Juyin Juyin Halitta Mai Zurfi
Zurfafawar juzu'ai masu iya rabuwa suna haifar da daidaitaccen juyi zuwa matakai biyu masu rahusa, yana yanke adadin ninkawa da sigogi.
Dubawa
They are the trick that lets neural networks run on phones and edge devices without melting the battery.
Zurfafa nutsewa
Daidaitaccen juzu'i yana haɗa bayanai a cikin sararin samaniya da tashoshi a cikin aiki mai yawa guda ɗaya, mai tsada. Juyin juyayi mai zurfi mai zurfi ya raba wannan zuwa matakai biyu. Na farko, mataki mai zurfi yana aiki da ƙaramin tacewa kowane tashar shigarwa da kanta, yana ɗaukar tsarin sararin samaniya a cikin kowane tashoshi amma ba zai taɓa haɗa tashoshi ba. Na biyu, matakin da ya dace yana amfani da juzu'in 1x1 don haɗa tashoshi a kowane pixel, haɗa bayanan tashar ba tare da kallon maƙwabta ba. Ta hanyar ɓata matattarar sararin samaniya daga haɗawar tashoshi, jimlar ƙididdigewa tana faɗuwa da ƙarfi, sau 8 zuwa 9 don tacewa 3x3, tare da ƙarancin daidaito kaɗan. Wannan factorization shine kashin bayan MobileNet da Xception.
Fahimtar Fasaha
Don taswirar shigarwar kernel 3x3 M zuwa abubuwan N akan taswirar fasali, daidaitaccen juyi yana kashe kusan sau 9 sau M N ninka-ƙara kowane wuri. Sigar keɓancewar ta biya sau 9 M don ɓangaren zurfin hikima da M sau N don madaidaicin 1x1. Matsakaicin shine kusan 1/N + 1/9, don haka ga manyan N tanadin yana kusanci 1/9 na sararin samaniya.
Dabarun Tasiri
Kudin da kasafin kuɗi
Hukunce-hukuncen gine-gine suna haifar da aiki da tsadar aiki na shekaru.
Shawarwari masu haske
Ilimin fasaha yana taimaka wa ƙungiyoyi su zaɓi tari mai kyau, ba kawai sabon abu ba.
Kula da inganci
Zaɓuɓɓukan injiniya mafi kyau suna rage abin dogaro a cikin samarwa.
Makomar Juyin Juyin Halitta Mai Zurfi
Juyin juyayi mai zurfi mai zurfi ya kasance babban madaidaicin ƙirar hangen nesa kuma yana ƙara bayyana a cikin ƙirar CNN-canzawa kamar MobileViT da ConvNeXt tubalan. Yayin da na'urar AI ke girma, masu haɓaka kayan aiki suna ƙara tallafi na asali don ops mai zurfi. Yi tsammanin ci gaba da amfani a cikin hangen nesa na ainihi, na'urori masu auna firikwensin, da kowane saiti inda latency, ƙwaƙwalwar ajiya, da kasafin kuɗin makamashi ke da ƙarfi, galibi haɗe tare da ƙididdigewa da binciken gine-ginen jijiyoyi.
Aiwatar da Gaskiyar Duniya
MobileNet da MobileNetV2 suna amfani da su don gudanar da rarrabuwar hoto kai tsaye akan wayoyin hannu tare da ƙarancin jinkiri
Rabe-raben hoto na ainihin lokaci da blur bango a cikin aikace-aikacen kiran bidiyo sun dogara da kasusuwa masu nauyi masu nauyi.
Gano abu akan na'urar a cikin kyamarori masu tsaro da jirage marasa matuki, inda wuta da lissafi ke iyakance
Xception yana amfani da su a sikelin don tura daidaiton ImageNet yayin sarrafa ƙidayar siga
Hatsari & Tsare-tsare
Haɓaka ma'auni ɗaya na iya ɓoye manyan raunin tsarin.
Sau da yawa ana raina kayan more rayuwa da kuma kuɗin kulawa.
Tsaro da gibin lura na iya girma yayin da tsarin ke ƙara haɓaka.
Taswirar Hanya
Ƙayyade latency, inganci, da maƙasudin farashi kafin aiwatarwa.
Alamar ma'auni a ƙarƙashin ainihin kaya da yanayin bayanai.
Kula da kayan aiki don kurakurai, ɗigo, da tasirin mai amfani.
Shirya bijirowa da hanyoyin mayar da martani kafin sikeli.
Ci gaba da Bincike
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Depthwise Separable Convolutions quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Jagora na gaba
Juyin Juyin Halitta
Tambayoyin da ake yawan yi
What is Depthwise Separable Convolutions?
Zurfafawar juzu'ai masu iya rabuwa suna haifar da daidaitaccen juyi zuwa matakai biyu masu rahusa, yana yanke adadin ninkawa da sigogi. Su ne dabarar da ke ba da damar hanyoyin sadarwa na jijiyoyi suyi aiki akan wayoyi da na'urorin gefen ba tare da narkar da baturi ba.
Wadanne matakai guda biyu ne da zurfin juzu'in rabewa ke raba ma'auni juzu'i zuwa ciki?
Mataki mai zurfi yana tace kowane tashoshi a sarari da kansa, kuma matakin 1x1 mai ma'ana sannan ya haɗa bayanai a cikin tashoshi.
A cikin zurfin mataki, yaya ake kula da tashoshi na shigarwa?
Juyin zurfafa yana aiwatar da keɓantaccen tacewa na sarari ga kowane tashar shigarwa ba tare da haɗa tashoshi ba, wanda shine abin da ya sa ya zama mai arha.
Kusan nawa 3x3 zurfin juzu'in rabuwa mai zurfi zai iya rage ƙididdigewa idan aka kwatanta da daidaitaccen juyi na 3x3 tare da tashoshi masu fitarwa da yawa?
Don kwaya 3x3 rabon tanadi yana kusanci 1/9 lokacin da adadin tashoshin fitarwa ya yi girma, yana ba da ƙarancin ayyuka sau 8 zuwa 9.
Menene rawar juzu'i mai ma'ana (1x1) a cikin wannan ƙira?
Juyin juzu'i mai ma'ana 1x1 yana haɗa tashoshi a kowane pixel, wanda mataki mai zurfi ya guji yin da gangan.
Wanne sanannen gine-gine ne ya bazu cikin zurfin juzu'i mai iya rabuwa don na'urorin hannu?
MobileNet an gina shi ne a kusa da zurfin juzu'ai masu iya rabuwa musamman don ba da damar ingantacciyar fahimta akan wayoyi.