Igihombo Cyibanze Kumenyekanisha Kuringaniza
Igihombo cyibanze nigikorwa cyahinduwe cyo gutakaza uburemere-buremereye ingero zoroshye kuburyo detector ishobora kwibanda kubintu bikomeye, bidasanzwe.
Incamake
It solved the extreme background-versus-object imbalance that crippled one-stage object detectors.
Kwibira cyane
Mugushakisha ibintu, ishusho irashobora kuba irimo ibintu bike gusa ariko ibihumbi icumi byabakandida, hafi ya byose byoroshye. Hamwe na cross-entropie isanzwe, uyu mwuzure wibintu byoroshye byiganje kuri gradient kandi urohama ibyiza bidasanzwe. Igihombo cyibanze, cyatangijwe mu mpapuro za RetinaNet 2017 na Lin na bagenzi be kuri Facebook AI, gikosora ibi mugwiza cross-entropy kubintu (1 - p_t) ^ gamma. Iyo icyitegererezo cyashyizwe mubyizere kandi neza, p_t iri hafi ya 1, bityo ibintu bigabanuka kuri zeru kandi urugero rwiza rwatanze umusanzu. Ingero zikomeye, zidahwitse zigumana hafi-yuzuye uburemere. Hamwe na gamma hafi 2, RetinaNet yahuye cyangwa ikubita buhoro buhoro ibyiciro bibiri byihuta nka Faster R-CNN mugihe ugumye umuyoboro woroheje.
Ubushishozi
Ibyibanze byibanze gamma igenzura uburyo ingero zoroshye zahagaritswe: kuri gamma 0 igihombo cyibanze gihwanye na cross-entropy isanzwe, kandi gamma yo hejuru ikarishye kwibanda kubibazo bikomeye. Kuringaniza uburemere alfa (akenshi 0,25 kubiciro bidasanzwe) mubisanzwe bihujwe nayo. By'umwihariko ibintu bihindura imiterere ya gradients, ntabwo ari agaciro k'igihombo gusa, kubwibyo gusubiranamo mubisanzwe byibanda kuburugero rudasobanutse nta ntoki zikomeye-ubucukuzi bw'amabuye y'agaciro cyangwa gusubiramo.
Ingaruka z'Ingamba
Igiciro na bije
Ibyemezo byubwubatsi bitwara imikorere nigiciro cyimikorere kumyaka.
Ibyemezo bisobanutse
Ubuhanga bwa tekinike bufasha amakipe guhitamo umurongo ukwiye, ntabwo ari shyashya gusa.
Kugenzura ubuziranenge
Guhitamo neza bya injeniyeri bigabanya ibintu byizewe mubikorwa.
Ejo hazaza h'igihombo Cyibanze Kumenyekanisha Kuringaniza
Igihombo cyibanze cyahindutse ikintu kirenze RetinaNet, kigaragara muri disiketi nka FCOS, mubice, no murwego rurerure. Ibihinduka nkibihombo byibanze, gukwirakwiza igihombo, hamwe nigihombo cya varifocal birayinonosora kubushakashatsi bugezweho butagira ankeri kandi bushingiye kuri transformateur. Nkuko gutahura bigenda byerekanwa-byerekanwe nka DETR ikoresha ibice bibiri bihuye, kwibanda-muburyo bwo kwisubiramo bikomeza kuba igikoresho gifatika ahantu hose imirongo ikurikirana.
Gushyira mu bikorwa Isi
Gutahura ibyapa bito byumuhanda cyangwa abanyamaguru ba kure mumashusho yigenga-atwara ibinyabiziga aho pigiseli nyinshi ziri inyuma.
Kubona ibibyimba bidasanzwe cyangwa ibikomere muri scan yubuvuzi yiganjemo ingirangingo nzima.
Kugaragara inenge kumurongo wo gukora aho igice kinini cyagenzuwe gisanzwe.
Kumenya amato mato cyangwa ibinyabiziga muri satelite nini n'amashusho yo mu kirere.
Ingaruka & Kurinda
Gutezimbere igipimo kimwe gishobora guhisha intege nke za sisitemu.
Ibikorwa Remezo no kubungabunga akenshi usanga bidahabwa agaciro.
Icyuho cyumutekano no kwitegereza birashobora kwiyongera uko sisitemu igenda igorana.
Igishushanyo mbonera
Sobanura ubukererwe, ubuziranenge, nigiciro cyibiciro mbere yo kubishyira mubikorwa.
Ibipimo byerekana umutwaro ufatika hamwe namakuru yimiterere.
Gukurikirana ibikoresho kubikosa, drift, ningaruka zabakoresha.
Tegura inzira yo gusubiza ibyabaye mbere yo gupima.
Komeza Ubushakashatsi
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 Focal Loss for Imbalanced Detection 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
Ubuyobozi bukurikira
Imiyoboro ya Siamese hamwe nigihombo cya gatatu
Ibibazo bikunze kubazwa
What is Focal Loss for Imbalanced Detection?
Igihombo cyibanze nigikorwa cyahinduwe cyo gutakaza uburemere-buremereye ingero zoroshye kuburyo detector ishobora kwibanda kubintu bikomeye, bidasanzwe. Byakemuye ibyakabije bikabije-nibintu-bitaringaniza byamugaye icyiciro kimwe.
Ni ikihe kibazo cy'ibanze mu cyiciro kimwe cyo gutahura cyari igihombo cyibanze kugirango gikemuke?
Mu gutahura kwinshi, ingero zoroshye zurugero ziruta cyane ibintu bifatika kandi biganje kurwego rusanzwe rwa cross-entropy. Igihombo cyibanze kiremereye cyane, ingero zikomeye zibara byinshi.
Ni ikihe kintu cyerekana ko igihombo cyibanze kigwira kuri cross-entropy?
Igihombo cyibanze ni - (1 - p_t) ^ gamma * log (p_t). (1 - p_t) term ijambo rya gamma rigabanuka kugera kuri zeru kugirango wizere neza ibyahanuwe.
Bigenda bite kubura igihombo mugihe kwibanda kuri gamma byashyizwe kuri 0?
Kuri gamma = 0 ikintu (1 - p_t) ^ 0 kingana na 1, bityo igihombo cyibanze kigabanuka neza kuri cross-entropy.
Niki kongera gamma ikora kurugero rworoshye, rwashyizwe mubikorwa neza?
Gama yo hejuru ituma (1 - p_t) term ijambo rya gamma rigabanuka vuba kuri p_t ndende, ingero zoroshye rero zitanga umusanzu muke kubihombo na gradient.
Ni ubuhe bwubatsi bwa detector bwatangije igihombo cyibanze?
Igihombo cyibanze cyatangijwe mu mpapuro za RetinaNet ya 2017, bituma icyuma kimwe gishobora guhangana nuburyo bubiri.