Ngaba ukhe wachukunyiswa kukukwazi kwekhamera ye-smartphone yakho ukubona ubuso kwifoto yeqela?
Mhlawumbi uye wothuswa yindlela iimoto eziziqhubayo ezithi zihambe ngaphandle komthungo kwitrafikhi, zichonga abahambi ngeenyawo nezinye iimoto ezichane ngendlela emangalisayo.
Ezi mpu melelo zibonakala zingaphaya kwendalo zenziwa ukuba zenzeke ngokubhaqwa kwezinto, umxholo onika umdla wophando. Ukutsho nje, ukuchongwa kwento kukuchonga kunye nokubekwa kwezinto ngaphakathi kwemifanekiso okanye iividiyo.
Bubuchwephesha obuvumela iikhompyutha ukuba "zibone" kwaye ziqonde ihlabathi elizingqongileyo.
Kodwa le nkqubo ingakholelekiyo isebenza njani? Siyayibona loo nto ukufunda okunzulu kuye iguqule indawo yokuchongwa kwento. Ivula indlela yothotho lwezicelo ezinempembelelo ngqo kubomi bethu bemihla ngemihla.
Kwesi sithuba, siya kutyhutyha ummandla onomdla wokuchongwa kwento enzulu esekelwe ekufundeni, sifunde ukuba inamandla okulungisa indlela esisebenzisana ngayo neteknoloji.
Yintoni kanye kanye ukufunyanwa kwento?
Enye yezona umbono osisiseko wekhompyuter imisebenzi kukubona into, ebandakanya ukufumana kunye nokubeka izinto ezahlukeneyo kumfanekiso okanye ividiyo.
Xa kuthelekiswa nokuhlelwa komfanekiso, apho ileyibhile yodidi lwento nganye imiselwe, ubhaqo lwento luhamba inyathelo elinye phambili ngokuchonga kuphela ubukho bento nganye kodwa nokuzoba iibhokisi zokubopha kwindawo nganye.
Ngenxa yoko, sinokubona ngaxeshanye iintlobo zezinto ezinomdla kwaye sizibeke ngokuchanekileyo.
Ukukwazi ukubona izinto kubalulekile kwizicelo ezininzi, kubandakanya ukuqhuba ngokuzimela, ukubek' esweni, ukubonwa kobuso, kunye nemifanekiso yezonyango.
Ukujongana nalo mceli mngeni unzima ngokuchaneka okubalaseleyo kunye nokusebenza kwexesha lokwenyani, ubuchule obunzulu bokufunda obusekwe kuguqule ukufunyanwa kwezinto.
Ukufunda okunzulu kutshanje kuye kwavela njengesicwangciso esinamandla sokunqoba obu bunzima, ukutshintsha imboni yento yokuqaphela.
Usapho lwe-R-CNN kunye ne YOLO usapho ziintsapho ezimbini ezaziwayo eziyimodeli ekuchongeni into eza kuphononongwa kweli nqaku.
I-R-CNN Family: Ukufunyanwa kweNto yobuvulindlela
Uphando lokuqondwa kwezinto kwangethuba lubone inkqubela phambili emandla kusapho lwe-R-CNN, olubandakanya i-R-CNN, i-R-CNN ekhawulezayo, kunye ne-Faster R-CNN.
Ngoyilo lweemodyuli ezintathu, imimandla ecetywayo ye-R-CNN isebenzise i-CNN ukukhupha iimpawu, kunye nezinto ezihleliweyo zisebenzisa i-SVMs.
I-R-CNN yayichanekile, nangona kuthathe ixesha kuba iibhidi zengingqi zabagqatswa zazifuneka. Oku kwajongwana ne-Fast R-CNN, eyandisa ukusebenza kakuhle ngokudibanisa zonke iimodyuli zibe yimodeli enye.
Ngokongeza iNethiwekhi yeSindululo soMmandla (RPN) edale kwaye yaphucula izindululo zengingqi ngexesha loqeqesho, iR-CNN ekhawulezayo iphucule kakhulu ukusebenza kwaye yafumana ukuqondwa kwento yexesha lokwenyani.
Ukusuka kwi-R-CNN ukuya kwi-Faster R-CNN
Usapho lwe-R-CNN, olumele “Isekwe kwiNgingqi Convolutional Neural Networks,” iqhubele phambili ekubhaqweni kwento.
Le ntsapho iquka i-R-CNN, i-Fast R-CNN, kunye ne-Faster R-CNN, zonke ezilungiselelwe ukujongana nezinto zasekuhlaleni kunye nemisebenzi yokuqaphela.
I-R-CNN yasekuqaleni, eyaziswa kwi-2014, ibonise ukusetyenziswa okuyimpumelelo kwe-convolutional networks ze-neural zokubona into kunye nokufakwa kwendawo.
Kwathatha isicwangciso samanyathelo amathathu abandakanya isiphakamiso sommandla, i-extraction ye-CNN, kunye nokuhlelwa kwezinto kunye nabahluli be-Vector Machine (SVM).
Ukulandela ukuqaliswa kwe-Fast R-CNN kwi-2015, iingxaki zesantya zaye zasonjululwa ngokudibanisa isiphakamiso sommandla kunye nokuhlelwa kwimodeli enye, ukunciphisa ngokukhawuleza uqeqesho kunye nexesha le-inference.
Ukukhawuleza i-R-CNN, ekhutshwe kwi-2016, isantya esiphuculweyo kunye nokuchaneka ngokubandakanya i-Regional Proposal Network (RPN) ngexesha loqeqesho ukuphakamisa ngokukhawuleza nokuhlaziya iindawo.
Ngenxa yoko, i-Faster R-CNN iye yaziseka njengenye ye-algorithms ehamba phambili kwimisebenzi yokufumanisa into.
Ukudityaniswa kwabadidiyeli be-SVM kwakubalulekile kwimpumelelo yosapho lwe-R-CNN, ukutshintsha indawo yombono wekhompyuter kunye nokubeka indlela yempumelelo yexesha elizayo ekubhaqweni okunzulu kwento esekwe kufundo.
Amandla:
- Ukuchaneka kwento yokufumana indawo ephezulu.
- Ukuchaneka kunye nokusebenza kakuhle kuhambelana noyilo oludibeneyo lwe-R-CNN ekhawulezayo.
Ubuthathaka:
- Ukungeniswa nge-R-CNN kunye ne-R-CNN ekhawulezayo kunokuba nzima kakhulu.
- Ukuze i-R-CNN ekhawulezileyo isebenze ngokusemandleni ayo, uninzi lwezindululo zengingqi zisenokuba yimfuneko.
Intsapho ye-YOLO: Ukufunyanwa kweNto ngexesha langempela
Usapho lwe-YOLO, olusekwe kwingqikelelo ethi "Ujonge Kanye Kanye" igxininisa ukuqondwa kwezinto ngexesha langempela ngelixa uncama ukuchaneka.
Imodeli yoqobo ye-YOLO ibiquka inethiwekhi ye-neural enye eqikelele ngokuthe ngqo iibhokisi zokubopha kunye neelebhile zeklasi.
Ngaphandle kokuchaneka okuncinci koqikelelo, i-YOLO inokusebenza ngesantya ukuya kuthi ga kwi-155 yefreyimu ngomzuzwana. I-YOLOv2, ekwabizwa ngokuba yi-YOLO9000, yajongana nezinye zeentsilelo zemodeli yoqobo ngokuqikelela iiklasi zezinto ezingama-9,000 kunye nokubandakanya iibhokisi zeankile zoqikelelo oluluqilima ngakumbi.
I-YOLOv3 iphuculwe ngakumbi, kunye nenethiwekhi yesixhobo esibanzi ngakumbi.
Ukusebenza kwangaphakathi kweYOLO yoSapho
Iimodeli zokuchongwa kwento kwiYOLO (Ujonge Kanye Kanye) usapho luvele njengempumelelo eqaphelekayo kumbono wekhompyuter.
I-YOLO, eyaziswa kwi-2015, ibeka phambili isantya kunye nexesha langempela lokuchongwa kwezinto ngokulindela ngokuthe ngqo iibhokisi ezibophezelayo kunye neelebhile zeklasi.
Nangona ukuchaneka okuthile kubingelelwe, ihlalutya iifoto ngexesha lokwenyani, iyenza ibe luncedo kwizicelo ezibalulekileyo zexesha.
I-YOLOv2 ifake iibhokisi ze-ankile zokujongana nezikali zezinto ezahlukeneyo kwaye ziqeqeshwe kwiiseti zedatha ezininzi ukulindela ngaphezu kweeklasi zezinto ezingama-9,000.
Ngo-2018, i-YOLOv3 yomeleza usapho ngakumbi ngenethiwekhi yesixhobo esinzulu, iphucula ukuchaneka ngaphandle kokuncama ukusebenza.
Usapho lwe-YOLO luqikelela iibhokisi ezibophezelayo, izinto ezinokwenzeka eklasini, kunye namanqaku aphikisanayo ngokwahlula umfanekiso ube yigridi. Idibanisa ngokufanelekileyo isantya kunye nokuchaneka, iyenza ilungele ukusetyenziswa izithuthi ezizimeleyo, ukucupha, ukhathalelo lwempilo, kunye nezinye iinkalo.
Uchungechunge lwe-YOLO luguqule ukuchongwa kwezinto ngokubonelela ngezisombululo zexesha langempela ngaphandle kokuncama ukuchaneka okubalulekileyo.
Ukusuka kwi-YOLO ukuya kwi-YOLOv2 kunye ne-YOLOv3, olu sapho lwenze inkqubela phambili enkulu ekuphuculeni ukuqondwa kwezinto kuwo wonke amashishini, ukuseka umgangatho weenkqubo zale mihla zokufunda ezisekelwe kwizinto ezinzulu.
Amandla:
- Ukubona izinto ngexesha lokwenyani kumazinga aphezulu esakhelo.
- Ukuzinza kwiingqikelelo zebhokisi ezibophezelayo kwaziswa kwi-YOLOv2 kunye ne-YOLOv3.
Ubuthathaka:
- Iimodeli ze-YOLO zinokuncama ukuchaneka okuthile ngokutshintshiselana ngesantya.
Ufaniso lweNtsapho engumzekelo: Ukuchaneka vs
Xa i-R-CNN kunye neentsapho ze-YOLO zifaniswa, kuyacaca ukuba ukuchaneka kunye nokusebenza kakuhle zizinto ezibalulekileyo zokurhweba. Iimodeli zosapho ze-R-CNN zigqwesa ngokuchaneka kodwa zicotha ngexesha lokucinga ngenxa yoyilo lweemodyuli ezintathu.
Intsapho ye-YOLO, ngakolunye uhlangothi, ibeka phambili ukusebenza kwexesha langempela, ukubonelela ngesantya esibalaseleyo ngelixa ulahlekelwa ngokuchanekileyo. Isigqibo phakathi kwezi ntsapho zemodeli sigqitywa ziimfuno ezithile zesicelo.
Iimodeli zosapho lwe-R-CNN zinokukhethwa kumthwalo womsebenzi ofuna ukuchaneka okugqithisileyo, ngelixa iimodeli zosapho lwe-YOLO zilungele usetyenziso lwexesha lokwenyani.
Ngaphaya kweNgcaciso yeNgcaciso: Ii-Reality-World Applications
Ngaphaya kwemisebenzi yokuqondwa kwezinto ezisemgangathweni, ukufunyanwa kwento enzulu esekelwe ekufundeni kufumene uluhlu olubanzi losetyenziso.
Ukuguquguquka kwayo kunye nokuchaneka kwayo kudale amathuba amatsha kumacandelo ahlukeneyo, ukulungisa imingeni enzima kunye nokuguqula amashishini.
Izithuthi eziZimelayo: Ukumisela uMgangatho wokuQhuba ngokuKhuselekileyo
Ukufunyaniswa kwento kubalulekile kwiimoto ezizimeleyo zokuqinisekisa ukuhamba ngokukhuselekileyo nokuthembekileyo.
Iimodeli zokufunda ezinzulu ukubonelela ngolwazi olubalulekileyo lweenkqubo zokuqhuba ezizimeleyo ngokuqaphela kunye nokwazisa abahambi ngeenyawo, abakhweli beebhayisikile, ezinye iimoto, kunye neengozi zendlela ezinokuthi zibekho.
Ezi modeli zivumela izithuthi ukuba zithathe ukhetho lwexesha lokwenyani kwaye zithintele ukungqubana, zisisondeza kwikamva apho iimoto eziziqhubayo zihlala kunye nabaqhubi abangabantu.
Ukwandiswa kokuSebenza kunye noKhuseleko kuShishino loRhwebo
Ishishini lokuthengisa liye lamkela into enzulu yokufunda-esekelwe ekuboneni ukuphucula kakhulu ukusebenza kwayo.
Izixhobo zokufumanisa izinto ekuchongeni nasekulandeleni iimveliso kwiishelufu zevenkile, okuvumela ukubuyisela kwakhona okusebenzayo kunye nokunciphisa iimeko eziphumayo.
Ngaphaya koko, iinkqubo zokucupha ezinee-algorithms zokubona into zinceda ekuthinteleni ubusela kunye nokugcinwa kokhuseleko lweevenkile.
INkcazo yezoNyango kuNonophelo lwezeMpilo
Ukufunyanwa kwezinto ezisekelwe kufundo olunzulu kuye kwaba sisixhobo esibalulekileyo kwimifanekiso yezonyango kwicandelo lezempilo.
Inceda abaqeqeshi bezempilo ekuboneni izinto ezingaqhelekanga kwi-X-reyi, ukuskena kwe-MRI, kunye neminye imifanekiso yezonyango, efana nomhlaza okanye ukungahambi kakuhle.
Izixhobo zokuchongwa kwento ekuxilongweni kwangaphambili kunye nokucwangciswa kwonyango ngokuchonga kunye nokugqamisa iindawo ezithile ezixhalabisayo.
Ukwandiswa koKhuseleko ngoKhuseleko kunye nokuBeka iliso
Ukufunyanwa kwento kunokuba luncedo kakhulu kukhuseleko kunye nezicelo zokucupha.
Ii-algorithms zokufunda nzulu nceda abantu ababukeleyo, babone indlela yokuziphatha ekrokrisayo, kwaye babone iingozi ezinokubakho kwiindawo zikawonke-wonke, kwizikhululo zeenqwelomoya nakwiindawo zokuhamba.
Ezi nkqubo zinokulumkisa iingcali zokhuseleko ngexesha lokwenyani ngokuhlala zivavanya ukutya kwevidiyo, ukuthintela ukophulwa kokhuseleko, kunye nokuqinisekisa ukhuseleko loluntu.
Imiqobo yangoku kunye namaThemba exesha elizayo
Ngaphandle kwenkqubela phambili ebalulekileyo ekubhaqweni okunzulu okusekwe kufundo, iingxaki zihleli. Ubumfihlo bedatha yinkxalabo enkulu, njengoko ukuchongwa kwento rhoqo kubandakanya ukulawula ulwazi olubuthathaka.
Enye ingxaki ephambili kukuqinisekisa ukomelela ngokuchasene nohlaselo lweentshaba.
Abaphandi basakhangela iindlela zokwandisa imodeli ngokubanzi kunye nokutolika.
Ngophando oluqhubekayo olugxile ekuchongeni izinto ezininzi, ukulandelela into yevidiyo, kunye nexesha lokwenyani lokuqondwa kwento ye-3D, ikamva libonakala liqaqambile.
Kufuneka silindele izisombululo ezichane ngakumbi nezisebenzayo kwakamsinyane njengoko iimodeli zokufunda ezinzulu ziqhubeka nokukhula.
isiphelo
Ukufunda nzulu kuye kwaguqula ukuchongwa kwezinto, kwangenisa ixesha lokuchaneka okukhulu nokusebenza kakuhle. Iintsapho ze-R-CNN kunye ne-YOLO ziye zadlala indima ebalulekileyo, nganye inamandla ahlukeneyo kwizicelo ezithile.
Ukuchongwa kwento esekelwe kufundo olunzulu kukuguqula amacandelo kunye nokuphucula ukhuseleko nokusebenza kakuhle, ukusuka kwizithuthi ezizimeleyo ukuya kukhathalelo lwezempilo.
Ikamva lokufunyanwa kwento libonakala liqaqambile kunangaphambili njengoko uphando luqhubela phambili, ukulungisa iingxaki kunye nokuhlola iindawo ezintsha.
Sibona ukuzalwa kwexesha elitsha kumbono wekhompyuter njengoko samkela amandla okufunda okunzulu, kunye nokufumanisa into ehamba phambili.
Shiya iMpendulo