{"id":"https://openalex.org/W7163904553","doi":"https://doi.org/10.48550/arxiv.2606.06850","title":"CFRNet: Cycle-Consistent Fixed-Point Training for Real-Time Blind Face Restoration on Consumer Embedded NPUs","display_name":"CFRNet: Cycle-Consistent Fixed-Point Training for Real-Time Blind Face Restoration on Consumer Embedded NPUs","publication_year":2026,"publication_date":"2026-06-05","ids":{"openalex":"https://openalex.org/W7163904553","doi":"https://doi.org/10.48550/arxiv.2606.06850"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.06850","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06850","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.2606.06850","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138123226","display_name":"Fuchen Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Fuchen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138133658","display_name":"Xinyang Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xinyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100705245","display_name":"Yahui Zhang","orcid":"https://orcid.org/0009-0000-1841-375X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yahui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138139356","display_name":"Yuhan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yuhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138142906","display_name":"Jiahong Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Jiahong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138170545","display_name":"Zhuohan Qin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Zhuohan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5021713410","display_name":"Wenbo Ma","orcid":"https://orcid.org/0000-0001-6409-549X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Wenbo","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/T12301","display_name":"Facial Nerve Paralysis Treatment and Research","score":0.4577000141143799,"subfield":{"id":"https://openalex.org/subfields/2728","display_name":"Neurology"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},"topics":[{"id":"https://openalex.org/T12301","display_name":"Facial Nerve Paralysis Treatment and Research","score":0.4577000141143799,"subfield":{"id":"https://openalex.org/subfields/2728","display_name":"Neurology"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.21310000121593475,"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/T11448","display_name":"Face recognition and analysis","score":0.1137000024318695,"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/convolutional-neural-network","display_name":"Convolutional neural network","score":0.5533000230789185},{"id":"https://openalex.org/keywords/codebook","display_name":"Codebook","score":0.4966000020503998},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.46959999203681946},{"id":"https://openalex.org/keywords/prior-probability","display_name":"Prior probability","score":0.4235999882221222},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.4050000011920929},{"id":"https://openalex.org/keywords/inpainting","display_name":"Inpainting","score":0.39160001277923584},{"id":"https://openalex.org/keywords/face","display_name":"Face (sociological concept)","score":0.3885999917984009},{"id":"https://openalex.org/keywords/image-restoration","display_name":"Image restoration","score":0.3540000021457672},{"id":"https://openalex.org/keywords/perception","display_name":"Perception","score":0.3377000093460083}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7534999847412109},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.5533000230789185},{"id":"https://openalex.org/C127759330","wikidata":"https://www.wikidata.org/wiki/Q637416","display_name":"Codebook","level":2,"score":0.4966000020503998},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48910000920295715},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.46959999203681946},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.4235999882221222},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4050000011920929},{"id":"https://openalex.org/C11727466","wikidata":"https://www.wikidata.org/wiki/Q1628157","display_name":"Inpainting","level":3,"score":0.39160001277923584},{"id":"https://openalex.org/C2779304628","wikidata":"https://www.wikidata.org/wiki/Q3503480","display_name":"Face (sociological concept)","level":2,"score":0.3885999917984009},{"id":"https://openalex.org/C106430172","wikidata":"https://www.wikidata.org/wiki/Q6002272","display_name":"Image restoration","level":4,"score":0.3540000021457672},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3400000035762787},{"id":"https://openalex.org/C26760741","wikidata":"https://www.wikidata.org/wiki/Q160402","display_name":"Perception","level":2,"score":0.3377000093460083},{"id":"https://openalex.org/C126780896","wikidata":"https://www.wikidata.org/wiki/Q899871","display_name":"Distortion (music)","level":4,"score":0.3361999988555908},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.33169999718666077},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.3206999897956848},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.3124000132083893},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.31029999256134033},{"id":"https://openalex.org/C55020928","wikidata":"https://www.wikidata.org/wiki/Q3813865","display_name":"Image quality","level":3,"score":0.3059000074863434},{"id":"https://openalex.org/C2781137444","wikidata":"https://www.wikidata.org/wiki/Q237105","display_name":"Sharpening","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2896000146865845},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.2822999954223633},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.27709999680519104},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.27390000224113464},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.2694000005722046},{"id":"https://openalex.org/C2781235140","wikidata":"https://www.wikidata.org/wiki/Q275131","display_name":"Scratch","level":2,"score":0.26820001006126404},{"id":"https://openalex.org/C14185376","wikidata":"https://www.wikidata.org/wiki/Q30232","display_name":"Agile software development","level":2,"score":0.26350000500679016}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.06850","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06850","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.2606.06850","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06850","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Blind":[0],"face":[1,144],"restoration":[2],"on":[3,29,35,52,100,234,301],"consumer":[4,61,101],"devices":[5],"has":[6],"to":[7,48,72,75,130,141,246,273,291],"balance":[8],"image":[9],"quality":[10,257],"against":[11],"speed":[12],"and":[13,20,34,42,50,74,82,119,162,167,215,220,268,293],"memory.":[14],"Strong":[15],"methods":[16],"such":[17,37],"as":[18,38,132,254],"GFPGAN":[19],"CodeFormer":[21],"give":[22],"good":[23],"perceptual":[24,201],"quality,":[25],"but":[26,69],"they":[27,70],"rely":[28],"large":[30],"pretrained":[31],"generative":[32],"priors":[33],"operators":[36],"attention,":[39],"codebook":[40],"lookup,":[41],"style":[43],"modulation":[44],"that":[45,137,247,259,278,287],"are":[46],"hard":[47],"compile":[49],"quantize":[51],"the":[53,79,96,114,148,188,199,217,240,279,296],"small":[54],"neural":[55],"processing":[56],"units":[57],"(NPUs)":[58],"used":[59],"in":[60,227,232,298],"hardware.":[62],"Small":[63],"convolutional":[64],"restorers":[65],"run":[66,295],"fast":[67],"enough,":[68],"tend":[71],"over-smooth":[73],"leave":[76],"artifacts":[77],"around":[78],"eyes,":[80],"nose,":[81],"mouth.":[83],"We":[84,275],"present":[85],"CFRNet,":[86],"a":[87,133,142,163,193,235,255,284],"2.0,M-parameter":[88],"ResNet-style":[89],"restorer":[90],"for":[91,116],"on-device":[92],"use":[93],"at":[94,172,187,205,222,266],"$256\\times256$,":[95],"common":[97],"face-crop":[98],"size":[99],"NPUs.":[102],"The":[103,249],"main":[104],"idea":[105,281],"is":[106,209,264,288],"Cycle-Consistent":[107],"Fixed-Point":[108],"Training":[109],"(CCFP).":[110],"Instead":[111],"of":[112],"training":[113,153],"network":[115],"one":[117,213],"pass":[118],"then":[120],"running":[121],"it":[122,129,139,168],"several":[123],"times":[124],"by":[125],"hand,":[126],"we":[127,181,294],"train":[128],"act":[131],"fixed-point":[134],"operator,":[135],"so":[136],"applying":[138],"again":[140],"restored":[143],"does":[145],"not":[146],"change":[147],"face.":[149],"CCFP":[150],"uses":[151],"three":[152,206],"losses,":[154],"namely":[155],"progressive":[156],"multi-cycle":[157],"supervision,":[158],"an":[159,302],"idempotence":[160],"loss,":[161,166],"re-degradation":[164],"cycle":[165,231,250],"adds":[169],"no":[170,261],"cost":[171],"inference.":[173],"To":[174],"compare":[175],"fairly":[176],"under":[177],"our":[178],"deployment":[179],"limits,":[180],"retrain":[182],"all":[183],"baselines":[184,242],"from":[185],"scratch":[186],"same":[189,241,280],"$256\\times256$":[190],"resolution.":[191],"On":[192],"300-image":[194],"test":[195],"set,":[196],"CFRNet":[197],"reaches":[198],"best":[200,218,265],"score":[202],"(LPIPS":[203],"0.250":[204],"cycles,":[207],"which":[208],"31%":[210],"lower":[211],"than":[212],"cycle)":[214],"also":[216],"PSNR":[219,263],"SSIM":[221],"two":[223],"cycles.":[224],"It":[225],"runs":[226],"about":[228],"23,ms":[229],"per":[230],"INT8":[233],"HiSilicon":[236],"Hi3402":[237],"NPU,":[238],"while":[239],"cannot":[243],"be":[244],"compiled":[245],"chip.":[248],"count":[251],"$k$":[252],"acts":[253],"simple":[256],"knob":[258],"needs":[260],"retraining:":[262],"$k\\!=\\!2$":[267],"LPIPS":[269],"keeps":[270],"improving":[271],"up":[272],"$k\\!=\\!3$.":[274],"further":[276],"show":[277],"works":[282],"with":[283],"plain":[285],"CNN":[286],"even":[289],"easier":[290],"deploy,":[292],"model":[297],"real":[299],"time":[300],"in-car":[303],"driver-monitoring":[304],"board.":[305]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-09T00:00:00"}