are also evaluated on a perceptual similarity dataset to evaluate whether adapting to a ranking affects their prior performance. The findings show that DPS metrics can be adapted with high performance. While the adapted metrics have difficulties with the same contexts as baselines, performance is improved in 99% of cases. Finally, it is shown that the adaption is not significantly detrimental to prior performance on perceptual similarity. The implementation of this work is available online. Full text license: CC BY 4.0; Conference Object Arctic Luleå University of Technology Publications (DiVA) |
institution |
Open Polar |
collection |
Luleå University of Technology Publications (DiVA) |
op_collection_id |
ftluleatu |
language |
English |
topic |
Computer and Information Sciences Data- och informationsvetenskap Electrical Engineering Electronic Engineering Information Engineering Elektroteknik och elektronik |
spellingShingle |
Computer and Information Sciences Data- och informationsvetenskap Electrical Engineering Electronic Engineering Information Engineering Elektroteknik och elektronik Pihlgren, Gustav Grund Sandin, Fredrik Liwicki, Marcus Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
topic_facet |
Computer and Information Sciences Data- och informationsvetenskap Electrical Engineering Electronic Engineering Information Engineering Elektroteknik och elektronik |
description |
This work examines the adaptability of Deep Perceptual Similarity (DPS) metrics to context beyond those that align with average human perception and contexts in which the standard metrics have been shown to perform well. Prior works have shown that DPS metrics are good at estimating human perception of similarity, so-called perceptual similarity. However, it remains unknown whether such metrics can be adapted to other contexts. In this work, DPS metrics are evaluated for their adaptability to different contradictory similarity contexts. Such contexts are created by randomly ranking six image distortions. Metrics are adapted to consider distortions more or less disruptive to similarity depending on their place in the random rankings. This is done by training pretrained CNNs to measure similarity according to given contexts. The adapted metrics are also evaluated on a perceptual similarity dataset to evaluate whether adapting to a ranking affects their prior performance. The findings show that DPS metrics can be adapted with high performance. While the adapted metrics have difficulties with the same contexts as baselines, performance is improved in 99% of cases. Finally, it is shown that the adaption is not significantly detrimental to prior performance on perceptual similarity. The implementation of this work is available online. Full text license: CC BY 4.0;Â |
format |
Conference Object |
author |
Pihlgren, Gustav Grund Sandin, Fredrik Liwicki, Marcus |
author_facet |
Pihlgren, Gustav Grund Sandin, Fredrik Liwicki, Marcus |
author_sort |
Pihlgren, Gustav Grund |
title |
Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
title_short |
Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
title_full |
Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
title_fullStr |
Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
title_full_unstemmed |
Deep Perceptual Similarity is Adaptable to Ambiguous Contexts |
title_sort |
deep perceptual similarity is adaptable to ambiguous contexts |
publisher |
Luleå tekniska universitet, EISLAB |
publishDate |
2024 |
url |
http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-105093 |
genre |
Arctic |
genre_facet |
Arctic |
op_relation |
Proceedings of Machine Learning Research 233 Proceedings of Machine Learning Research, PMLR : Volume 233: Northern Lights Deep Learning Conference, 9-11 January 2024, UiT The Arctic University, Tromsø, Norway, p. 212-219 orcid:0000-0001-5662-825X orcid:0000-0003-4029-6574 http://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-105093 Scopus 2-s2.0-85189301791 |
op_rights |
info:eu-repo/semantics/openAccess |
_version_ |
1798837729930248192 |