{"id":"https://openalex.org/W7167635491","doi":"https://doi.org/10.48550/arxiv.2607.03050","title":"OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models","display_name":"OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models","publication_year":2026,"publication_date":"2026-07-03","ids":{"openalex":"https://openalex.org/W7167635491","doi":"https://doi.org/10.48550/arxiv.2607.03050"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.03050","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.03050","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,"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.2607.03050","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140160550","display_name":"Shijie Cao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cao, Shijie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140202908","display_name":"Qingyu Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Qingyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057321039","display_name":"Boxi Yu","orcid":"https://orcid.org/0000-0001-5213-7189"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Boxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132569403","display_name":"Yuzhong Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yuzhong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140212847","display_name":"Boxi Cao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cao, Boxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085698173","display_name":"Ye Lu","orcid":"https://orcid.org/0000-0002-2376-4519"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Yaojie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140214050","display_name":"Hongyu Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Hongyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140211768","display_name":"Xianpei Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Xianpei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5121042310","display_name":"L G Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Le","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/T10860","display_name":"Speech and Audio Processing","score":0.38089999556541443,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T10860","display_name":"Speech and Audio Processing","score":0.38089999556541443,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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":0.17509999871253967,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.1673000007867813,"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/security-token","display_name":"Security token","score":0.8646000027656555},{"id":"https://openalex.org/keywords/locality","display_name":"Locality","score":0.605400025844574},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5770999789237976},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5078999996185303},{"id":"https://openalex.org/keywords/salient","display_name":"Salient","score":0.4909000098705292},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.487199991941452},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.47099998593330383},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.4068000018596649}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8646000027656555},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8068000078201294},{"id":"https://openalex.org/C2779808786","wikidata":"https://www.wikidata.org/wiki/Q6664603","display_name":"Locality","level":2,"score":0.605400025844574},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5770999789237976},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5078999996185303},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.4909000098705292},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.487199991941452},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.47099998593330383},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.4068000018596649},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.39719998836517334},{"id":"https://openalex.org/C27602214","wikidata":"https://www.wikidata.org/wiki/Q1868547","display_name":"Locality of reference","level":3,"score":0.37549999356269836},{"id":"https://openalex.org/C2780226545","wikidata":"https://www.wikidata.org/wiki/Q6888030","display_name":"Modality (human\u2013computer interaction)","level":2,"score":0.36399999260902405},{"id":"https://openalex.org/C2779903281","wikidata":"https://www.wikidata.org/wiki/Q6888026","display_name":"Modalities","level":2,"score":0.3587999939918518},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.35260000824928284},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.3357999920845032},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3319999873638153},{"id":"https://openalex.org/C102462485","wikidata":"https://www.wikidata.org/wiki/Q904595","display_name":"Token bus network","level":3,"score":0.3075999915599823},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.30379998683929443},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.303600013256073},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.2890999913215637},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.28679999709129333},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.2825999855995178},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.27000001072883606},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2648000121116638},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.26190000772476196},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2549000084400177}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.03050","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.03050","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,"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.2607.03050","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.03050","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,"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":{"Omni":[0],"modal":[1],"large":[2,22],"language":[3],"models":[4],"(OmniLLMs)":[5],"have":[6],"attracted":[7],"wide":[8],"attention":[9],"for":[10,76,83],"their":[11],"ability":[12],"to":[13,29,163,168],"jointly":[14],"process":[15],"audio":[16],"and":[17,53,85,135],"video,":[18],"but":[19],"they":[20],"generate":[21],"token":[23,35,73,132,146,154],"sequences":[24],"under":[25],"audio-visual":[26,34,51,99,121],"inputs,":[27],"leading":[28],"substantial":[30],"inference":[31],"cost.":[32],"Existing":[33],"compression":[36,74,90],"methods":[37],"often":[38],"rely":[39],"on":[40,114,139],"unimodal":[41],"guidance,":[42],"overlooking":[43],"the":[44,57,103,115,169],"temporal":[45],"locality":[46],"of":[47],"query-relevant":[48],"evidence":[49,96],"in":[50],"inputs":[52],"implicitly":[54],"assuming":[55],"that":[56,78,92,107,124],"two":[58],"modalities":[59],"share":[60],"a":[61,70,88,173],"temporally":[62],"aligned":[63],"information":[64],"density":[65],"distribution.":[66],"We":[67],"propose":[68],"\\textbf{OmniFocus},":[69],"training-free":[71],"query-guided":[72],"method":[75],"OmniLLMs":[77],"performs":[79],"independent":[80],"importance":[81],"estimation":[82],"video":[84],"audio,":[86],"enabling":[87],"modality-symmetric":[89],"design":[91],"preserves":[93],"modality-specific":[94],"salient":[95],"while":[97,160],"maintaining":[98],"alignment,":[100],"thereby":[101],"mitigating":[102],"modality":[104],"bias":[105],"issue":[106],"can":[108],"arise":[109],"from":[110],"unimodal-guided":[111],"compression.":[112],"Experiments":[113],"Qwen2.5-Omni":[116],"model":[117],"family":[118],"across":[119],"four":[120],"benchmarks":[122],"show":[123],"OmniFocus":[125,156],"maintains":[126,157],"strong":[127],"compressed":[128],"performance":[129],"at":[130,144,152],"low":[131],"retention":[133],"ratios":[134],"outperforms":[136],"existing":[137],"baselines":[138],"several":[140],"major":[141],"benchmark":[142],"scores":[143],"25\\%":[145,153],"retention.":[147],"On":[148],"DailyOmni":[149],"with":[150],"Qwen2.5-Omni-7B":[151],"retention,":[155],"59.40":[158],"accuracy":[159],"delivering":[161],"up":[162],"1.38$\\times$":[164],"prefill":[165],"speedup":[166],"relative":[167],"full-token":[170],"baseline,":[171],"highlighting":[172],"favorable":[174],"practical":[175],"accuracy-efficiency":[176],"trade-off.":[177]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-08T00:00:00"}