Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning

Deep learning based object detectors struggle generalizing to a new target domain bearing significant variations in object and background. Most current methods align domains by using image or instance-level adversarial feature alignment. This often suffers due to unwanted background and lacks class-...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2023-11
Hauptverfasser: Muhammad Akhtar Munir, Khan, Muhammad Haris, M Saquib Sarfraz, Ali, Mohsen
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title arXiv.org
container_volume
creator Muhammad Akhtar Munir
Khan, Muhammad Haris
M Saquib Sarfraz
Ali, Mohsen
description Deep learning based object detectors struggle generalizing to a new target domain bearing significant variations in object and background. Most current methods align domains by using image or instance-level adversarial feature alignment. This often suffers due to unwanted background and lacks class-specific alignment. A straightforward approach to promote class-level alignment is to use high confidence predictions on unlabeled domain as pseudo-labels. These predictions are often noisy since model is poorly calibrated under domain shift. In this paper, we propose to leverage model's predictive uncertainty to strike the right balance between adversarial feature alignment and class-level alignment. We develop a technique to quantify predictive uncertainty on class assignments and bounding-box predictions. Model predictions with low uncertainty are used to generate pseudo-labels for self-training, whereas the ones with higher uncertainty are used to generate tiles for adversarial feature alignment. This synergy between tiling around uncertain object regions and generating pseudo-labels from highly certain object regions allows capturing both image and instance-level context during the model adaptation. We report thorough ablation study to reveal the impact of different components in our approach. Results on five diverse and challenging adaptation scenarios show that our approach outperforms existing state-of-the-art methods with noticeable margins.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2887705791</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2887705791</sourcerecordid><originalsourceid>FETCH-proquest_journals_28877057913</originalsourceid><addsrcrecordid>eNqNjNEKgjAYhUcQJOU7DLoW5sy0y8yii6CLvA3509-Y2GbbtNdvQg_Q1QfnO-fMiMejKAzSDecL4hvTMsb4NuFxHHnknqsXCEn3NfRWjEivjxYrS3O0DkJJOgqgGXQgKyGfNEP7QZT0hl0TFNpNpxRk7R5G1Aa0gI5eEPQkVmTeQGfQ_3FJ1qdjcTgHvVbvAY0tWzVo6VTJ0zRJWJzswui_1he4tUNX</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2887705791</pqid></control><display><type>article</type><title>Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning</title><source>Free E- Journals</source><creator>Muhammad Akhtar Munir ; Khan, Muhammad Haris ; M Saquib Sarfraz ; Ali, Mohsen</creator><creatorcontrib>Muhammad Akhtar Munir ; Khan, Muhammad Haris ; M Saquib Sarfraz ; Ali, Mohsen</creatorcontrib><description>Deep learning based object detectors struggle generalizing to a new target domain bearing significant variations in object and background. Most current methods align domains by using image or instance-level adversarial feature alignment. This often suffers due to unwanted background and lacks class-specific alignment. A straightforward approach to promote class-level alignment is to use high confidence predictions on unlabeled domain as pseudo-labels. These predictions are often noisy since model is poorly calibrated under domain shift. In this paper, we propose to leverage model's predictive uncertainty to strike the right balance between adversarial feature alignment and class-level alignment. We develop a technique to quantify predictive uncertainty on class assignments and bounding-box predictions. Model predictions with low uncertainty are used to generate pseudo-labels for self-training, whereas the ones with higher uncertainty are used to generate tiles for adversarial feature alignment. This synergy between tiling around uncertain object regions and generating pseudo-labels from highly certain object regions allows capturing both image and instance-level context during the model adaptation. We report thorough ablation study to reveal the impact of different components in our approach. Results on five diverse and challenging adaptation scenarios show that our approach outperforms existing state-of-the-art methods with noticeable margins.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Ablation ; Adaptation ; Alignment ; Deep learning ; Labels ; Object recognition ; Predictions ; Tiling ; Uncertainty</subject><ispartof>arXiv.org, 2023-11</ispartof><rights>2023. This work is published under http://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>776,780</link.rule.ids></links><search><creatorcontrib>Muhammad Akhtar Munir</creatorcontrib><creatorcontrib>Khan, Muhammad Haris</creatorcontrib><creatorcontrib>M Saquib Sarfraz</creatorcontrib><creatorcontrib>Ali, Mohsen</creatorcontrib><title>Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning</title><title>arXiv.org</title><description>Deep learning based object detectors struggle generalizing to a new target domain bearing significant variations in object and background. Most current methods align domains by using image or instance-level adversarial feature alignment. This often suffers due to unwanted background and lacks class-specific alignment. A straightforward approach to promote class-level alignment is to use high confidence predictions on unlabeled domain as pseudo-labels. These predictions are often noisy since model is poorly calibrated under domain shift. In this paper, we propose to leverage model's predictive uncertainty to strike the right balance between adversarial feature alignment and class-level alignment. We develop a technique to quantify predictive uncertainty on class assignments and bounding-box predictions. Model predictions with low uncertainty are used to generate pseudo-labels for self-training, whereas the ones with higher uncertainty are used to generate tiles for adversarial feature alignment. This synergy between tiling around uncertain object regions and generating pseudo-labels from highly certain object regions allows capturing both image and instance-level context during the model adaptation. We report thorough ablation study to reveal the impact of different components in our approach. Results on five diverse and challenging adaptation scenarios show that our approach outperforms existing state-of-the-art methods with noticeable margins.</description><subject>Ablation</subject><subject>Adaptation</subject><subject>Alignment</subject><subject>Deep learning</subject><subject>Labels</subject><subject>Object recognition</subject><subject>Predictions</subject><subject>Tiling</subject><subject>Uncertainty</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>BENPR</sourceid><recordid>eNqNjNEKgjAYhUcQJOU7DLoW5sy0y8yii6CLvA3509-Y2GbbtNdvQg_Q1QfnO-fMiMejKAzSDecL4hvTMsb4NuFxHHnknqsXCEn3NfRWjEivjxYrS3O0DkJJOgqgGXQgKyGfNEP7QZT0hl0TFNpNpxRk7R5G1Aa0gI5eEPQkVmTeQGfQ_3FJ1qdjcTgHvVbvAY0tWzVo6VTJ0zRJWJzswui_1he4tUNX</recordid><startdate>20231108</startdate><enddate>20231108</enddate><creator>Muhammad Akhtar Munir</creator><creator>Khan, Muhammad Haris</creator><creator>M Saquib Sarfraz</creator><creator>Ali, Mohsen</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20231108</creationdate><title>Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning</title><author>Muhammad Akhtar Munir ; Khan, Muhammad Haris ; M Saquib Sarfraz ; Ali, Mohsen</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28877057913</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Ablation</topic><topic>Adaptation</topic><topic>Alignment</topic><topic>Deep learning</topic><topic>Labels</topic><topic>Object recognition</topic><topic>Predictions</topic><topic>Tiling</topic><topic>Uncertainty</topic><toplevel>online_resources</toplevel><creatorcontrib>Muhammad Akhtar Munir</creatorcontrib><creatorcontrib>Khan, Muhammad Haris</creatorcontrib><creatorcontrib>M Saquib Sarfraz</creatorcontrib><creatorcontrib>Ali, Mohsen</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Muhammad Akhtar Munir</au><au>Khan, Muhammad Haris</au><au>M Saquib Sarfraz</au><au>Ali, Mohsen</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning</atitle><jtitle>arXiv.org</jtitle><date>2023-11-08</date><risdate>2023</risdate><eissn>2331-8422</eissn><abstract>Deep learning based object detectors struggle generalizing to a new target domain bearing significant variations in object and background. Most current methods align domains by using image or instance-level adversarial feature alignment. This often suffers due to unwanted background and lacks class-specific alignment. A straightforward approach to promote class-level alignment is to use high confidence predictions on unlabeled domain as pseudo-labels. These predictions are often noisy since model is poorly calibrated under domain shift. In this paper, we propose to leverage model's predictive uncertainty to strike the right balance between adversarial feature alignment and class-level alignment. We develop a technique to quantify predictive uncertainty on class assignments and bounding-box predictions. Model predictions with low uncertainty are used to generate pseudo-labels for self-training, whereas the ones with higher uncertainty are used to generate tiles for adversarial feature alignment. This synergy between tiling around uncertain object regions and generating pseudo-labels from highly certain object regions allows capturing both image and instance-level context during the model adaptation. We report thorough ablation study to reveal the impact of different components in our approach. Results on five diverse and challenging adaptation scenarios show that our approach outperforms existing state-of-the-art methods with noticeable margins.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier EISSN: 2331-8422
ispartof arXiv.org, 2023-11
issn 2331-8422
language eng
recordid cdi_proquest_journals_2887705791
source Free E- Journals
subjects Ablation
Adaptation
Alignment
Deep learning
Labels
Object recognition
Predictions
Tiling
Uncertainty
title Domain Adaptive Object Detection via Balancing Between Self-Training and Adversarial Learning
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-28T17%3A20%3A42IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Domain%20Adaptive%20Object%20Detection%20via%20Balancing%20Between%20Self-Training%20and%20Adversarial%20Learning&rft.jtitle=arXiv.org&rft.au=Muhammad%20Akhtar%20Munir&rft.date=2023-11-08&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2887705791%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2887705791&rft_id=info:pmid/&rfr_iscdi=true