{"id":"https://openalex.org/W4386065629","doi":"https://doi.org/10.1109/cvpr52729.2023.01554","title":"Bit-shrinking: Limiting Instantaneous Sharpness for Improving Post-training Quantization","display_name":"Bit-shrinking: Limiting Instantaneous Sharpness for Improving Post-training Quantization","publication_year":2023,"publication_date":"2023-06-01","ids":{"openalex":"https://openalex.org/W4386065629","doi":"https://doi.org/10.1109/cvpr52729.2023.01554"},"language":"en","primary_location":{"id":"doi:10.1109/cvpr52729.2023.01554","doi":"https://doi.org/10.1109/cvpr52729.2023.01554","is_oa":false,"landing_page_url":"https://doi.org/10.1109/cvpr52729.2023.01554","pdf_url":null,"source":{"id":"https://openalex.org/S4210176548","display_name":"IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings","issn_l":"1063-6919","issn":["1063-6919","2332-564X","2575-7075"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-1"],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100746862","display_name":"Lin Chen","orcid":"https://orcid.org/0000-0001-5144-7855","observed_orcids":["https://orcid.org/0000-0001-5144-7855"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chen Lin","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101685717","display_name":"Bo Peng","orcid":"https://orcid.org/0000-0003-1514-6881","observed_orcids":["https://orcid.org/0000-0003-1514-6881"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bo Peng","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046994319","display_name":"Zheyang Li","orcid":"https://orcid.org/0000-0002-0229-8707","observed_orcids":["https://orcid.org/0000-0002-0229-8707"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zheyang Li","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031951156","display_name":"Wenming Tan","orcid":"https://orcid.org/0000-0003-1338-4536","observed_orcids":["https://orcid.org/0000-0003-1338-4536"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wenming Tan","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100413326","display_name":"Ye Ren","orcid":"https://orcid.org/0000-0003-4558-5834","observed_orcids":["https://orcid.org/0000-0003-4558-5834"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ye Ren","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022098030","display_name":"Jun Ying Xiao","orcid":"https://orcid.org/0000-0002-6077-2155","observed_orcids":["https://orcid.org/0000-0002-6077-2155"]},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jun Xiao","raw_affiliation_strings":["Zhe Jiang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Zhe Jiang University","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5085955762","display_name":"Shiliang Pu","orcid":"https://orcid.org/0000-0001-5269-7821","observed_orcids":["https://orcid.org/0000-0001-5269-7821"]},"institutions":[{"id":"https://openalex.org/I4401727007","display_name":"Hikvision (China)","ror":"https://ror.org/02jzypx27","country_code":"CN","type":"company","lineage":["https://openalex.org/I4401727007"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shiliang Pu","raw_affiliation_strings":["Hikvision Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hikvision Research Institute","institution_ids":["https://openalex.org/I4401727007"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.3238,"has_fulltext":false,"cited_by_count":13,"citation_normalized_percentile":{"value":0.88340678,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":88,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"16196","last_page":"16205"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9984999895095825,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9984999895095825,"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":0.0003000000142492354,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.00019999999494757503,"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/post-training-quantization","display_name":"post-training quantization","score":0.9929999709129333},{"id":"https://openalex.org/keywords/loss-landscape","display_name":"loss landscape","score":0.7229999899864197},{"id":"https://openalex.org/keywords/quantization-noise","display_name":"quantization noise","score":0.8389999866485596},{"id":"https://openalex.org/keywords/local-minima","display_name":"local minima","score":0.746999979019165},{"id":"https://openalex.org/keywords/low-bit-quantization","display_name":"low-bit quantization","score":0.7250000238418579},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.6570000052452087},{"id":"https://openalex.org/keywords/vision-transformer","display_name":"vision transformer","score":0.6980000138282776}],"concepts":[{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.6781785488128662},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6645603179931641},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6569207310676575},{"id":"https://openalex.org/C117011727","wikidata":"https://www.wikidata.org/wiki/Q1278488","display_name":"Bit (key)","level":2,"score":0.5066028833389282},{"id":"https://openalex.org/C56281022","wikidata":"https://www.wikidata.org/wiki/Q11308039","display_name":"Resizing","level":3,"score":0.4452197551727295},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3851955533027649},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2724837064743042},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.13595271110534668},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.0681394636631012},{"id":"https://openalex.org/C2910001868","wikidata":"https://www.wikidata.org/wiki/Q458","display_name":"European union","level":2,"score":0.0},{"id":"https://openalex.org/C105639569","wikidata":"https://www.wikidata.org/wiki/Q582577","display_name":"Economic policy","level":1,"score":0.0},{"id":"https://openalex.org/C78519656","wikidata":"https://www.wikidata.org/wiki/Q101333","display_name":"Mechanical engineering","level":1,"score":0.0},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/cvpr52729.2023.01554","doi":"https://doi.org/10.1109/cvpr52729.2023.01554","is_oa":false,"landing_page_url":"https://doi.org/10.1109/cvpr52729.2023.01554","pdf_url":null,"source":{"id":"https://openalex.org/S4210176548","display_name":"IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings","issn_l":"1063-6919","issn":["1063-6919","2332-564X","2575-7075"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-1"],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":49,"referenced_works":["https://openalex.org/W639708223","https://openalex.org/W1522301498","https://openalex.org/W2108598243","https://openalex.org/W2143163787","https://openalex.org/W2183341477","https://openalex.org/W2194775991","https://openalex.org/W2300242332","https://openalex.org/W2469490737","https://openalex.org/W2515385951","https://openalex.org/W2550403740","https://openalex.org/W2612445135","https://openalex.org/W2798742790","https://openalex.org/W2808868252","https://openalex.org/W2884561390","https://openalex.org/W2896718327","https://openalex.org/W2911595986","https://openalex.org/W2915478146","https://openalex.org/W2950141105","https://openalex.org/W2954582488","https://openalex.org/W2962933129","https://openalex.org/W2963363373","https://openalex.org/W2963521187","https://openalex.org/W2982479999","https://openalex.org/W2985339713","https://openalex.org/W2995738369","https://openalex.org/W3020212829","https://openalex.org/W3034528892","https://openalex.org/W3034719990","https://openalex.org/W3034940165","https://openalex.org/W3035078980","https://openalex.org/W3035855442","https://openalex.org/W3091401866","https://openalex.org/W3094502228","https://openalex.org/W3103385169","https://openalex.org/W3116489684","https://openalex.org/W3122499249","https://openalex.org/W3138516171","https://openalex.org/W3165976180","https://openalex.org/W3166792421","https://openalex.org/W3170874841","https://openalex.org/W3174042481","https://openalex.org/W3175617644","https://openalex.org/W3214891595","https://openalex.org/W3215741372","https://openalex.org/W4221159612","https://openalex.org/W4283687200","https://openalex.org/W4411245276","https://openalex.org/W6784333009","https://openalex.org/W6804113219"],"related_works":["https://openalex.org/W3035855442","https://openalex.org/W4411086485","https://openalex.org/W7168245335","https://openalex.org/W4220934237","https://openalex.org/W4386076113","https://openalex.org/W4306312216","https://openalex.org/W3038470071","https://openalex.org/W4408841131","https://openalex.org/W2954582488","https://openalex.org/W4221159612"],"abstract_inverted_index":{"Post-training":[0],"quantization":[1,89,107,172],"(PTQ)":[2],"is":[3,28,61,82],"an":[4],"effective":[5],"compression":[6],"method":[7],"to":[8,117,150,178],"reduce":[9],"the":[10,42,54,79,85,100,104,111,119,131,142,151,155,176,190,193,199,217,240],"model":[11,19],"size":[12],"and":[13,30,123,174,186,205,211,228,234],"computational":[14],"cost.":[15],"However,":[16],"quantizing":[17],"a":[18,21,96,137,160],"into":[20,65],"low-bit":[22,73],"one,":[23],"e.g.,":[24],"lower":[25],"than":[26],"4,":[27],"difficult":[29],"often":[31],"results":[32],"in":[33,72,144,196],"non-negligible":[34],"performance":[35],"degradation.":[36],"To":[37,91,109],"address":[38],"this,":[39],"we":[40,94,115,135],"investigate":[41],"loss":[43,59,101,113],"landscapes":[44],"of":[45,87,106,128,192],"quantized":[46,223],"networks":[47],"with":[48,56],"various":[49],"bit-widths.":[50],"We":[51],"show":[52],"that":[53],"network":[55,177],"more":[57,62],"ragged":[58,80],"surface,":[60,114],"easily":[63],"trapped":[64],"bad":[66],"local":[67],"minima,":[68],"which":[69,102,238],"mostly":[70],"appears":[71],"quantization.":[74],"A":[75],"deeper":[76],"analysis":[77],"indicates,":[78],"surface":[81],"caused":[83],"by":[84,153],"injection":[86],"excessive":[88],"noise.":[90,108],"this":[92],"end,":[93],"detach":[95],"sharpness":[97,120,157],"term":[98,121,158],"from":[99,147],"reflects":[103],"impact":[105],"smooth":[110],"rugged":[112],"propose":[116],"limit":[118],"small":[122,170],"stable":[124],"during":[125],"optimization.":[126],"Instead":[127],"directly":[129],"optimizing":[130],"target":[132,152],"bit":[133],"network,":[134],"design":[136],"self-adapted":[138],"shrinking":[139],"scheduler":[140],"for":[141],"bit-width":[143,149],"continuous":[145],"domain":[146],"high":[148],"limiting":[154],"increasing":[156],"within":[159,209,226],"proper":[161],"range.":[162],"It":[163],"can":[164],"be":[165],"viewed":[166],"as":[167],"iteratively":[168],"adding":[169],"\u201cinstant\u201d":[171],"noise":[173],"adjusting":[175],"eliminate":[179],"its":[180],"impact.":[181],"Widely":[182],"experiments":[183],"including":[184],"classification":[185],"detection":[187],"tasks":[188],"demonstrate":[189],"effectiveness":[191],"Bit-shrinking":[194],"strategy":[195],"PTQ.":[197],"On":[198,216],"Vision":[200],"Transformer":[201],"models,":[202],"our":[203,221],"INT8":[204],"INT6":[206],"models":[207,224],"drop":[208,225],"0.5%":[210],"1.5%":[212],"Top-1":[213,230],"accuracy,":[214],"respectively.":[215],"traditional":[218],"CNN":[219],"networks,":[220],"INT4":[222],"1.3%":[227],"3.5%":[229],"accuracy":[231],"on":[232],"ResNet18":[233],"MobileNetV2":[235],"without":[236],"fine-tuning,":[237],"achieves":[239],"state-of-the-art":[241],"performance.":[242]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":6},{"year":2024,"cited_by_count":4},{"year":2023,"cited_by_count":1}],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2023-08-23T00:00:00"}