{"id":"https://openalex.org/W7160931161","doi":"https://doi.org/10.48550/arxiv.2605.08985","title":"LLaVA-UHD v4: What Makes Efficient Visual Encoding in MLLMs?","display_name":"LLaVA-UHD v4: What Makes Efficient Visual Encoding in MLLMs?","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7160931161","doi":"https://doi.org/10.48550/arxiv.2605.08985"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08985","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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.2605.08985","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5111082343","display_name":"Kechen Fang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fang, Kechen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135931204","display_name":"Yihua Qin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Yihua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135917502","display_name":"Chongyi Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Chongyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134931733","display_name":"Wenshuo Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Wenshuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135980300","display_name":"Tianyu Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Tianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135955564","display_name":"Yuan Yao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Yuan","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":0.3946000039577484,"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":0.3946000039577484,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.1987999975681305,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.05700000002980232,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.8517000079154968},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5846999883651733},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.5041999816894531},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4986000061035156},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.4120999872684479},{"id":"https://openalex.org/keywords/flops","display_name":"FLOPS","score":0.39100000262260437},{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.3813000023365021}],"concepts":[{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.8517000079154968},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.771399974822998},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5846999883651733},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.5041999816894531},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4986000061035156},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4120999872684479},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.39100000262260437},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.3813000023365021},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3662000000476837},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.3610000014305115},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.3598000109195709},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.3467000126838684},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3440999984741211},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.3310000002384186},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.30869999527931213},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.2948000133037567},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.29019999504089355},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2590999901294708},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.2524999976158142},{"id":"https://openalex.org/C2780186347","wikidata":"https://www.wikidata.org/wiki/Q11414","display_name":"Subnetwork","level":2,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08985","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.08985","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08985","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Visual":[0],"encoding":[1,24,30,64,76,79,130,141],"constitutes":[2],"a":[3,148,189],"major":[4],"computational":[5],"bottleneck":[6],"in":[7,110],"Multimodal":[8],"Large":[9],"Language":[10],"Models":[11],"(MLLMs),":[12],"especially":[13],"for":[14,98,144,193],"high-resolution":[15,145,195],"image":[16],"inputs.":[17,146],"The":[18],"prevailing":[19],"practice":[20],"typically":[21],"adopts":[22],"global":[23,78,96],"followed":[25],"by":[26,165],"post-ViT":[27,36],"compression.":[28,69],"Global":[29],"produces":[31],"massive":[32],"token":[33,49,68],"sequences,":[34],"while":[35,119,167],"compression":[37,126],"incurs":[38],"the":[39,45,63,128],"full":[40],"quadratic":[41],"attention":[42,97],"cost":[43],"of":[44,151],"ViT":[46,112],"before":[47],"any":[48],"reduction":[50],"takes":[51],"place.":[52],"In":[53],"this":[54,58],"work,":[55],"we":[56,102,132],"revisit":[57],"convention":[59],"along":[60],"two":[61],"dimensions:":[62],"strategy":[65],"and":[66,114,138,157,200],"visual":[67,140],"First,":[70],"controlled":[71],"experiments":[72],"show":[73],"that":[74,83,177],"slice-based":[75,129],"outperforms":[77],"across":[80],"benchmarks,":[81],"suggesting":[82],"preserving":[84,120],"local":[85],"details":[86],"through":[87],"sliced":[88],"views":[89],"can":[90,180],"be":[91,181,203],"more":[92],"beneficial":[93],"than":[94],"applying":[95],"fine-grained":[99],"perception.":[100],"Second,":[101],"introduce":[103],"intra-ViT":[104,125],"early":[105],"compression,":[106],"which":[107],"reduces":[108,162],"tokens":[109],"shallow":[111],"layers":[113],"substantially":[115,182],"lowers":[116],"visual-encoding":[117,163,178],"FLOPs":[118,164],"downstream":[121,186],"performance.":[122,173],"By":[123],"integrating":[124],"into":[127],"framework,":[131],"present":[133],"LLaVA-UHD":[134,160],"v4,":[135],"an":[136],"efficient":[137,194],"compute-controllable":[139],"scheme":[142],"tailored":[143],"Across":[147],"diverse":[149],"set":[150],"benchmarks":[152],"covering":[153],"document":[154],"understanding,":[155],"OCR,":[156],"general":[158],"VQA,":[159],"v4":[161],"55.8%":[166],"matching":[168],"or":[169],"even":[170],"surpassing":[171],"baseline":[172],"These":[174],"results":[175],"suggest":[176],"efficiency":[179],"improved":[183],"without":[184],"sacrificing":[185],"performance,":[187],"providing":[188],"practical":[190],"design":[191],"direction":[192],"MLLMs.":[196],"All":[197],"model":[198],"weights":[199],"code":[201],"will":[202],"publicly":[204],"released":[205],"to":[206],"support":[207],"further":[208],"research.":[209]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-13T00:00:00"}