Nisa Farko na Fréchet
Fréchet Inception Distance (FID) shine ma'auni na ma'auni don yin la'akari da yadda ainihin saƙon da aka ƙirƙira ya kasance.
Dubawa
It compares the statistics of real and generated images in a deep feature space — lower scores mean the fakes look closer to the real thing.
Zurfafa nutsewa
FID, wanda Heusel et al. a cikin 2017, ya gyara maɓalli mai mahimmanci a cikin Makin Farko na farko: bai taɓa kwatanta hotunan da aka ƙirƙira zuwa ainihin bayanan ba. FID tana ciyar da hotuna na gaske da waɗanda aka ƙirƙira ta hanyar hanyar sadarwa ta Inception-v3 da aka riga aka horar da ita kuma tana karanta siffa mai girman 2048 daga madogara mai zurfi don kowane hoto. Sannan ya ƙirƙira kowane saitin fasali azaman Gaussian mai yawa, yana taƙaita su ta ma'anar vector da matrix covariance. An ƙididdige nisa tsakanin Gaussians biyu tare da nisan Fréchet (wanda ake kira nisa 2-Wasserstein). Ƙarƙashin FID yana nufin ma'anar rarraba da aka samar kuma ya bazu daidai da ainihin hotuna, yana ɗaukar aminci (suna kama da gaske?) da bambancin (suna rufe nau'ikan bayanan gaske?).
Fahimtar Fasaha
Ƙididdigar FID ita ce bambancin murabba'i na ma'ana biyu na ma'ana tare da alamar ( jimlar covariances ban da sau biyu tushen tushen matrix na samfurin su). Saboda yana amfani da cikakken haɗin kai, FID yana ladabtar da duka biyun blur, abubuwan da ba na gaskiya ba da rugujewar yanayi inda ƙirar ke samar da ɗanɗano iri-iri. Yana da kula da girman samfurin - ƙananan hotuna suna nuna son zuciya ga ƙiyasin sama - don haka masu aiki yawanci suna lissafta shi sama da dubun dubatar hotuna, galibi 50,000.
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 Nisan Faruwar Fréchet
FID ya kasance tsohowar filin, amma rauninsa shine madaidaicin hanyoyin. Masu bincike sun nuna cewa ya gaji ImageNet son rai daga Inception-v3 kuma zai iya saba wa hukuncin ɗan adam, yana haifar da ma'auni kamar FID da aka lissafta akan fasalulluka na CLIP (wani lokacin da ake kira FDD ko CMMD), Kernel Inception Distance (KID) don ƙananan samfurori, da ma'auni mai mahimmanci / tunawa da ke raba aminci daga bambancin. Yi tsammanin mafi arziƙi, fasalin-kashin baya-agnostic da tsinkaye mai daidaituwa, musamman kamar yadda rubutu-zuwa-hoto da tsarar bidiyo suka fi girma takaitattun lamba ɗaya.
Aiwatar da Gaskiyar Duniya
Ƙididdigar GANs kamar StyleGAN, inda ƙungiyoyi ke ba da rahoton FID akan ma'ajin bayanai kamar FFHQ don kwatanta ingancin haɓakar fuska.
Bibiyar ci gaban horon samfurin yaduwa ta hanyar lissafin FID a wuraren bincike don ganin lokacin da ingancin hoto ya daina inganta.
Kwatanta ƙirar rubutu-zuwa-hoto masu fafatawa akan tsarin bayanan COCO, inda aka ambata ƙananan FID a matsayin shaidar ƙarin abubuwan da suka dace.
Gano yanayin rugujewa a cikin janareta, tun lokacin da FID's covariance term ya tashi lokacin da ƙirar ta haifar da ƙarancin bambancin hoto.
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 Fréchet Inception Distance 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
Ayyukan Nisa Sa hannu
Tambayoyin da ake yawan yi
What is Fréchet Inception Distance?
Fréchet Inception Distance (FID) shine ma'auni na ma'auni don yin la'akari da yadda ainihin saƙon da aka ƙirƙira ya kasance. Yana kwatanta ƙididdiga na ainihin hotuna da aka ƙirƙira a cikin sararin fasali mai zurfi - ƙananan maki yana nufin fakes suna kallon kusa da ainihin abu.
Menene ƙananan maki FID ke nunawa?
FID yana auna nisa tsakanin rarrabuwar siffa ta gaske da wacce aka samar, don haka ƙaramin ƙima yana nufin saitin da aka ƙirƙira ya fi dacewa da ainihin bayanai cikin aminci da bambancin.
Wace cibiyar sadarwa da aka riga aka horar da ita daidaitattun FID ke amfani da ita don cire fasalin hoto?
FID tana ɗaukar hotuna ta hanyar hanyar sadarwa ta Inception-v3 da aka riga aka horar kuma tana amfani da fasalulluka-2048-dimensional pooling-Layer don duka hotuna da aka ƙirƙira.
Ta yaya FID ke taƙaita kowane saitin fasali na hoto kafin kwatanta su?
Kowane saitin fasali an tsara shi azaman Gaussian mai yawa, kuma FID yana ƙididdige nisan Fréchet (2-Wasserstein) tsakanin Gaussians biyu.
Wane mahimmin gazawar Makin Farko ya yi adireshin FID?
Makin Ƙaddamarwa kawai ya kimanta hotunan da aka ƙirƙira a keɓe; FID ta inganta akansa ta hanyar kwatanta rarrabuwar hotuna da aka ƙirƙira kai tsaye.
Me yasa FID ke tashi lokacin da janareta ke fama da rugujewar yanayi?
FID tana amfani da cikakkiyar haɗin kai na fasali, don haka lokacin da samfurin ya samar da ɗanɗano iri-iri yaɗuwar sa ya bambanta daga ainihin bayanai, yana tura maki sama.