Mitigating Low-Level Visual Hallucinations Requires Self-Awareness: Database, Model and Training Strategy

The rapid development of multimodal large language models has resulted in remarkable advancements in visual perception and understanding, consolidating several tasks into a single visual question-answering framework. However, these models are prone to hallucinations, which limit their reliability as...

Full description

Saved in:

Bibliographic Details
Published in:	IEEE transactions on circuits and systems for video technology p. 1
Main Authors:	Sun, Yinan, Min, Xiongkuo, Zhang, Zicheng, Gao, Yixuan, Cao, Yuqin, Zhai, Guangtao
Format:	Journal Article
Language:	English
Published:	IEEE 2025
Subjects:	Accuracy Analytical models Feature extraction hallucination Image quality image quality assessment Large language models low-level vision Multimodal large language models Reliability Training Visual databases Visual perception Visualization
ISSN:	1051-8215, 1558-2205
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Abstract	The rapid development of multimodal large language models has resulted in remarkable advancements in visual perception and understanding, consolidating several tasks into a single visual question-answering framework. However, these models are prone to hallucinations, which limit their reliability as artificial intelligence systems. While this issue is extensively researched in natural language processing and image captioning, there remains a lack of investigation of hallucinations in Low-level Visual Perception and Understanding (HLPU), especially in the context of image quality assessment tasks. We consider that these hallucinations arise from an absence of clear self-awareness within the models. To address this issue, we first introduce the HLPU instruction database, the first instruction database specifically focused on hallucinations in low-level vision tasks. This database contains approximately 200K question-answer pairs and comprises four subsets, each covering different types of instructions. Subsequently, we propose the Self-Awareness Failure Elimination (SAFEQA) model, which utilizes image features, salient region features and quality features to improve the perception and comprehension abilities of the model in low-level vision tasks. Furthermore, we propose the Enhancing Self-Awareness Preference Optimization (ESA-PO) framework to increase the model's awareness of knowledge boundaries, thereby mitigating the incidence of hallucination. Finally, we conduct comprehensive experiments on low-level vision tasks, with the results demonstrating that our proposed method significantly enhances self-awareness of the model in these tasks and reduces hallucinations. Notably, our proposed method improves both accuracy and self-awareness of the proposed model and outperforms close-source models in terms of various evaluation metrics. This research contributes to the advancement of self-awareness capabilities in multimodal large language models, particularly for low-level visual perception and understanding tasks.
AbstractList	The rapid development of multimodal large language models has resulted in remarkable advancements in visual perception and understanding, consolidating several tasks into a single visual question-answering framework. However, these models are prone to hallucinations, which limit their reliability as artificial intelligence systems. While this issue is extensively researched in natural language processing and image captioning, there remains a lack of investigation of hallucinations in Low-level Visual Perception and Understanding (HLPU), especially in the context of image quality assessment tasks. We consider that these hallucinations arise from an absence of clear self-awareness within the models. To address this issue, we first introduce the HLPU instruction database, the first instruction database specifically focused on hallucinations in low-level vision tasks. This database contains approximately 200K question-answer pairs and comprises four subsets, each covering different types of instructions. Subsequently, we propose the Self-Awareness Failure Elimination (SAFEQA) model, which utilizes image features, salient region features and quality features to improve the perception and comprehension abilities of the model in low-level vision tasks. Furthermore, we propose the Enhancing Self-Awareness Preference Optimization (ESA-PO) framework to increase the model's awareness of knowledge boundaries, thereby mitigating the incidence of hallucination. Finally, we conduct comprehensive experiments on low-level vision tasks, with the results demonstrating that our proposed method significantly enhances self-awareness of the model in these tasks and reduces hallucinations. Notably, our proposed method improves both accuracy and self-awareness of the proposed model and outperforms close-source models in terms of various evaluation metrics. This research contributes to the advancement of self-awareness capabilities in multimodal large language models, particularly for low-level visual perception and understanding tasks.
Author	Min, Xiongkuo Zhang, Zicheng Gao, Yixuan Cao, Yuqin Sun, Yinan Zhai, Guangtao
Author_xml	– sequence: 1 givenname: Yinan orcidid: 0009-0004-3054-4268 surname: Sun fullname: Sun, Yinan – sequence: 2 givenname: Xiongkuo orcidid: 0000-0001-5693-0416 surname: Min fullname: Min, Xiongkuo – sequence: 3 givenname: Zicheng orcidid: 0000-0002-7247-7938 surname: Zhang fullname: Zhang, Zicheng – sequence: 4 givenname: Yixuan orcidid: 0000-0002-6292-0529 surname: Gao fullname: Gao, Yixuan – sequence: 5 givenname: Yuqin surname: Cao fullname: Cao, Yuqin – sequence: 6 givenname: Guangtao orcidid: 0000-0001-8165-9322 surname: Zhai fullname: Zhai, Guangtao
BookMark	eNpFkF1LwzAYhYNMcJv-AfEiP8DOvEmTdt6N-TGhQ3B1t-Vtk45ITTXpHPv3dm7g1TlweM7FMyID1zpDyDWwCQCb3uXz1TqfcMblRCiYSpmekSH0EXHO5KDvTEKUcpAXZBTCB2MQp3EyJHZpO7vBzroNzdpdlJkf09C1DVts6AKbZltZ18-tC_TNfG-tN4GuTFNHsx1640wI9_QBOywxmFu6bHWPo9M092jd4XXVeezMZn9Jzmtsgrk65Zi8Pz3m80WUvT6_zGdZVIFQXSRrBF0qzZIqQdQx04mSlWJTUaVCoZKJqGPQMUcDNRPAy5KBFoiSMVYqFGPCj7-Vb0Pwpi6-vP1Evy-AFQdZxZ-s4iCrOMnqoZsjZI0x_wDANJFCil8S9mlk
CODEN	ITCTEM
ContentType	Journal Article
DBID	97E RIA RIE AAYXX CITATION
DOI	10.1109/TCSVT.2025.3619558
DatabaseName	IEEE All-Society Periodicals Package (ASPP) 2005–Present IEEE All-Society Periodicals Package (ASPP) 1998–Present IEEE/IET Electronic Library (IEL) CrossRef
DatabaseTitle	CrossRef
DatabaseTitleList
Database_xml	– sequence: 1 dbid: RIE name: IEEE Electronic Library (IEL) url: https://ieeexplore.ieee.org/ sourceTypes: Publisher
DeliveryMethod	fulltext_linktorsrc
Discipline	Engineering
EISSN	1558-2205
EndPage	1
ExternalDocumentID	10_1109_TCSVT_2025_3619558 11197535
Genre	orig-research
GroupedDBID	-~X 0R~ 29I 4.4 5GY 6IK 97E AAJGR AASAJ AAWTH ABAZT ABQJQ ABVLG ACGFO ACGFS ACIWK AENEX AGQYO AHBIQ AKJIK AKQYR ALMA_UNASSIGNED_HOLDINGS ASUFR ATWAV BEFXN BFFAM BGNUA BKEBE BPEOZ CS3 DU5 EBS HZ~ IFIPE IPLJI JAVBF LAI O9- OCL P2P RIA RIE RNS RXW TAE TN5 5VS AAYXX AETIX AGSQL AI. AIBXA ALLEH CITATION EJD H~9 ICLAB IFJZH M43 VH1
ID	FETCH-LOGICAL-c136t-5fa1db6d07c7aad40d765c6093c836a6573f41d42ae1f0312bb01d3aa5000b6a3
IEDL.DBID	RIE
ISSN	1051-8215
IngestDate	Sat Nov 29 07:11:06 EST 2025 Wed Oct 15 14:20:46 EDT 2025
IsPeerReviewed	true
IsScholarly	true
Language	English
License	https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html https://doi.org/10.15223/policy-029 https://doi.org/10.15223/policy-037
LinkModel	DirectLink
MergedId	FETCHMERGED-LOGICAL-c136t-5fa1db6d07c7aad40d765c6093c836a6573f41d42ae1f0312bb01d3aa5000b6a3
ORCID	0000-0001-5693-0416 0009-0004-3054-4268 0000-0001-8165-9322 0000-0002-7247-7938 0000-0002-6292-0529
PageCount	1
ParticipantIDs	crossref_primary_10_1109_TCSVT_2025_3619558 ieee_primary_11197535
PublicationCentury	2000
PublicationDate	2025-00-00
PublicationDateYYYYMMDD	2025-01-01
PublicationDate_xml	– year: 2025 text: 2025-00-00
PublicationDecade	2020
PublicationTitle	IEEE transactions on circuits and systems for video technology
PublicationTitleAbbrev	TCSVT
PublicationYear	2025
Publisher	IEEE
Publisher_xml	– name: IEEE
SSID	ssj0014847
Score	2.4584467
Snippet	The rapid development of multimodal large language models has resulted in remarkable advancements in visual perception and understanding, consolidating several...
SourceID	crossref ieee
SourceType	Index Database Publisher
StartPage	1
SubjectTerms	Accuracy Analytical models Feature extraction hallucination Image quality image quality assessment Large language models low-level vision Multimodal large language models Reliability Training Visual databases Visual perception Visualization
Title	Mitigating Low-Level Visual Hallucinations Requires Self-Awareness: Database, Model and Training Strategy
URI	https://ieeexplore.ieee.org/document/11197535
hasFullText	1
inHoldings	1
isFullTextHit
isPrint
journalDatabaseRights	– providerCode: PRVIEE databaseName: IEEE Electronic Library (IEL) customDbUrl: eissn: 1558-2205 dateEnd: 99991231 omitProxy: false ssIdentifier: ssj0014847 issn: 1051-8215 databaseCode: RIE dateStart: 19910101 isFulltext: true titleUrlDefault: https://ieeexplore.ieee.org/ providerName: IEEE
link	http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV05T8MwFLZoxQADZxHlkgc2SJvEcZywVYWqQ6kQDVW3yFdQpChFTULFv8d2UujCwGZFthS9Z_v5Xd8HwG0oNC6WRJbncGx5SsXqHmQ6XUiD0LV5gIkwZBNkOg0Wi_ClaVY3vTBSSlN8Jnt6aHL5YskrHSrrOzrnhRFugRYhft2s9ZMy8ALDJqbeC44VKEO26ZCxw340nM0j5Qu6uIeUw4A1v_uWFdqiVTFWZXT4z_85AgfN8xEOan0fgx2Zn4D9LVDBU5A-pzVuRv4OJ8u1NdFlQXCeFpVaOKZZVvG0DgEW8FXqQmBZwJnMEmuw1p1h6up7gI-0pNrA3UNNlpZBmgsYNWwSsEG0_eqAt9FTNBxbDaGCxR3klxZOqCOYL2zCCaXCswXxMfftEPEA-dTHBCWeIzyXSidRp91lzHYEooY1gfkUnYF2vszlOYAahl8qb0RTmHkBDVnoKWc3YUy4iTrkrAvuNgKOP2rcjNj4G3YYG3XEWh1xo44u6Gjp_s5sBHvxx_dLsKeX16GQK9AuV5W8Brv8s0yL1Y3ZF99M6rVG
linkProvider	IEEE
linkToHtml	http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3LT8IwGG8UTdSDT4z47MGbTra13cMbQQlGIEYm4bb0NbOEDMOYxP_ethvKxYO3pdmW5vvafv1evx8A16HQuFgSWdjhxMJKxeocZDpdSIPQtXlAfGHIJvzBIBiPw5eqWd30wkgpTfGZvNOPJpcvprzQobKmo3NeBJF1sEEwdu2yXesnaYADwyembgyOFShTtuyRscNm1B6OIuUNuuQOKZeBaIb3FTu0Qqxi7Epn758z2ge71QUStkqNH4A1mR2CnRVYwSOQ9tMSOSN7h73pwurpwiA4SvNCfdilk0nB0zIImMNXqUuBZQ6HcpJYrYXuDVOH3z18oHOqTdwt1HRpE0gzAaOKTwJWmLZfdfDWeYzaXauiVLC4g7y5RRLqCOYJ2-c-pQLbwvcI9-wQ8QB51CM-SrAjsEulk6j97jJmOwJRw5vAPIqOQS2bZvIEQA3EL5U_oknMcEBDFmLl7iaMCTdR25w1wM1SwPFHiZwRG4_DDmOjjlirI67U0QB1Ld3fNyvBnv4xfgW2ulG_F_eeBs9nYFv_qgyMnIPafFbIC7DJP-dpPrs0a-Qb6Eu4jQ
openUrl	ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Mitigating+Low-Level+Visual+Hallucinations+Requires+Self-Awareness%3A+Database%2C+Model+and+Training+Strategy&rft.jtitle=IEEE+transactions+on+circuits+and+systems+for+video+technology&rft.au=Sun%2C+Yinan&rft.au=Min%2C+Xiongkuo&rft.au=Zhang%2C+Zicheng&rft.au=Gao%2C+Yixuan&rft.date=2025&rft.issn=1051-8215&rft.eissn=1558-2205&rft.spage=1&rft.epage=1&rft_id=info:doi/10.1109%2FTCSVT.2025.3619558&rft.externalDBID=n%2Fa&rft.externalDocID=10_1109_TCSVT_2025_3619558
thumbnail_l	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1051-8215&client=summon
thumbnail_m	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1051-8215&client=summon
thumbnail_s	http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1051-8215&client=summon