{"id":"https://openalex.org/W7162643627","doi":"https://doi.org/10.48550/arxiv.2605.27960","title":"Mags-RL: Wearing Multimodal LLMs a Magnifying Glass via Agentic Reinforcement Learning For Complex Scene Reasoning","display_name":"Mags-RL: Wearing Multimodal LLMs a Magnifying Glass via Agentic Reinforcement Learning For Complex Scene Reasoning","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162643627","doi":"https://doi.org/10.48550/arxiv.2605.27960"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.27960","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27960","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.2605.27960","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137202192","display_name":"Xuanzhao Dong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Xuanzhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137271295","display_name":"Wenhui Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Wenhui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058590658","display_name":"Peijie Qiu","orcid":"https://orcid.org/0000-0002-1591-5436"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiu, Peijie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137300205","display_name":"Xiwen Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Xiwen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137241880","display_name":"Xiaobing Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Xiaobing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137208758","display_name":"Xin Li","orcid":"https://orcid.org/0000-0003-2067-2763"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Xin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137270122","display_name":"Zhipeng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zhipeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137219723","display_name":"Shao Tang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Shao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137229421","display_name":"Gen Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Gen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137283486","display_name":"Yujian Xiong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiong, Yujian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137257263","display_name":"Hao Wang","orcid":"https://orcid.org/0009-0009-6316-3572"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137255169","display_name":"Yanxi Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yanxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137247738","display_name":"Prayag Tiwari","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tiwari, Prayag","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137253075","display_name":"Yalin Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yalin","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":0.3483999967575073,"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":0.3483999967575073,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.20260000228881836,"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":0.1776999980211258,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.6273999810218811},{"id":"https://openalex.org/keywords/visual-reasoning","display_name":"Visual reasoning","score":0.6190000176429749},{"id":"https://openalex.org/keywords/popularity","display_name":"Popularity","score":0.5238999724388123},{"id":"https://openalex.org/keywords/object","display_name":"Object (grammar)","score":0.5012999773025513},{"id":"https://openalex.org/keywords/bounding-overwatch","display_name":"Bounding overwatch","score":0.4909999966621399},{"id":"https://openalex.org/keywords/active-learning","display_name":"Active learning (machine learning)","score":0.36090001463890076},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.3483000099658966}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6273999810218811},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.6190000176429749},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5726000070571899},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.565500020980835},{"id":"https://openalex.org/C2780586970","wikidata":"https://www.wikidata.org/wiki/Q1357284","display_name":"Popularity","level":2,"score":0.5238999724388123},{"id":"https://openalex.org/C2781238097","wikidata":"https://www.wikidata.org/wiki/Q175026","display_name":"Object (grammar)","level":2,"score":0.5012999773025513},{"id":"https://openalex.org/C63584917","wikidata":"https://www.wikidata.org/wiki/Q333286","display_name":"Bounding overwatch","level":2,"score":0.4909999966621399},{"id":"https://openalex.org/C77967617","wikidata":"https://www.wikidata.org/wiki/Q4677561","display_name":"Active learning (machine learning)","level":2,"score":0.36090001463890076},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3483000099658966},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.3391999900341034},{"id":"https://openalex.org/C2779321571","wikidata":"https://www.wikidata.org/wiki/Q7936605","display_name":"Visual learning","level":2,"score":0.3248000144958496},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3127000033855438},{"id":"https://openalex.org/C178253425","wikidata":"https://www.wikidata.org/wiki/Q162668","display_name":"Visual perception","level":3,"score":0.31139999628067017},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3066999912261963},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.30550000071525574},{"id":"https://openalex.org/C83248878","wikidata":"https://www.wikidata.org/wiki/Q344000","display_name":"Active appearance model","level":3,"score":0.30480000376701355},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.2842999994754791},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.28369998931884766},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.27649998664855957}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.27960","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27960","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.2605.27960","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27960","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":[{"score":0.5327211022377014,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Despite":[0],"their":[1,18],"popularity":[2],"and":[3,28,105,127,133,172,192],"success,":[4],"Multimodal":[5],"Large":[6],"Language":[7],"Models":[8],"(MLLMs)":[9],"often":[10,56],"struggle":[11],"to":[12,125,138,164],"interpret":[13],"images":[14],"accurately,":[15],"which":[16],"limits":[17],"reasoning":[19,137,186],"capability":[20],"in":[21,96,116],"complex":[22,29],"scenarios":[23],"(e.g.,":[24],"high":[25],"object":[26],"density":[27],"background":[30],"clutter).":[31],"Prior":[32],"work":[33],"mainly":[34],"addresses":[35],"this":[36],"limitation":[37],"by":[38],"incorporating":[39],"explicit":[40],"visual":[41,189],"cues":[42],"like":[43],"bounding":[44],"boxes":[45],"that":[46,60,76,151],"require":[47,62],"extra":[48],"annotations.":[49],"In":[50],"addition,":[51],"the":[52,91,97,117,140],"resulting":[53],"low-resolution":[54],"crops":[55],"miss":[57],"fine-grained":[58,88],"details":[59],"MLLMs":[61,78],"for":[63,86],"accurate":[64],"reasoning.":[65],"Therefore,":[66],"we":[67],"propose":[68],"Mags-RL,":[69],"an":[70,80,102],"Agentic":[71],"Reinforcement":[72],"Learning":[73],"(RL)":[74],"framework":[75],"equips":[77],"with":[79,187],"external":[81],"super-resolution":[82,123],"\"magnifying":[83],"glass\"":[84],"agent":[85,124],"high-resolution":[87],"inspection.":[89],"Specifically,":[90],"model":[92],"performs":[93],"two-round":[94],"reasoning:":[95],"first":[98],"round,":[99,119],"it":[100,120],"generates":[101],"initial":[103],"rationale":[104],"autonomously":[106],"identifies":[107],"regions":[108],"of":[109],"interest":[110],"without":[111],"relying":[112],"on":[113,169],"additional":[114],"annotations;":[115],"second":[118],"invokes":[121],"a":[122,146],"crop":[126],"upscale":[128],"those":[129],"regions,":[130],"then":[131],"revisits":[132],"verifies":[134],"its":[135,176],"earlier":[136],"produce":[139],"final":[141],"answer.":[142],"We":[143],"also":[144],"introduce":[145],"novel":[147],"curriculum":[148],"learning":[149],"strategy":[150],"enables":[152],"data-efficient":[153],"RL":[154],"training,":[155],"needing":[156],"as":[157,159],"few":[158],"only":[160],"40":[161],"training":[162],"samples":[163],"achieve":[165],"reasonable":[166],"performance.":[167],"Experiments":[168],"VSR,":[170],"TallyQA,":[171],"GQA":[173],"subsets":[174],"show":[175],"superior":[177],"performance":[178],"against":[179],"recent":[180],"strong":[181],"competing":[182],"methods,":[183],"demonstrating":[184],"high-quality":[185],"precise":[188],"grounding.":[190],"Code":[191],"weights":[193],"will":[194],"be":[195],"released":[196],"soon.":[197]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-29T00:00:00"}