Juyin Juyin Halitta
Ƙunƙarar rikice-rikice na barin cibiyar sadarwa ta jijiyoyi ta lanƙwasa grid ɗin samfurin don bin ainihin siffar abubuwa maimakon tilasta ta ta tagar murabba'i mai tsauri.
Dubawa
This makes models far better at handling odd shapes, scale changes, and geometric distortion.
Zurfafa nutsewa
Samfuran juzu'i na yau da kullun yana yin pixels a ƙayyadaddun kashewa - tsayayyen grid 3x3 wanda ya tsaya akan kowane wuri. Wannan yana aiki da kyau don laushi amma yana gwagwarmaya lokacin da abubuwa suka karkata, shimfiɗa, ko siffa mara kyau. Abubuwan da ba a iya gyarawa, waɗanda Dai da abokan aiki suka gabatar a Microsoft Bincike a cikin 2017, ƙara ƙaramin koyo ga kowane ɗayan waɗannan wuraren samarwa. Cibiyar sadarwa tana duba shigarwar kuma tana tsinkayar canjin 2D ga kowane matsayi na grid, don haka filin karɓa zai iya jujjuyawa don rungumar gefuna mai lanƙwasa ko bin wata gaɓa mai lankwasa. Rukunin RoI mai lalacewa yana amfani da ra'ayi iri ɗaya ga fasalin yanki. Siffar 2 (2018) ta ƙara ma'aunin daidaitawa na kowane-aya, yana barin Layer ya datse ko haɓaka kowane samfurin, wanda ya ƙayyadad da daidaiton gano abu akan alamomi kamar COCO.
Fahimtar Fasaha
Ana samar da ɓangarorin ta hanyar ƙarin juzu'in juzu'in da ke gudana a layi daya, yana fitar da ƙimar 2N don kernel N-point (dx ɗaya, dy ɗaya a kowane aya). Saboda abubuwan da aka annabta sun kasance juzu'i, ana ƙididdige ƙimar ƙimar pixel da aka ƙididdige su tare da interpolation na bilinear, wanda ke sa gabaɗayan aikin ya bambanta. Ana koyan kashe-kashe-ƙarshe-zuwa-ƙarshe ta hanyar yaɗa baya-bayan nan na yau da kullun-babu wata kulawa ta dabam da ke gaya wa cibiyar sadarwa inda ya kamata. Ƙarin farashin yana da ƙanƙanta saboda reshen da aka kashe yana da nauyi dangane da babban taswirorin fasali.
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 Juyin Juyin Halitta
Hankali mara kyau ya zama ƙashin bayan gano zamani: Deformable DETR yana amfani da ƙwaƙƙwaran ƙira don sanya hankalin taranfofi ya yi rauni da sauri, yana yanke lokacin horo sosai tare da ainihin DETR. Yi tsammanin ƙa'idar da za ta ci gaba da yaɗuwa cikin bidiyo, gizagizai na maki 3D, da ƙirar harshe-hangen nesa, inda samfurin daidaitawa yana taimakawa wajen sarrafa motsi, ɓoyewa, da lissafi mara daidaituwa. Kamar yadda goyon bayan hardware don samun damar ƙwaƙwalwar ajiya ba bisa ka'ida ba ya inganta, masu aikin nakasa ya kamata su sami rahusa kuma ana tura su gabaɗaya akan na'urori.
Aiwatar da Gaskiyar Duniya
Gano abu akan COCO, inda yadudduka masu lalacewa suna haɓaka daidaito akan abubuwa masu tsayi ko jujjuya kamar jiragen ƙasa da raƙuman ruwa.
Rarraba yanayin yanayin titi, masu taimaka wa samfura don gano alamomin layi mai lanƙwasa da ƙayyadaddun gine-gine marasa tsari.
Deformable DETR don gano ƙarshen-zuwa-ƙarshe, ta amfani da ɓangarorin da aka koya don sanya kulawar taswira mai inganci
Hoton likita, inda ciwace-ciwacen daji da gabobin ke da sifofi marasa ƙarfi waɗanda ƙayyadaddun grid ɗin ke kama da kyau
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 Deformable Convolutions 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
Juyin Juyin Halitta Mai Zurfi
Tambayoyin da ake yawan yi
What is Deformable Convolutions?
Ƙunƙarar rikice-rikice na barin cibiyar sadarwa ta jijiyoyi ta lanƙwasa grid ɗin samfurin don bin ainihin siffar abubuwa maimakon tilasta ta ta tagar murabba'i mai tsauri. Wannan yana sa ƙirar ta fi kyau wajen sarrafa sifar da ba ta dace ba, sauye-sauyen sikeli, da hargitsin lissafi.
Menene juzu'in nakasassu ke ƙara idan aka kwatanta da daidaitaccen juyi?
Juzu'i masu lalacewa suna tsinkaya koma baya da aka koya (dx, dy) ga kowane wurin samfurin, yana barin grid ɗin ya yi daidai da sifofin abu.
Ta yaya ake samar da ɓangarorin ƙima a cikin juzu'i mai lalacewa?
Wani reshe mai kama da juzu'i yana fitar da lada, waɗanda aka koya daga ƙarshe zuwa ƙarshe ta hanyar yaɗa baya.
Saboda abubuwan da aka annabta yawanci juzu'i ne, ta yaya ake ƙididdige ƙimar pixel a waɗannan wurare?
Matsakaicin juzu'i yana buƙatar haɗakarwar bilinear, wanda ke sa aikin ya bambanta don horo.
Menene Deformable ConvNets v2 (2018) ya ƙara akan asali?
v2 ya gabatar da juzu'i, ma'aunin nauyi da aka koya kowane wurin samfur wanda zai iya haɓaka ko murkushe tasirinsa, inganta daidaito.
Me yasa rikicewar rikice-rikice ke taimakawa musamman ga abubuwa marasa siffa?
Ta hanyar lanƙwasa grid ɗin samfur, ingantaccen filin karɓa yana daidaitawa da jumlolin abu maimakon ƙaƙƙarfan murabba'i.