{"id":"https://openalex.org/W4414449239","doi":"https://doi.org/10.48550/arxiv.2508.20691","title":"MobileCLIP2: Improving Multi-Modal Reinforced Training","display_name":"MobileCLIP2: Improving Multi-Modal Reinforced Training","publication_year":2025,"publication_date":"2025-08-28","ids":{"openalex":"https://openalex.org/W4414449239","doi":"https://doi.org/10.48550/arxiv.2508.20691"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2508.20691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2508.20691","pdf_url":"https://arxiv.org/pdf/2508.20691","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2508.20691","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5036601505","display_name":"Fartash Faghri","orcid":"https://orcid.org/0000-0001-5975-5158","observed_orcids":["https://orcid.org/0000-0001-5975-5158"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Faghri, Fartash","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013120724","display_name":"Pavan Kumar Anasosalu Vasu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vasu, Pavan Kumar Anasosalu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104313177","display_name":"Cem Koc","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Koc, Cem","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112327867","display_name":"Vaishaal Shankar","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shankar, Vaishaal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5064265174","display_name":"Alexander Toshev","orcid":"https://orcid.org/0000-0003-0925-638X","observed_orcids":["https://orcid.org/0000-0003-0925-638X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Toshev, Alexander","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028613002","display_name":"Oncel Tuzel","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tuzel, Oncel","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5059295598","display_name":"Hadi Pouransari","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pouransari, Hadi","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":true,"cited_by_count":11,"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":0.9262999892234802,"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":0.9262999892234802,"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":0.03759999945759773,"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":0.03400000184774399,"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/image-text-models","display_name":"image-text models","score":0.6639999747276306},{"id":"https://openalex.org/keywords/knowledge-distillation","display_name":"knowledge distillation","score":0.9070000052452087},{"id":"https://openalex.org/keywords/zero-shot-image-classification","display_name":"zero-shot image classification","score":0.7519999742507935},{"id":"https://openalex.org/keywords/imagenet-1k","display_name":"ImageNet-1K","score":0.7730000019073486},{"id":"https://openalex.org/keywords/thermal-tuning","display_name":"temperature tuning","score":0.8029999732971191}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7390999794006348},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6154000163078308},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.5370000004768372},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.5263000130653381},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4871000051498413},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4794999957084656},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.46540001034736633},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4431000053882599},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4068000018596649},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.33340001106262207},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.2815000116825104},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2685000002384186},{"id":"https://openalex.org/C3020594445","wikidata":"https://www.wikidata.org/wiki/Q1550460","display_name":"Initial training","level":2,"score":0.2612999975681305}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2508.20691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2508.20691","pdf_url":"https://arxiv.org/pdf/2508.20691","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2508.20691","doi":"https://doi.org/10.48550/arxiv.2508.20691","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2508.20691","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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2508.20691","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2508.20691","pdf_url":"https://arxiv.org/pdf/2508.20691","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414449239.pdf","grobid_xml":"https://content.openalex.org/works/W4414449239.grobid-xml"},"referenced_works_count":28,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W1889081078","https://openalex.org/W2950141105","https://openalex.org/W3094502228","https://openalex.org/W3138516171","https://openalex.org/W3195577433","https://openalex.org/W3207750165","https://openalex.org/W3209532394","https://openalex.org/W4221143402","https://openalex.org/W4224821742","https://openalex.org/W4361194507","https://openalex.org/W4386185623","https://openalex.org/W4390873312","https://openalex.org/W4391591725","https://openalex.org/W4396986653","https://openalex.org/W4399657827","https://openalex.org/W4400572780","https://openalex.org/W4403012121","https://openalex.org/W4403570351","https://openalex.org/W4403853618","https://openalex.org/W4404988257","https://openalex.org/W4407815020","https://openalex.org/W4413144805","https://openalex.org/W4413145886","https://openalex.org/W6931203431","https://openalex.org/W7133195915","https://openalex.org/W7133209731","https://openalex.org/W7133243546"],"related_works":["https://openalex.org/W4402727234","https://openalex.org/W4386065600","https://openalex.org/W7123768003","https://openalex.org/W4382458283","https://openalex.org/W4416574112","https://openalex.org/W4414421801","https://openalex.org/W4376122763","https://openalex.org/W7138423787","https://openalex.org/W4382498719","https://openalex.org/W4405036137"],"abstract_inverted_index":{"Foundation":[0],"image-text":[1,21],"models":[2,22,148,202],"such":[3,111],"as":[4,112],"CLIP":[5,58,78],"with":[6,29,172,222],"zero-shot":[7,31,155,179],"capabilities":[8],"enable":[9],"a":[10,17,45,98,144],"wide":[11],"array":[12],"of":[13,20,73,101,115,124,147,181],"applications.":[14],"MobileCLIP":[15,37,74],"is":[16],"recent":[18],"family":[19,146],"at":[23,157,194],"3-15ms":[24],"latency":[25],"and":[26,41,44,57,62,95,130,151,189,204],"50-150M":[27],"parameters":[28],"state-of-the-art":[30,153],"accuracy.":[32],"The":[33,210],"main":[34],"ingredients":[35],"in":[36,118,166],"were":[38],"its":[39],"low-latency":[40],"light":[42],"architectures":[43],"novel":[46],"multi-modal":[47,70],"reinforced":[48,71,220],"training":[49,72],"that":[50],"made":[51],"knowledge":[52,120],"distillation":[53],"from":[54,134],"multiple":[55,140],"caption-generators":[56],"teachers":[59,89,224],"efficient,":[60],"scalable,":[61],"reproducible.":[63],"In":[64,160],"this":[65],"paper,":[66],"we":[67,162],"improve":[68],"the":[69,83,92,113,122,131,178,205],"through:":[75],"1)":[76],"better":[77],"teacher":[79],"ensembles":[80],"trained":[81,90],"on":[82,91,97,183,191],"DFN":[84,93,192],"dataset,":[85],"2)":[86],"improved":[87],"captioner":[88],"dataset":[94],"fine-tuned":[96],"diverse":[99],"selection":[100],"high-quality":[102],"image-caption":[103],"datasets.":[104],"We":[105,142,198],"discover":[106],"new":[107,145,219],"insights":[108],"through":[109],"ablations":[110],"importance":[114],"temperature":[116],"tuning":[117],"contrastive":[119],"distillation,":[121],"effectiveness":[123],"caption-generator":[125],"fine-tuning":[126],"for":[127,169],"caption":[128],"diversity,":[129],"additive":[132],"improvement":[133,165],"combining":[135],"synthetic":[136],"captions":[137],"generated":[138],"by":[139],"models.":[141],"train":[143],"called":[149],"MobileCLIP2":[150],"achieve":[152],"ImageNet-1k":[154,167,184],"accuracies":[156],"low":[158],"latencies.":[159],"particular,":[161],"observe":[163],"2.2%":[164],"accuracy":[168,180],"MobileCLIP2-B":[170],"compared":[171],"MobileCLIP-B":[173],"architecture.":[174],"Notably,":[175],"MobileCLIP2-S4":[176],"matches":[177],"SigLIP-SO400M/14":[182],"while":[185],"being":[186],"2$\\times$":[187],"smaller":[188],"improves":[190],"ViT-L/14":[193],"2.5$\\times$":[195],"lower":[196],"latency.":[197],"release":[199],"our":[200],"pretrained":[201],"(https://github.com/apple/ml-mobileclip)":[203],"data":[206,211],"generation":[207,212],"code":[208,213],"(https://github.com/apple/ml-mobileclip-dr).":[209],"makes":[214],"it":[215],"easy":[216],"to":[217],"create":[218],"datasets":[221],"arbitrary":[223],"using":[225],"distributed":[226],"scalable":[227],"processing.":[228]},"counts_by_year":[{"year":2026,"cited_by_count":9},{"year":2025,"cited_by_count":2}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-10T00:00:00"}