Masu bincike na Microsoft sun ƙirƙiri tsarin leƙen asiri na wucin gadi wanda zai iya samar da bayanan hoto waɗanda, a lokuta da yawa, sun zama mafi inganci fiye da kwatancin da mutane suka yi. Wannan ci gaban ya nuna babban ci gaba a yunƙurin Microsoft na mai da samfuransa da aiyukan sa ga duk masu amfani.
"Bayyana hoto yana daya daga cikin manyan ayyuka na hangen nesa na kwamfuta, wanda ke ba da dama ga ayyuka masu yawa," in ji Xuedong Huang (), Jami'in Fasaha na Microsoft da CTO na Azure AI Cognitive Services a Redmond, Washington.
Sabon samfurin yanzu yana samuwa ga masu amfani ta hanyar Computer Vision a , wanda shine ɓangare na Azure AI, kuma yana ba masu haɓaka damar amfani da wannan fasalin don inganta wadatar ayyukan su. Ana kuma haɗa shi a cikin manhajar Seeing AI kuma za a samu nan gaba a wannan shekara a cikin Microsoft Word da Outlook na Windows da Mac, da kuma PowerPoint don Windows, Mac da kuma kan yanar gizo.
Bayanin atomatik yana taimaka wa masu amfani samun damar samun mahimman abun ciki na kowane hoto, ko hoto ne da aka dawo a sakamakon bincike ko hoto don gabatarwa.
"Amfani da rubutun da ke bayyana abubuwan da ke cikin hotuna (wanda ake kira madadin ko madadin rubutu) a kan shafukan yanar gizo da takardu yana da mahimmanci ga makafi ko masu nakasa," in ji Saqib Sheikh (), Manajan Software a Microsoft's AI Platform Group a Redmond.
Misali, tawagarsa tana amfani da ingantaccen fasalin bayanin hoto a cikin app don makafi da nakasassu. , wanda ke gane abin da kyamarar ke ɗauka kuma ta faɗi game da shi. Ka'idar tana amfani da rubutun da aka ƙirƙira don bayyana hotuna, gami da kan kafofin watsa labarun.
"Mai kyau, kowa ya kamata ya ƙara alt rubutu zuwa duk hotuna a cikin takardu, a yanar gizo, a shafukan sada zumunta, saboda wannan yana bawa makafi damar shiga abubuwan da ke ciki kuma su shiga cikin tattaunawa. Amma kash, mutane ba sa yin haka,” in ji Sheikh. "Duk da haka, akwai wasu ƙa'idodi waɗanda ke amfani da fasalin bayanin hoton don ƙara madadin rubutu lokacin da ya ɓace."
Liruan Wang, babban manajan bincike na Microsoft's Redmond Lab, ya jagoranci tawagar binciken da ta cimma kuma ta zarce sakamakon dan Adam. Hoto: Dan DeLong.
Bayanin sabbin abubuwa
"Bayyana hotuna yana daya daga cikin manyan ayyuka na hangen nesa na kwamfuta, wanda ke buƙatar tsarin fasaha na wucin gadi don fahimta da kuma kwatanta babban abun ciki ko aikin da aka gabatar a cikin hoton," in ji Liruan Wang (), babban manajan bincike a dakin binciken Redmond na Microsoft.
"Kuna buƙatar fahimtar abin da ke faruwa, gano menene alaƙar da ke tsakanin abubuwa da ayyuka, sannan ku taƙaita kuma ku kwatanta shi duka a cikin jumla cikin harshen da mutum zai iya karantawa," in ji ta.
Wang ya jagoranci tawagar bincike, wanda a cikin benchmarking (rubutun labari na labari a ma'auni, babban bayanin sabbin abubuwa) ya sami sakamako mai kama da na mutum, kuma ya zarce shi. Wannan gwajin yana ba ku damar kimanta yadda tsarin AI ke samar da kwatancen abubuwan da aka kwatanta waɗanda ba a haɗa su cikin saitin bayanan da aka horar da ƙirar a kansu.
Yawanci, ana horar da tsarin bayanin hoto akan saitin bayanai waɗanda ke ɗauke da hotuna tare da bayanin rubutu na waɗannan hotuna, wato, akan saitin hotunan da aka sa hannu.
"Gwajin nocaps ya nuna yadda tsarin zai iya kwatanta sababbin abubuwan da ba a samo su a cikin bayanan horo ba," in ji Wang.
Don magance wannan matsala, ƙungiyar Microsoft ta riga ta horar da babban samfurin AI akan babban ma'aunin bayanai mai ɗauke da hotuna masu alamar kalma, kowannensu yana da alaƙa da wani takamaiman abu a cikin hoton.
Saitin hotuna tare da alamun kalmomi maimakon cikakkun bayanai sun fi dacewa don ƙirƙira, yana bawa ƙungiyar Wang damar ciyar da bayanai da yawa a cikin ƙirar su. Wannan hanya ta ba da samfurin abin da ƙungiyar ta kira ƙamus na gani.
Kamar yadda Huang ya bayyana, hanyar koyo kafin koyo ta hanyar amfani da ƙamus na gani yana kama da shirya yara don karantawa: na farko, ana amfani da littafin hoto wanda ake danganta kalmomi ɗaya da hotuna, misali, ƙarƙashin hoton apple an rubuta "apple" kuma a ƙarƙashin hoton cat akwai kalmar "cat".
“Wannan horo na farko tare da ƙamus na gani shine ainihin ilimin farko da ake buƙata don horar da tsarin. Wannan shine yadda muke ƙoƙarin haɓaka nau'in ƙwaƙwalwar ajiyar mota, "in ji Huang.
Sa'an nan kuma ana tace samfurin da aka riga aka horar tare da saitin bayanai gami da hotuna masu lakabi. A wannan mataki na horo, samfurin ya koyi yin jimloli. Idan hoton da ke ɗauke da sababbin abubuwa ya bayyana, tsarin AI yana amfani da ƙamus na gani don ƙirƙirar cikakkun bayanai.
"Don yin aiki tare da sababbin abubuwa yayin gwaji, tsarin yana haɗa abubuwan da ya koya yayin horo da kuma lokacin gyare-gyare na gaba," in ji Wang.
A cewar sakamakon , Lokacin da aka kimanta akan gwaje-gwajen nocaps, tsarin AI ya samar da ƙarin ma'ana da cikakkun bayanai fiye da yadda mutane suka yi don hotuna iri ɗaya.
Saurin sauyawa zuwa yanayin aiki
Daga cikin wasu abubuwa, sabon tsarin bayanin hoto ya ninka samfurin da aka yi amfani da shi a cikin samfuran Microsoft da sabis tun daga 2015, idan aka kwatanta da wani ma'auni na masana'antu.
La'akari da fa'idodin da duk masu amfani da samfura da sabis na Microsoft za su samu daga wannan haɓakawa, Huang ya haɓaka haɗa sabon ƙirar zuwa yanayin aikin Azure.
"Muna ɗaukar wannan fasaha ta AI mai ruguzawa zuwa Azure a matsayin dandamali don hidimar abokan ciniki da yawa," in ji shi. “Kuma wannan ba ci gaba ba ne kawai a cikin bincike. Lokacin da aka ɗauka don haɗa wannan ci gaba a cikin yanayin samar da Azure shi ma wani ci gaba ne. "
Huang ya kara da cewa samun sakamako irin na dan Adam na ci gaba da kasancewa a cikin tsarin bayanan sirri na Microsoft.
"A cikin shekaru biyar da suka gabata, mun sami sakamako mai kama da mutum a manyan fannoni biyar: a cikin fahimtar magana, a cikin fassarar injin, a cikin amsa tambayoyi, a cikin karatun injin da fahimtar rubutu, kuma a cikin 2020, duk da COVID-19, a cikin bayanin hoto. ' Juan ya ce.
Ta hanyar magana
- Ziyarci shafin don ƙarin koyo game da hangen nesa na kwamfuta ()
- Karanta labarin
Kwatanta sakamakon bayanin hotunan da tsarin ya bayar a baya da kuma yanzu ta amfani da AI
Hakkin mallakar hoto Getty Images. Bayanin da ya gabata: Kusa da wani mutum yana shirya kare mai zafi a kan katako. Sabon bayanin: Mutum yana yin burodi.
Hakkin mallakar hoto Getty Images. Bayanin da ya gabata: Wani mutum yana zaune a faɗuwar rana. Sabon bayanin: Wuta a bakin teku.
Hakkin mallakar hoto Getty Images. Bayanin da ya gabata: Wani mutum a cikin riga mai shuɗi. Sabon bayanin: Mutane da yawa sanye da abin rufe fuska na tiyata.
Hakkin mallakar hoto Getty Images. Bayanin da ya gabata: Wani mutum a kan skateboard ya tashi sama da bango. Sabon bayanin: Dan wasan baseball ya kama kwallo.
source: www.habr.com