{"id":"https://openalex.org/W6929265781","doi":"https://doi.org/10.48550/arxiv.2508.04567","title":"Analyzing and Mitigating Object Hallucination: A Training Bias Perspective","display_name":"Analyzing and Mitigating Object Hallucination: A Training Bias Perspective","publication_year":2025,"publication_date":"2025-08-06","ids":{"openalex":"https://openalex.org/W6929265781","doi":"https://doi.org/10.48550/arxiv.2508.04567"},"language":"en","primary_location":{"id":"doi:10.48550/arxiv.2508.04567","doi":"https://doi.org/10.48550/arxiv.2508.04567","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2508.04567","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2508.04567","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Li, Yifan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yifan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zhou, Kun","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Kun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zhao, Wayne Xin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Wayne Xin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Fang, Lei","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fang, Lei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125635347","display_name":"Jirong Wen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Ji-Rong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":true,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/object-hallucination","display_name":"object hallucination","score":0.9919999837875366},{"id":"https://openalex.org/keywords/training-bias","display_name":"training bias","score":0.8880000114440918},{"id":"https://openalex.org/keywords/large-vision-language-models","display_name":"large vision-language models","score":0.9769999980926514},{"id":"https://openalex.org/keywords/unlearning","display_name":"unlearning","score":0.7120000123977661},{"id":"https://openalex.org/keywords/fine-tuning","display_name":"fine-tuning","score":0.6610000133514404},{"id":"https://openalex.org/keywords/hallucination-mitigation","display_name":"hallucination mitigation","score":0.7369999885559082}],"concepts":[{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.7081999778747559},{"id":"https://openalex.org/C2911011789","wikidata":"https://www.wikidata.org/wiki/Q130741","display_name":"Hallucinating","level":2,"score":0.7052000164985657},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6934000253677368},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6450999975204468},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.6273999810218811},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.6266999840736389},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.5914999842643738},{"id":"https://openalex.org/C2780440489","wikidata":"https://www.wikidata.org/wiki/Q5227278","display_name":"Data-driven","level":2,"score":0.46720001101493835},{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.4287000000476837},{"id":"https://openalex.org/C2781238097","wikidata":"https://www.wikidata.org/wiki/Q175026","display_name":"Object (grammar)","level":2,"score":0.4104999899864197},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.4092000126838684},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.40560001134872437},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.40299999713897705},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.3869999945163727},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.3767000138759613},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.3677000105381012},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.32679998874664307},{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.310699999332428},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.28200000524520874},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2809000015258789},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.27950000762939453},{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.2757999897003174},{"id":"https://openalex.org/C94124525","wikidata":"https://www.wikidata.org/wiki/Q912550","display_name":"Categorization","level":2,"score":0.2709999978542328},{"id":"https://openalex.org/C177212765","wikidata":"https://www.wikidata.org/wiki/Q627335","display_name":"Workflow","level":2,"score":0.2678000032901764},{"id":"https://openalex.org/C108154423","wikidata":"https://www.wikidata.org/wiki/Q1469792","display_name":"Salience (neuroscience)","level":2,"score":0.26750001311302185},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.26350000500679016},{"id":"https://openalex.org/C64876066","wikidata":"https://www.wikidata.org/wiki/Q5141226","display_name":"Cognitive neuroscience of visual object recognition","level":3,"score":0.26190000772476196},{"id":"https://openalex.org/C2780922921","wikidata":"https://www.wikidata.org/wiki/Q255189","display_name":"Paraphrase","level":2,"score":0.25540000200271606},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2508.04567","doi":"https://doi.org/10.48550/arxiv.2508.04567","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2508.04567","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2508.04567","doi":"https://doi.org/10.48550/arxiv.2508.04567","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2508.04567","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10","score":0.6917446255683899}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"scaling":[1],"up":[2],"training":[3,43,61,81,89,132,162,177,210,244],"data":[4,44,62,90,178,211,245,260],"has":[5],"significantly":[6,220],"improved":[7],"the":[8,21,30,40,60,125,138,167,176,195,201,209,217],"general":[9],"multimodal":[10],"capabilities":[11],"of":[12,42,55,63,203,216],"Large":[13],"Vision-Language":[14],"Models":[15],"(LVLMs),":[16],"they":[17,83,101],"still":[18],"suffer":[19,79],"from":[20,59,80],"hallucination":[22,160,222,252],"issue,":[23,119],"generating":[24],"text":[25],"that":[26,76,130,192],"is":[27,134],"inconsistent":[28],"with":[29,65,234],"visual":[31],"input.":[32],"This":[33],"phenomenon":[34],"motivates":[35],"us":[36],"to":[37,85,111,157,236,241,251],"systematically":[38],"investigate":[39],"role":[41],"in":[45,137],"hallucination.":[46,256],"We":[47],"introduce":[48],"a":[49,180,186],"new":[50],"benchmark,":[51],"POPEv2,":[52,73],"which":[53],"consists":[54],"counterfactual":[56,105],"images":[57,96],"collected":[58],"LVLMs":[64,78],"certain":[66],"objects":[67],"masked.":[68],"Through":[69],"comprehensive":[70],"evaluation":[71],"on":[72,95,104,124,144,175],"we":[74,120,147],"find":[75],"current":[77],"bias:":[82],"fail":[84],"fully":[86],"leverage":[87],"their":[88],"and":[91,152,172,184,188,212,226,243,247,259],"hallucinate":[92],"more":[93],"frequently":[94],"seen":[97],"during":[98],"training.":[99],"Specifically,":[100],"perform":[102],"poorly":[103],"images,":[106],"often":[107],"incorrectly":[108],"answering":[109],"``Yes''":[110],"questions":[112],"about":[113],"masked":[114],"objects.":[115],"To":[116],"understand":[117],"this":[118,131],"conduct":[121],"probing":[122],"experiments":[123,199],"models'":[126],"internal":[127],"components,":[128],"revealing":[129],"bias":[133,163,183],"primarily":[135],"located":[136],"language":[139],"modeling":[140],"(LM)":[141],"head.":[142,197],"Based":[143],"these":[145],"findings,":[146],"propose":[148],"Obliviate,":[149],"an":[150],"efficient":[151],"lightweight":[153],"unlearning":[154],"method":[155],"designed":[156],"mitigate":[158],"object":[159],"via":[161],"unlearning.":[164],"Obliviate":[165,219],"identifies":[166],"discrepancy":[168],"between":[169],"ground-truth":[170],"labels":[171],"model":[173,238],"outputs":[174],"as":[179],"proxy":[181],"for":[182],"adopts":[185],"parameter-":[187],"data-efficient":[189],"fine-tuning":[190],"strategy":[191],"only":[193,207],"updates":[194],"LM":[196],"Extensive":[198],"demonstrate":[200],"effectiveness":[202],"our":[204],"approach.":[205],"While":[206],"reusing":[208],"updating":[213],"approximately":[214],"2\\%":[215],"parameters,":[218],"reduces":[221],"across":[223],"both":[224,237],"discriminative":[225],"generative":[227],"tasks.":[228],"Furthermore,":[229],"it":[230],"demonstrates":[231],"strong":[232],"scalability":[233],"respect":[235],"size":[239],"(2B":[240],"72B)":[242],"volume,":[246],"exhibits":[248],"promising":[249],"generalization":[250],"types":[253],"beyond":[254],"object-level":[255],"Our":[257],"code":[258],"will":[261],"be":[262],"publicly":[263],"released.":[264]},"counts_by_year":[],"updated_date":"2026-10-13T10:08:12.033832","created_date":"2025-10-10T00:00:00"}