{"id":"https://openalex.org/W7164828831","doi":"https://doi.org/10.48550/arxiv.2606.14703","title":"Gaze Heads: How VLMs Look at What They Describe","display_name":"Gaze Heads: How VLMs Look at What They Describe","publication_year":2026,"publication_date":"2026-06-12","ids":{"openalex":"https://openalex.org/W7164828831","doi":"https://doi.org/10.48550/arxiv.2606.14703"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.14703","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.14703","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.14703","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138648319","display_name":"Rohit Gandikota","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gandikota, Rohit","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138628398","display_name":"David Bau","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bau, David","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":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.8647000193595886,"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":0.8647000193595886,"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/T11094","display_name":"Face Recognition and Perception","score":0.010900000110268593,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.008100000210106373,"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/gaze","display_name":"Gaze","score":0.8550000190734863},{"id":"https://openalex.org/keywords/comic-strip","display_name":"Comic strip","score":0.6155999898910522},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5918999910354614},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.5295000076293945},{"id":"https://openalex.org/keywords/comics","display_name":"Comics","score":0.4408999979496002},{"id":"https://openalex.org/keywords/mechanism","display_name":"Mechanism (biology)","score":0.4291999936103821},{"id":"https://openalex.org/keywords/lever","display_name":"Lever","score":0.42820000648498535},{"id":"https://openalex.org/keywords/image","display_name":"Image (mathematics)","score":0.4131999909877777}],"concepts":[{"id":"https://openalex.org/C2779916870","wikidata":"https://www.wikidata.org/wiki/Q14467155","display_name":"Gaze","level":2,"score":0.8550000190734863},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6520000100135803},{"id":"https://openalex.org/C3020586399","wikidata":"https://www.wikidata.org/wiki/Q838795","display_name":"Comic strip","level":3,"score":0.6155999898910522},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5918999910354614},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5295000076293945},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5040000081062317},{"id":"https://openalex.org/C529099274","wikidata":"https://www.wikidata.org/wiki/Q1004","display_name":"Comics","level":2,"score":0.4408999979496002},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.4291999936103821},{"id":"https://openalex.org/C107524782","wikidata":"https://www.wikidata.org/wiki/Q40164","display_name":"Lever","level":2,"score":0.42820000648498535},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.4131999909877777},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.4074999988079071},{"id":"https://openalex.org/C56461940","wikidata":"https://www.wikidata.org/wiki/Q970687","display_name":"Eye tracking","level":2,"score":0.38359999656677246},{"id":"https://openalex.org/C2776608160","wikidata":"https://www.wikidata.org/wiki/Q4785462","display_name":"Natural (archaeology)","level":2,"score":0.32910001277923584},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.32510000467300415},{"id":"https://openalex.org/C31395832","wikidata":"https://www.wikidata.org/wiki/Q1318674","display_name":"Testbed","level":2,"score":0.320499986410141},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.30880001187324524},{"id":"https://openalex.org/C2776538412","wikidata":"https://www.wikidata.org/wiki/Q989963","display_name":"Storytelling","level":3,"score":0.28029999136924744},{"id":"https://openalex.org/C199033989","wikidata":"https://www.wikidata.org/wiki/Q1318295","display_name":"Narrative","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C100609095","wikidata":"https://www.wikidata.org/wiki/Q1335050","display_name":"Embodied cognition","level":2,"score":0.2700999975204468},{"id":"https://openalex.org/C2775936607","wikidata":"https://www.wikidata.org/wiki/Q466845","display_name":"Tracking (education)","level":2,"score":0.26910001039505005},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.2653000056743622},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C46312422","wikidata":"https://www.wikidata.org/wiki/Q11024","display_name":"Communication","level":1,"score":0.260699987411499},{"id":"https://openalex.org/C65682993","wikidata":"https://www.wikidata.org/wiki/Q1056451","display_name":"Reflection (computer programming)","level":2,"score":0.2581999897956848},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.2572999894618988},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.2531999945640564},{"id":"https://openalex.org/C16345878","wikidata":"https://www.wikidata.org/wiki/Q107472979","display_name":"Orientation (vector space)","level":2,"score":0.251800000667572}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.14703","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.14703","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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.2606.14703","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.14703","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"How":[0],"a":[1,22,27,57,62,70,96,182],"vision-language":[2],"model":[3,20,49,168,204,245],"internally":[4],"solves":[5],"the":[6,19,45,48,87,100,112,123,135,144,162,167,178,187],"task":[7],"of":[8,30,119],"describing":[9],"an":[10],"image":[11,46,88],"is":[12,50,76],"far":[13],"from":[14,61,206],"obvious.":[15],"We":[16,53],"find":[17,54],"that":[18,104,229],"develops":[21],"specific":[23],"mechanism":[24,200],"for":[25,242],"this:":[26],"small":[28],"set":[29],"attention":[31,43,94],"heads":[32,82,140,150],"in":[33,195],"its":[34,171],"language-model":[35],"backbone,":[36],"which":[37],"we":[38],"call":[39],"gaze":[40,81,114,163],"heads,":[41,115,121],"whose":[42],"tracks":[44],"region":[47,98,105],"currently":[51],"describing.":[52],"them":[55],"with":[56],"simple":[58],"correlation":[59],"score":[60],"few":[63,183],"forward":[64],"passes,":[65],"using":[66],"comic":[67,129],"strips":[68],"as":[69,238],"controlled":[71],"testbed":[72],"where":[73],"narrative":[74],"order":[75],"laid":[77],"out":[78],"spatially.":[79],"These":[80],"do":[83],"not":[84],"just":[85],"track":[86],"tokens":[89],"being":[90],"described:":[91],"redirecting":[92],"their":[93],"to":[95,102,126,142,158,177,192,208],"chosen":[97,128,193],"forces":[99],"VLM":[101,214],"describe":[103],"instead.":[106],"A":[107],"single":[108],"attention-mask":[109],"intervention":[110,137,189],"on":[111,138,148],"top-100":[113],"fewer":[116],"than":[117],"9%":[118],"all":[120,149],"steers":[122],"model's":[124],"answer":[125],"any":[127,248],"panel":[130,173],"at":[131,258],"83.1%":[132],"accuracy,":[133],"while":[134],"same":[136,154,188],"random":[139],"fails":[141],"redirect":[143],"answer,":[145],"and":[146,175,211,254],"intervening":[147],"destroys":[151],"generation.":[152],"The":[153,199],"lever":[155],"also":[156],"extends":[157],"continuous":[159],"control:":[160],"switching":[161],"target":[164],"mid-generation":[165],"makes":[166],"wrap":[169],"up":[170],"current":[172],"description":[174],"move":[176],"new":[179],"one":[180],"within":[181],"tokens.":[184],"Beyond":[185],"comics,":[186],"redirects":[190],"answers":[191],"regions":[194],"natural":[196],"COCO":[197],"images.":[198],"further":[201],"recurs":[202],"across":[203,212],"sizes":[205],"2B":[207],"32B":[209],"parameters":[210],"other":[213],"architectures,":[215],"although":[216],"some":[217],"frozen-encoder":[218],"families":[219],"show":[220],"no":[221],"comparable":[222],"head":[223],"set.":[224],"More":[225],"broadly,":[226],"this":[227],"shows":[228],"targeted":[230],"edits":[231],"identified":[232],"through":[233],"mechanistic":[234],"analysis":[235],"can":[236],"serve":[237],"practical":[239],"inference-time":[240],"levers":[241],"steering":[243],"multimodal":[244],"behavior,":[246],"without":[247],"retraining.":[249],"Our":[250],"code,":[251],"interactive":[252],"demo,":[253],"datasets":[255],"are":[256],"available":[257],"https://gaze.baulab.info/":[259]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-16T00:00:00"}