Mgbagha ndị nwere ike imebi
Mgbagwoju anya nrụrụ na-ahapụ netwọkụ akwara na-ehulata grid nlele ya ka ọ soro ụdị ihe dị n'ezie kama ịmanye ya site na windo square siri ike.
Nchịkọta
This makes models far better at handling odd shapes, scale changes, and geometric distortion.
Ime miri emi
Ntụle mgbanwe mgbanwe nkịtị na-esetịpụ pikselụ n'ụdị akwụghị ụgwọ - grid 3 × 3 dabara adaba na ebe ọ bụla. Nke ahụ na-arụ ọrụ nke ọma maka textures mana ọ na-agbasi mbọ ike mgbe ihe na-ekpudo, gbatịa, ma ọ bụ nke na-adịghị mma. Mgbagha ndị nwere ike ime mgbanwe, nke Dai na ndị ọrụ ibe webatara na Microsoft Nchọpụta na 2017, tinye obere mgbaghara mmụta na nke ọ bụla n'ime ihe nlele ahụ. Netwọk ahụ na-eleba anya na ntinye wee buru amụma mgbanwe 2D maka ọnọdụ grid ọ bụla, yabụ ubi nnabata nwere ike ịmakụ akụkụ gbagọrọ agbagọ ma ọ bụ soro akụkụ ụkwụ gbajiri agbaji. Pooling RoI nwere ike ime na-emetụta otu echiche na njirimara mpaghara. Ụdị 2 (2018) agbakwunyere ihe nha nha nha nha ọ bụla, na-eme ka oyi akwa ahụ daa ma ọ bụ welie ihe nlele ọ bụla, nke na-eme ka nchọpụta ihe dị nkọ na akara akara dị ka COCO.
Nghọta nka nka
A na-emepụta ihe nkwụghachi ahụ site na oyi akwa convolution ọzọ na-agba ọsọ n'otu aka ahụ, na-ewepụta ụkpụrụ 2N maka kernel N-point (otu dx, otu dy kwa isi). N'ihi na nkwụghachi amụma ndị e buru n'amụma bụ akụkụ dị nta, a na-agbakọ ụkpụrụ pikselụ egosipụtara site na njikọ bilinear, nke na-eme ka ọrụ niile dị iche. A na-amụta ihe nkwụghachi site na njedebe ruo na njedebe site na mgbasa ozi nkịtị - ọ nweghị nlekọta dị iche na-agwa netwọk ebe ọ ga-ele anya. Ọnụ ego a gbakwunyere dị obere n'ihi na alaka ụlọ ọrụ akwụghị ụgwọ dị fechaa n'ihe metụtara maapụ ndị bụ isi.
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 mgbanwe mgbanwe
Nlebara anya nrụrụ aka abụrụla ọkpụkpụ azụ nke nchọpụta ọgbara ọhụrụ: Deformable DETR na-eji ihe nlele nlele amụtara mee ka nlebara anya na-agbanwe agbanwe na ngwa ngwa, na-ebelata oge ọzụzụ n'ụzọ dị egwu megide DETR mbụ. Na-atụ anya ụkpụrụ na-emebi emebi ka ọ na-agbasa n'ime vidiyo, igwe ojii isi ihe 3D, na ụdị asụsụ ọhụụ, ebe nlele mgbanwe na-enyere aka ijikwa mmegharị, mgbachi, na geometry na-adịghị mma. Ka nkwado ngwaike maka ịnweta ebe nchekwa oge niile na-akawanye mma, ndị na-arụ ọrụ na-enweghị ike ga-adịkwa ọnụ ala ma na-ebuga ya na ngwaọrụ ihu.
Mmejuputa n'ezie n'ụwa
Nchọpụta ihe na COCO, ebe akwa akwa nwere ike na-akwalite izi ezi na ihe elongated ma ọ bụ gbagharịrị dị ka ụgbọ oloko na giraffes.
Nkewa semantic nke ihe nkiri okporo ụzọ, na-enyere ụdị aka ịchọpụta akara okporo ụzọ gbagoro agbagọ na nkọwapụta ụlọ na-adịghị mma.
DETR na-emebi emebi maka nchọpụta ngwụcha ruo na njedebe, na-eji nkwụghachi amụtara iji mee ka nleba anya nke ntụgharị dị mma
Onyonyo ahụike, ebe etuto na akụkụ ahụ nwere ọdịdị na-esighi ike nke grid a kapịrị ọnụ na-ejide nke ọma
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 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
Ntuziaka na-esote
Mgbagha ndị nwere ike kewapụta nke omimi
Ajụjụ a na-ajụkarị
What is Deformable Convolutions?
Mgbagwoju anya nrụrụ na-ahapụ netwọkụ akwara na-ehulata grid nlele ya ka ọ soro ụdị ihe dị n'ezie kama ịmanye ya site na windo square siri ike. Nke a na-eme ka ụdịdị ka mma n'ịjikwa ọdịdị na-adịghị mma, mgbanwe ọnụ ọgụgụ, na mgbagha geometric.
Kedu ihe convolution nwere ike ịgbakwụnye ma e jiri ya tụnyere ọkọlọtọ ọkọlọtọ?
Mgbanwe mgbanwe na-ebu amụma nkwụghachi amụtara (dx, dy) maka ọnọdụ nlele ọ bụla, na-ahapụ grid ahụ ka ọ dakọọ n'ụdị ihe.
Kedu otu esi ewepụta ihe nhụpụta n'ụdị ngbanwe nwere ike ime?
Alaka convolution yiri ya na-ewepụta mgbaghara, nke a na-amụta site na njedebe ruo na njedebe site n'ịgbasa azụ.
N'ihi na nkwụghachi amụma amụma na-abụkarị akụkụ dị nta, kedu ka esi atụpụta ụkpụrụ pikselụ n'ọkwa ndị ahụ?
Ihe nkwụghachi nke irighiri ihe chọrọ interpolation bilinear, nke na-eme ka ọrụ ahụ dị iche maka ọzụzụ.
Kedu ihe Deformable ConvNets v2 (2018) gbakwunyere na mbụ?
v2 ewebata modulation, ibu mmụta n'otu ebe nlele nke nwere ike ịkwalite ma ọ bụ kwụsị mmetụta ya, na-emeziwanye izi ezi.
Kedu ihe kpatara mgbagha ndị nwere ike ime ka ọ na-enye aka karịsịa maka ihe ndị na-emebeghị ka ọ dị?
Site n'ihu grid nlele, ogige nnabata dị mma dabara na geometry ihe ahụ kama ịbụ square siri ike.