{"id":"https://openalex.org/W4417068532","doi":"https://doi.org/10.1109/iccv51701.2025.01522","title":"DiTFastAttnV2: Head-Wise Attention Compression for Multi-Modality Diffusion Transformers","display_name":"DiTFastAttnV2: Head-Wise Attention Compression for Multi-Modality Diffusion Transformers","publication_year":2025,"publication_date":"2025-10-19","ids":{"openalex":"https://openalex.org/W4417068532","doi":"https://doi.org/10.1109/iccv51701.2025.01522"},"language":"en","primary_location":{"id":"doi:10.1109/iccv51701.2025.01522","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.01522","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2503.22796","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102956076","display_name":"Hanling Zhang","orcid":"https://orcid.org/0000-0001-5954-1424"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hanling Zhang","raw_affiliation_strings":["Tsinghua University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111108543","display_name":"Ruohan Su","orcid":"https://orcid.org/0009-0000-8367-9140"},"institutions":[{"id":"https://openalex.org/I24943067","display_name":"Fudan University","ror":"https://ror.org/013q1eq08","country_code":"CN","type":"education","lineage":["https://openalex.org/I24943067"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Rundong Su","raw_affiliation_strings":["Fudan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Fudan University","institution_ids":["https://openalex.org/I24943067"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5084059508","display_name":"Zhihang Yuan","orcid":"https://orcid.org/0000-0001-9980-7717"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhihang Yuan","raw_affiliation_strings":["Tsinghua University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012986920","display_name":"Pengtao Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I24943067","display_name":"Fudan University","ror":"https://ror.org/013q1eq08","country_code":"CN","type":"education","lineage":["https://openalex.org/I24943067"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Pengtao Chen","raw_affiliation_strings":["Fudan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Fudan University","institution_ids":["https://openalex.org/I24943067"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Mingzhu Shen","orcid":null},"institutions":[{"id":"https://openalex.org/I47508984","display_name":"Imperial College London","ror":"https://ror.org/041kmwe10","country_code":"GB","type":"education","lineage":["https://openalex.org/I47508984"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Mingzhu Shen","raw_affiliation_strings":["Imperial College London"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Imperial College London","institution_ids":["https://openalex.org/I47508984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049910854","display_name":"Shengen Yan","orcid":"https://orcid.org/0009-0005-3858-7972"},"institutions":[{"id":"https://openalex.org/I24943067","display_name":"Fudan University","ror":"https://ror.org/013q1eq08","country_code":"CN","type":"education","lineage":["https://openalex.org/I24943067"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yibo Fan","raw_affiliation_strings":["Fudan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Fudan University","institution_ids":["https://openalex.org/I24943067"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049910854","display_name":"Shengen Yan","orcid":"https://orcid.org/0009-0005-3858-7972"},"institutions":[{"id":"https://openalex.org/I4210148037","display_name":"Infinity (United States)","ror":"https://ror.org/03k7v4244","country_code":"US","type":"company","lineage":["https://openalex.org/I4210148037"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shengen Yan","raw_affiliation_strings":["Infinigence AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Infinigence AI","institution_ids":["https://openalex.org/I4210148037"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Guohao Dai","orcid":null},"institutions":[{"id":"https://openalex.org/I4210148037","display_name":"Infinity (United States)","ror":"https://ror.org/03k7v4244","country_code":"US","type":"company","lineage":["https://openalex.org/I4210148037"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Guohao Dai","raw_affiliation_strings":["Infinigence AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Infinigence AI","institution_ids":["https://openalex.org/I4210148037"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100445359","display_name":"Yu Wang","orcid":"https://orcid.org/0000-0003-3008-8712"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yu Wang","raw_affiliation_strings":["Tsinghua University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University","institution_ids":["https://openalex.org/I99065089"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":4,"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":"16399","last_page":"16409"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.6323000192642212,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.6323000192642212,"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/T10901","display_name":"Advanced Data Compression Techniques","score":0.10329999774694443,"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/T11019","display_name":"Image Enhancement Techniques","score":0.10000000149011612,"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/speedup","display_name":"Speedup","score":0.7160000205039978},{"id":"https://openalex.org/keywords/bridging","display_name":"Bridging (networking)","score":0.5900999903678894},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.5663999915122986},{"id":"https://openalex.org/keywords/flops","display_name":"FLOPS","score":0.4350999891757965},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.4332999885082245},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.42969998717308044},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.37790000438690186},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.36899998784065247}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7595999836921692},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.7160000205039978},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.5900999903678894},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.5663999915122986},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.4350999891757965},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.4332999885082245},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.42969998717308044},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3837999999523163},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.37790000438690186},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.36899998784065247},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3686999976634979},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.35760000348091125},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.35409998893737793},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.32019999623298645},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.3075999915599823},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.2973000109195709},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C25797200","wikidata":"https://www.wikidata.org/wiki/Q828137","display_name":"Compression ratio","level":3,"score":0.2800000011920929},{"id":"https://openalex.org/C13481523","wikidata":"https://www.wikidata.org/wiki/Q412438","display_name":"Image compression","level":4,"score":0.2741999924182892},{"id":"https://openalex.org/C71134354","wikidata":"https://www.wikidata.org/wiki/Q458825","display_name":"Kernel density estimation","level":3,"score":0.2685000002384186},{"id":"https://openalex.org/C57654395","wikidata":"https://www.wikidata.org/wiki/Q1097775","display_name":"Compression artifact","level":5,"score":0.26080000400543213},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.2563000023365021}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1109/iccv51701.2025.01522","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.01522","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2503.22796","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.22796","pdf_url":"https://arxiv.org/pdf/2503.22796","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2503.22796","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.22796","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":"pmh:oai:arXiv.org:2503.22796","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.22796","pdf_url":"https://arxiv.org/pdf/2503.22796","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G1129054162","display_name":null,"funder_award_id":"62325405,62104128,U19B2019,U21B2031,61832007,62204164","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320329777","display_name":"Beijing National Research Center For Information Science And Technology","ror":null}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Text-to-image":[0],"generation":[1,117,139],"models,":[2],"especially":[3],"Multimodal":[4],"Diffusion":[5],"Transformers":[6],"(MMDiT),":[7],"have":[8],"shown":[9],"remarkable":[10],"progress":[11],"in":[12,25,48,129],"generating":[13],"high-quality":[14],"images.":[15],"However,":[16],"these":[17],"models":[18],"often":[19],"face":[20],"significant":[21],"computational":[22],"bottlenecks,":[23],"particularly":[24],"attention":[26,47,56,70,77,130],"mechanisms,":[27],"which":[28],"hinder":[29],"their":[30],"scalability":[31],"and":[32,66,71,98,132],"efficiency.":[33],"In":[34],"this":[35,81],"paper,":[36],"we":[37,58],"introduce":[38],"DiTFastAttnV2,":[39],"a":[40,126],"post-training":[41],"compression":[42,110],"method":[43],"designed":[44],"to":[45,74,112],"accelerate":[46],"MMDiT.":[49],"Through":[50],"an":[51,86],"in-depth":[52],"analysis":[53],"of":[54],"MMDiT's":[55],"patterns,":[57],"identify":[59],"key":[60],"differences":[61],"from":[62],"prior":[63],"DiT-based":[64],"methods":[65,97],"propose":[67],"head-wise":[68],"arrow":[69],"caching":[72],"mechanisms":[73],"dynamically":[75],"adjust":[76],"heads,":[78],"effectively":[79],"bridging":[80],"gap.":[82],"We":[83],"also":[84],"design":[85],"Efficient":[87],"Fused":[88],"Kernel":[89],"for":[90,108],"further":[91],"acceleration.":[92],"By":[93],"leveraging":[94],"local":[95],"metric":[96],"optimization":[99],"techniques,":[100],"our":[101],"approach":[102],"significantly":[103],"reduces":[104],"the":[105,121],"search":[106],"time":[107],"optimal":[109],"schemes":[111],"just":[113],"minutes":[114],"while":[115],"maintaining":[116],"quality.":[118],"Furthermore,":[119],"with":[120],"customized":[122],"kernel,":[123],"DiTFastAttnV2":[124],"achieves":[125],"68%":[127],"reduction":[128],"FLOPs":[131],"1.5x":[133],"end-to-end":[134],"speedup":[135],"on":[136],"2K":[137],"image":[138],"without":[140],"compromising":[141],"visual":[142],"fidelity.":[143]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2025-10-10T00:00:00"}