Asarar Hankali da LPIPS
Asarar hasashe tana auna yadda hotuna biyu masu kama da juna suke kallon mutane ta hanyar kwatanta fasalin cibiyar sadarwa mai zurfi maimakon danyen pixels.
Dubawa
It matters because pixel-by-pixel comparison wrongly punishes tiny shifts and blurs detail, while perceptual loss rewards sharp, realistic results.
Zurfafa nutsewa
Hasara ta al'ada kamar L2 (kuskuren murabba'i) kwatanta hotuna pixel-by-pixel, don haka motsi-pixel guda ɗaya ko nau'in rubutu daban-daban yana kama da babban kuskure ko da yake ɗan adam ba sa lura. Asarar fahimta a maimakon haka tana gudanar da hotuna biyu ta hanyar sadarwar da aka riga aka horar (sau da yawa VGG) kuma tana kwatanta kunnawa daga tsaka-tsaki. Saboda waɗannan fasalulluka suna ɓoye gefuna, laushi, da sassan abu maimakon madaidaicin ƙimar pixel, asarar ta daidaita da kyau tare da hukuncin ɗan adam, yana ƙarfafa fitar da kaifi, amintaccen bayanan ma'ana. LPIPS (Kwantar da Hankalin Hoto Patch), wanda Zhang et al ya gabatar. a cikin 2018, ya tsara wannan: yana fitar da siffofi masu zurfi, yana daidaita su, kuma yana amfani da ma'aunin nauyi da aka kwatanta da dubban hukunce-hukuncen kamanni na ɗan adam, yana samar da maki ɗaya tazara inda ƙananan yana nufin mafi fahimta iri ɗaya.
Fahimtar Fasaha
LPIPS yana wucewa duka hotuna ta hanyar kafaffen kashin baya (VGG, AlexNet, ko SqueezeNet), naúrar tana daidaita kunna tashoshi a yadudduka da yawa, sannan ta ɗauki bambancin murabba'i a kowane wuri na sarari. Ƙananan ma'aunin ma'auni na kowane tashoshi da aka koya yana daidaita waɗannan bambance-bambancen kafin a ƙididdige su ta sarari da taƙaita su a cikin yadudduka. An horar da waɗancan ma'aunin nauyi akan tsarin bayanan BAPPS na hukunce-hukuncen zaɓi na zaɓi biyu na ɗan adam, don haka ma'aunin yana nuna abin da mutane ke fahimta a zahiri maimakon ɗan nesa mai nisa.
Dabarun Tasiri
Gudu da sikelin
Kayayyakin AI na iya sarrafa aiki da bincike, ganowa, da ayyuka masu alama a sikelin.
Gina zaɓuɓɓuka
Ƙungiyoyin ƙirƙira za su iya samar da ra'ayoyi cikin sauri tare da ƙarancin bita da hannu.
Ƙungiya da aikin aiki
Ayyuka na iya amfani da siginar hoto da bidiyo waɗanda a baya suke da wahalar aiwatarwa.
Makomar Asarar Hankali da LPIPS
Ma'auni na fahimta suna jujjuya daga kasusuwan CNN zuwa fasali daga nau'ikan kulawa da kai da hangen nesa kamar DINO da CLIP, waɗanda ke ɗaukar ingantattun ilimin tarukan. Yi tsammanin haɗin kai mai ƙarfi tare da horarwa-samfurin watsawa da kimanta rubutu-zuwa-hoto, da makin fahimta da aka kunna don daidaiton bidiyo na ɗan lokaci. Masu bincike kuma suna binciken wuraren makafi na LPIPS: ana iya yaudare shi da gaba kuma yana da alaƙa da inganci da aminci sosai, yana ƙarfafa sabbin ma'auni masu alaƙa da ɗan adam kamar DITS da hanyoyin haɗin gwiwa.
Aiwatar da Gaskiyar Duniya
Horar da manyan cibiyoyin sadarwa (misali, SRGAN) don haka hotuna masu girman gaske suna kallon kaifi da rubutu maimakon blush.
Ƙimar damfara hoto da codecs ta hanyar ƙididdige yadda fahimtar yanayin rufe hoton da aka yanke zuwa asali.
Canja wurin salon jagora, inda abun ciki ya dace ta hanyar zurfin fasalin VGG maimakon ainihin pixels.
Benchmarking GAN da masu samar da hoto na watsawa ta hanyar ba da rahoton tazarar LPIPS tsakanin hotuna da aka ƙirƙira da na gaske.
Hatsari & Tsare-tsare
Haƙƙoƙin hoto da yarda na iya zama haxarin doka idan ba a fayyace ba.
Ayyukan samfuri na iya bambanta a ko'ina cikin haske, ƙididdiga, da mahalli.
Ƙarya tabbataccen ƙila ba za a iya lura da shi ba sai dai idan an kula da ƙofofin amincewa.
Taswirar Hanya
Ƙayyade ma'auni na karɓa don daidaito, tunowa, da farashi na kuskure.
Gwada tare da bayanan da suka dace da ainihin yanayin samarwa.
Ƙara bita na ɗan adam don ƙarancin amincewa ko tsinkaya mai tasiri.
Bi diddigin ƙirar ƙira kuma sake ingantawa bayan canje-canjen kamara ko saitin bayanai.
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 Perceptual Loss and LPIPS 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
Asarar Hankali don Gano Rashin Daidaito
Tambayoyin da ake yawan yi
What is Perceptual Loss and LPIPS?
Asarar hasashe tana auna yadda hotuna biyu masu kama da juna suke kallon mutane ta hanyar kwatanta fasalin cibiyar sadarwa mai zurfi maimakon danyen pixels. Yana da mahimmanci saboda kwatancen pixel-by-pixel ba daidai ba yana azabtar da ƙananan canje-canje da ɓarna daki-daki, yayin da hasarar fahimi ke ba da sakamako mai kaifi.
Me yasa asarar L2 na tushen pixel sau da yawa yana yin kuskuren kamanni na hoto idan aka kwatanta da hasarar fahimta?
L2 yana kwatanta pixels kai tsaye, don haka ƙananan sauye-sauye na sarari ko bambance-bambancen rubutu suna yin rijista azaman manyan kurakurai ko da hoton yayi kyau ga mutum.
Menene LPIPS da farko ke kwatanta tsakanin hotuna biyu?
LPIPS yana fitar da ayyukan kunnawa mai zurfi daga kafaffen kashin baya kuma yana auna nisan su, wanda yayi daidai da fahimtar ɗan adam fiye da pixels.
Yaya aka ƙayyade ma'aunin tashoshi ɗaya a cikin LPIPS?
An koyi ma'aunin nauyi don dacewa da babban ma'aunin bayanai (BAPPS) na zaɓin kamanni-madaidaicin-tilasta-zaɓin ɗan adam.
Wace hanyar sadarwar kashin baya aka fi haɗawa da hasarar fahimi (fasali)?
Matsakaicin taswirorin fasalin VGG sun zama ma'auni don hasarar hasashe a cikin ayyuka kamar babban ƙuduri da canza salo.
Wane aiki ya fi fa'ida kai tsaye daga amfani da hasarar fahimta maimakon L2 mai tsabta?
Babban hanyoyin sadarwa waɗanda aka horar da hasarar hasashe suna haifar da ƙwaƙƙwaran ƙira, mafi inganci, yayin da L2 ke ƙoƙarin samar da matsakaita mara kyau.