Mfu nghọta na LPIPS
Ọnwụ nghọta na-atụ ka onyonyo abụọ yiri nke ahụ si ele mmadụ anya site n'ịtụle njirimara netwọkụ akwara dị omimi kama ịbụ pikselụ.
Nchịkọta
It matters because pixel-by-pixel comparison wrongly punishes tiny shifts and blurs detail, while perceptual loss rewards sharp, realistic results.
Ime miri emi
Ọnwụ ọdịnala dị ka L2 ( pụtara squared njehie) tụnyere ihe oyiyi pixel-by-pixel, yabụ ngbanwe otu pikselụ ma ọ bụ udidi dịtụ iche dị ka nnukwu njehie n'agbanyeghị na ụmụ mmadụ anaghị achọpụta. Ọnwụ nhụta kama na-eme onyonyo abụọ ahụ site na netwọk a zụrụ azụ (na-abụkarị VGG) ma na-atụnyere mmemme sitere na ọkwa etiti. N'ihi na njirimara ndị ahụ na-edobe akụkụ, textures na akụkụ ihe karịa kpọmkwem ụkpụrụ pikselụ, mfu ahụ dabara nke ọma na mkpebi mmadụ, na-agba ume dị nkọ, nke kwesị ntụkwasị obi. LPIPS (Myiri Nhụta Amụma Amụtara), nke Zhang et al webatara. na 2018, formalizes nke a: ọ extracts miri atụmatụ, normalizes ha, na-etinye amụtara arọ calibrated megide ọtụtụ puku mmadụ myirịta ikpe, na-emepụta otu anya akara ebe ala pụtara ọzọ nghọta n'otu n'otu.
Nghọta nka nka
LPIPS na-agafe onyonyo abụọ ahụ site na ọkpụkpụ azụ (VGG, AlexNet, ma ọ bụ SqueezeNet), otu-na-emezigharị ọrụ ọwa n'ọtụtụ ọkwa, wee were ọdịiche dị squared n'ọnọdụ ọ bụla. Obere ihe ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ̀tụ̀tụ̀ ọ bụla a mụtara na-atụle ndịiche ndị ahụ tupu a chịkọta ya n'ụzọ zuru ezu ma chịkọta ya n'ofe. A zụrụ ihe ọ̀tụ̀tụ̀ ndị ahụ na dataset BAPPS nke mmadụ abụọ-ọzọ mmanye-nhọpụta mmadụ, ya mere metrik ahụ na-egosipụta ihe ndị mmadụ na-aghọta n'ezie kama ịdị anya njirimara.
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 mfu nghọta na LPIPS
Metiriki nghọta na-agbanwe site na azụ azụ CNN gaa na njirimara sitere na ụdị onye na-elekọta onwe ya yana ihe ntụgharị ọhụụ dị ka DINO na CLIP, nke na-ejide usoro ọmụmụ bara ụba. Na-atụ anya mwekota siri ike site na nkuzi ụdị mgbasa ozi yana nlebanya ederede gaa na onyonyo, gbakwunyere akara nghọta na-ege ntị maka ngbanwe oge vidiyo. Ndị na-eme nchọpụta na-enyochakwa ntụpọ kpuru ìsì LPIPS: enwere ike ịghọgbu ya na mmegide na adịghị ike na-ejikọta ya na ịdịmma na ntụkwasị obi dị elu, na-akpali metrics ndị mmadụ na-ejikọta ọhụrụ dị ka DISTS na usoro nchịkọta.
Mmejuputa n'ezie n'ụwa
Ọzụzụ netwọkụ nwere mkpebi siri ike (dịka, SRGAN) foto ndị gbagoro agbago na-adị nkọ ma na-edobe anya karịa ka ọ dị nkọ.
Na-enyocha mkpakọ onyonyo na codecs site n'ịtụle ka nlebara anya na-emechi onyonyo emepụtara na nke izizi.
Nbufe ụdị na-eduzi, ebe a na-ejikọta ọdịnaya site na njirimara VGG dị omimi kama ịbụ pikselụ.
Benchmarking GAN na ndị na-emepụta ihe onyonyo na-ekesa site n'ịkọwa anya LPIPS n'etiti onyonyo emepụtara na nke adịchaghị.
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 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
Ntuziaka na-esote
Mfu n'uche maka nchọpụta na-ezighi ezi
Ajụjụ a na-ajụkarị
What is Perceptual Loss and LPIPS?
Ọnwụ nghọta na-atụ ka onyonyo abụọ yiri nke ahụ si ele mmadụ anya site n'ịtụle njirimara netwọkụ akwara dị omimi kama ịbụ pikselụ. Ọ dị mkpa n'ihi na ntụnyere pikselụ-site-pixel na-ata ahụhụ n'ụzọ na-ezighi ezi na-ata obere mgbanwe na nhụsianya nkọwa, ebe mfu nghọta na-akwụghachi ụgwọ dị nkọ, nsonaazụ ezi uche dị na ya.
Kedu ihe kpatara mfu L2 dabere na pixel na-echekarị myirịta onyonyo ma e jiri ya tụnyere enweghị nghọta?
L2 na-atụnyere pikselụ ozugbo, yabụ obere mgbanwe oghere ma ọ bụ ọdịiche udidi edebanye aha dị ka nnukwu njehie ọbụlagodi mgbe onyonyo mara mma n'anya mmadụ.
Kedu ihe LPIPS ji atụnyere n'etiti onyonyo abụọ?
LPIPS na-ewepụta ọrụ dị omimi site na ọkpụkpụ azụ a kapịrị ọnụ wee tụọ anya ha, nke dabara na nghọta mmadụ karịa pikselụ.
Kedu otu esi ekpebi oke ọwa ọ bụla na LPIPS?
Amụtara ihe ọ̀tụ̀tụ̀ ahụ ka ọ dakọọ na nnukwu dataset (BAPPS) nke mkpebi myirịta mmadụ abụọ ọzọ-mmanye-mmanye.
Kedu netwọkụ azụ ka a na-ejikọkarị na mfu nghọta (atụmatụ)?
Maapụ atụmatụ etiti VGG ghọrọ ọkọlọtọ maka mfu nhụsianya na ọrụ dị ka nnukwu mkpebi na mbufe ụdị.
Kedu ọrụ na-erite uru ozugbo site na iji mfu nghọta kama ịbụ L2 dị ọcha?
Netwọk nwere mkpebi siri ike zụrụ azụ na nhụta nghọta na-arụpụta textures dị nkọ karịa, ebe L2 na-achọ iwepụta ọnụ ọgụgụ na-adịghị mma.