Ebe ọdịda Fréchet
Fréchet Inception Distance (FID) bụ metric ọkọlọtọ maka ikpebi etu ihe onyonyo ewepụtara si bụrụ eziokwu yana dị iche iche.
Nchịkọta
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.
Ime miri emi
FID, nke Heusel et al webatara. na 2017, doziri ntụpọ isi na akara mmalite mmalite: ọ dịghị atụnyere onyonyo ewepụtara na ezigbo data. FID na-enye ma onyonyo dị adị na nke ewepụtara site na netwọk Inception-v3 a zụrụ azụ wee gụpụta vector akụkụ akụkụ 2048 site na akwa akwa mmiri miri emi maka onyonyo ọ bụla. Ọ na-emepụta atụmatụ nke ọ bụla dị ka Gaussian multivariate, na-achịkọta ha site na vector na covariance matrix. A na-agbakọ anya dị n'etiti ndị Gauss abụọ ahụ na anya Fréchet (nke a na-akpọkwa anya 2-Wasserstein). FID dị ala pụtara ihe nkesa ewepụtara wee gbasaa nke ọma dabara na ezigbo onyonyo, na-ewepụta ma ikwesị ntụkwasị obi (ha ọ dị adị?) na ụdị dị iche iche (ha na-ekpuchi ụdị data dị iche iche?).
Nghọta nka nka
Usoro FID bụ ọdịiche dị n'akụkụ akụkụ nke vector abụọ pụtara gbakwunyere akara nke (nchikota nke covariances mwepu ugboro abụọ matrix square mgbọrọgwụ nke ngwaahịa ha). N'ihi na ọ na-eji nkwado zuru oke, FID na-ata ahụhụ ma ihe na-adịghị mma, nsonaazụ na-adịghị mma yana ndakpọ ọnọdụ ebe ihe nlereanya na-emepụta obere ụdị dị iche iche. Ọ na-enwe mmetụta maka nha nlele - onyonyo ole na ole na-eleda atụmatụ elu - yabụ ndị na-eme ya na-agbakọ ya karịa iri puku kwuru iri puku onyonyo, na-abụkarị 50,000.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Visual AI nwere ike megharịa nyocha, nchọpụta na mkpado ọrụ n'ọtụtụ.
Mee nhọrọ
Otu ndị na-emepụta ihe nwere ike imepụta echiche ngwa ngwa site na ngbanwe akwụkwọ ntuziaka ole na ole.
Team na usoro ọrụ
Ọrụ nwere ike iji onyonyo na akara vidiyo siri ike ịhazi.
Ọdịnihu nke Ebe ọdịda Fréchet
FID ka bụ ndabara nke ubi ahụ, mana adịghị ike ya na-ebu ụzọ ọzọ. Ndị nchọpụta egosila na ọ na-eketa echiche ImageNet site na Inception-v3 ma nwee ike ekwenyeghị na ikpe mmadụ, na-eme ka metrics dị ka FID gbakọrọ na njirimara CLIP (mgbe ụfọdụ a na-akpọ FDD ma ọ bụ CMMD), Kernel Inception Distance (KID) maka obere ihe nlele, na nkenke / cheta metrics na-ekewapụ ikwesị ntụkwasị obi na ụdị dị iche iche. Na-atụ anya ntule bara ụba, njirimara-ọkpụkpụ-agnostic na nghọta dabara adaba, ọkachasị dị ka ederede gaa na onyonyo yana ọgbọ vidiyo na-abawanye nchịkọta otu ọnụọgụ.
Mmejuputa n'ezie n'ụwa
Benchmarking GANs dị ka StyleGAN, ebe ndị otu na-akọ FID na datasets dị ka FFHQ iji tụnyere ịdịmma ihu-ọgbọ.
Isochi ọganihu ọzụzụ nke ụdị mgbasa ozi site na ịgbakọ FID na ebe nlele iji hụ mgbe ogo onyonyo kwụsịrị imeziwanye.
Tụnyere ụdị ederede na onyonyo na-asọrịta mpi na dataset COCO, ebe akpọrọ FID dị ala dị ka ihe akaebe nke mpụta ezi uche karịa.
Ịchọta ọnọdụ ndakpọ n'ime igwe ọkụ, ebe ọ bụ na okwu nnabata FID na-ebili mgbe ihe nlereanya ahụ na-emepụta obere ihe oyiyi dị iche iche.
Ihe ize ndụ & okporo ụzọ nche
Ikike onyonyo na nkwenye nwere ike bụrụ ihe egwu dị n'iwu ma ọ bụrụ na edoghị anya.
Ọrụ nlereanya nwere ike ịdịgasị iche n'ofe ọkụ, igwe mmadụ, na gburugburu.
Enwere ike ghara ịhụ ihe dị mma ma ọ bụrụ na enyochaghị oke ntụkwasị obi.
Map mmejuputa
Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.
Nwalee na data dabara na ọnọdụ mmepụta n'ezie.
Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.
Sochie ihe nlere anya wee megharịa ka emechara mgbanwe igwefoto ma ọ bụ dataset.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Ọrụ Anya bịanyere aka na ya
Ajụjụ a na-ajụkarị
What is Fréchet Inception Distance?
Fréchet Inception Distance (FID) bụ metric ọkọlọtọ maka ikpebi etu ihe onyonyo ewepụtara si bụrụ eziokwu yana dị iche iche. Ọ na-atụnyere ọnụ ọgụgụ nke ihe oyiyi dị adị na nke emepụtara na oghere dị omimi - akara dị ala pụtara na adịgboroja na-ele anya nso na ihe ahụ n'ezie.
Kedu ihe akara FID dị ala na-egosi?
FID na-atụ anya ebe dị n'etiti nkesa n'ezie na nke emepụtara, yabụ uru dị ala pụtara na ntọala emepụtara dabara na ezigbo data na ntụkwasị obi na ụdị dị iche iche.
Kedu netwọk a zụrụ azụ ka FID ọkọlọtọ na-eji wepụta atụmatụ onyonyo?
FID na-enyefe onyonyo site na netwọk Inception-v3 a zụrụ azụ wee jiri atụmatụ 2048-dimensional pooling-layer maka foto dị adị na nke ewepụtara.
Kedu ka FID si achịkọta atụmatụ onyonyo ọ bụla tupu ha atụnyere ha?
A na-eme atụmatụ nhazi ọ bụla dị ka Gaussian multivariate, yana FID na-agbakọ anya Fréchet (2-Wasserstein) n'etiti ndị Gauss abụọ.
Kedu mmejọ igodo mmalite nke akara mmalite ka FID lebara anya?
Akara mmalite tụlere naanị onyonyo ewepụtara na iche; FID kwalitere na ya site n'ịtụle nkesa onyonyo emepụtara na n'ezie.
Kedu ihe kpatara FID ji ebili mgbe generator na-ata ahụhụ site na ndakpọ ọnọdụ?
FID na-eji njirimara njirimara zuru oke, yabụ mgbe ihe nlereanya na-emepụta obere ụdị dị iche iche mgbasa ya na-esi na ezigbo data na-ebuli akara ahụ elu.