\nNov\u00fd motor DNN-inferen\u010dn\u00e9ho syst\u00e9mu.<\/strong>
\nV module dnn sa objavil nov\u00fd motor, ktor\u00fd koexistuje so star\u00fdm. Lep\u0161ie pracuje s dynamick\u00fdmi formami tenzorov, podgrafmi a modern\u00fdmi mo\u017enos\u0165ami ONNX. Vyhl\u00e1sen\u00e1 pokrytie viac 80% \u0161pecifik\u00e1cie ONNX<\/strong>, zatia\u013e \u010do v OpenCV 4.x sa pokr\u00fdvalo menej ako 23%.
\nTo je pravdepodobne hlavn\u00e1 zmena vydania: OpenCV sa sna\u017e\u00ed sta\u0165 nielen obalen\u00edm okolo modelov, ale aj plnohodnotn\u00fdm integrovan\u00fdm motorom inferencie pre modern\u00e9 neur\u00f3nov\u00e9 siete.<\/p>\n<\/li>\n\nV\u00fdber motora pri na\u010d\u00edtan\u00ed modelu.<\/strong>
\nV cv::dnn::readNet() a pr\u00edbuzn\u00fdch funkci\u00e1ch sa objavil parameter ENGINE_AUTO. Predvolene sa OpenCV najprv pok\u00fa\u0161a o nov\u00fd motor a pri ne\u00faspechu prech\u00e1dza na klasick\u00fd. Spr\u00e1vanie je mo\u017en\u00e9 nastavi\u0165 aj cez premenn\u00fa prostredia OPENCV_FORCE_DNN_ENGINE: pou\u017e\u00edva\u0165 star\u00fd motor, nov\u00fd, automatick\u00fd re\u017eim alebo ONNX Runtime. Star\u00e9 projekty nemusia okam\u017eite prech\u00e1dza\u0165 na nov\u00fd motor, ale nov\u00e9 modely mo\u017eno postupne pres\u00fava\u0165 na neho.<\/p>\n<\/li>\n\nIntegr\u00e1cia s ONNX Runtime.<\/strong>
\nOpenCV je teraz mo\u017en\u00e9 zostavi\u0165 s pripojen\u00fdm ONNX Runtime. OpenCV pritom pou\u017e\u00edva vlastn\u00fd ONNX parser na kon\u0161trukciu grafu ORT, a preto nevy\u017eaduje pln\u00fd bal\u00edk ONNX. Pre zostavenie s\u00fa k dispoz\u00edcii parametre CMake -DWITH_ONNXRUNTIME=ON a -DDOWNLOAD_ONNXRUNTIME=ON; pre GPU NVIDIA \u2014 -DDOWNLOAD_ONNXRUNTIME_GPU=ON. Funkcia bude u\u017eito\u010dn\u00e1 pre t\u00fdch, ktor\u00ed potrebuj\u00fa GPU inferenciu u\u017e teraz, preto\u017ee nov\u00fd DNN motor OpenCV zatia\u013e funguje len na CPU.<\/p>\n<\/li>\n\nPodpora modelov v\u00edzie-jazyk.<\/strong>
\nNov\u00fd DNN motor zah\u0155\u0148a tokeniz\u00e1tory, vrstvy pozornosti, dek\u00f3dovacie bloky, post-spracovanie a KV-cache, ktor\u00e9 s\u00fa potrebn\u00e9 na spustenie VLM modelov \u201ez krabice\u201c. Kni\u017enica bola historicky spojen\u00e1 s spracovan\u00edm obr\u00e1zkov, a teraz vstupuje do oblasti modelov, ktor\u00e9 z\u00e1rove\u0148 pracuj\u00fa s obr\u00e1zkami a textom.<\/p>\n<\/li>\n\nPodpora GPU nov\u00e9ho DNN motora e\u0161te nie je pripraven\u00e1.<\/strong>
\nNov\u00fd motor v OpenCV 5.0 funguje iba na CPU. Nativn\u00fa podporu GPU s\u013eubuj\u00fa prida\u0165 v nasleduj\u00facich vydaniach; zatia\u013e sa pre GPU odpor\u00fa\u010da pou\u017ei\u0165 klasick\u00fd motor alebo zostavenie s ONNX Runtime a poskytovate\u013emi exek\u00facie NVIDIA. Pre serverov\u00fd alebo desktopov\u00fd CUDA inference OpenCV 5.0 samotn\u00fd e\u0161te nenahr\u00e1dza \u0161pecializovan\u00e9 runtime.<\/p>\n<\/li>\n\nBoli odstr\u00e1nen\u00e9 star\u00e9 parsers Darknet a Caffe.<\/strong>
\nV\u00fdvoj\u00e1ri odstr\u00e1nili parsers Darknet a Caffe, preto\u017ee v\u00e4\u010d\u0161ina modelov u\u017e bola preveden\u00e1 do ONNX. TFLite st\u00e1le zost\u00e1va cez klasick\u00fd motor, prenos na nov\u00fd motor je pl\u00e1novan\u00fd nesk\u00f4r. OpenCV sa vzd\u00e1va star\u00fdch form\u00e1tov a s\u00fastred\u00ed sa na ONNX ako hlavn\u00fd v\u00fdmenn\u00fd form\u00e1t modelov.<\/p>\n<\/li>\n\nVe\u013ek\u00e9 \u010distenie zastaran\u00e9ho API.<\/strong>
\nC API OpenCV 1.x bolo odstr\u00e1nen\u00e9: funkcie ako cvCreateMat() a star\u00e9 \u0161trukt\u00fary ako CvMat u\u017e nie s\u00fa s\u00fa\u010das\u0165ou hlavn\u00e9ho API. Podpora OpenVX bola takisto odstr\u00e1nen\u00e1, modul G-API bol presunut\u00fd do opencv_contrib, tam bol poslan\u00fd aj klasick\u00fd ML modul. Pre star\u00e9 projekty je to prelomov\u00e1 zmena, ale bola potrebn\u00e1: kni\u017enica sa zbavuje technick\u00e9ho dlhu, ktor\u00fd sa nahromadil za desa\u0165ro\u010dia.<\/p>\n<\/li>\n\nFeatures2D bol premenovan\u00fd na Features.<\/strong>
\nModul bol roz\u0161\u00edren\u00fd o modern\u00e9 vektory znakov z neur\u00f3nov\u00fdch siet\u00ed. V hlavnom repozit\u00e1ri zostali SIFT, ORB, FAST, GoodFeaturesToTrack a MSER, zatia\u013e \u010do zastaran\u00e9 detektory a deskriptor boli presunut\u00e9 do opencv_contrib. Pridan\u00e9 s\u00fa ALIKED, DISK a porovn\u00e1va\u010d LightGlue. OpenCV sa jasne prisp\u00f4sobuje modern\u00fdm pipeline, kde s\u00fa znaky obr\u00e1zkov \u010doraz \u010dastej\u0161ie extrahovan\u00e9 nie klasick\u00fdmi algoritmami, ale neur\u00f3nov\u00fdmi sie\u0165ami.<\/p>\n<\/li>\n\nNov\u00e9 typy \u00fadajov v Core.<\/strong>
\nPridan\u00e9 CV_16BF \/ cv::bfloat, CV_32U, CV_64U, CV_64S a CV_Bool. Taktie\u017e bola vylep\u0161en\u00e1 podpora N-dimenzion\u00e1lnych pol\u00ed, 1D pol\u00ed a 0D skalarov. LAPACK je teraz v\u017edy dostupn\u00fd: ak extern\u00e1 kni\u017enica nie je, OpenCV zostavuje zabudovan\u00e9 minim\u00e1lne podmno\u017estvo. Zmeny s\u00fa d\u00f4le\u017eit\u00e9 pre DNN, numerick\u00e9 algoritmy a kompatibilitu s modern\u00fdmi form\u00e1tmi d\u00e1t, kde sa bfloat16 a roz\u0161\u00edren\u00e9 typy vyskytuj\u00fa \u010doraz \u010dastej\u0161ie.<\/p>\n<\/li>\n\nSpracovanie obr\u00e1zkov bolo ur\u00fdchlen\u00e9.<\/strong>
\nV imgproc boli prepracovan\u00e9 warpAffine, warpPerspective a remap. Zlep\u0161enia v\u00fdkonnosti s\u00fa deklarovan\u00e9 v rozmedz\u00ed od 10 % do viac ako 300 % v z\u00e1vislosti od platformy, ve\u013ekosti obr\u00e1zka, typu \u00fadajov a vlajok oper\u00e1cie.
\nToto u\u017e nie je o AI module \u2014 tak\u00e9to optimaliz\u00e1cie s\u00fa priamo d\u00f4le\u017eit\u00e9 pre klasick\u00e9 \u00falohy: video streamy, stabiliz\u00e1ciu, projek\u010dn\u00e9 transform\u00e1cie a pr\u00edpravu \u00fadajov pred inferenciou.<\/p>\n<\/li>\n\nVylep\u0161en\u00fd v\u00fdstup textu.<\/strong>
\nStar\u00fd mechanizmus renderovania textu bol nahraden\u00fd STB-based TrueType enginom s integrovan\u00fdm variabiln\u00fdm p\u00edsmom Rubik. M\u00f4\u017eete pripoji\u0165 pou\u017e\u00edvate\u013esk\u00e9 p\u00edsma, bola zaveden\u00e1 \u0161irok\u00e1 podpora Unicode. Pritom zlo\u017een\u00e9 p\u00edsma ako arabsk\u00e9 a devanagari sa zatia\u013e nezobrazuj\u00fa spr\u00e1vne bez Harfbuzz, a farebn\u00e9 emoji nie s\u00fa podporovan\u00e9. OpenCV sa st\u00e1va pohodlnej\u0161\u00edm na generovanie anot\u00e1ci\u00ed, lad\u00edc\u00edch obrazov a vizualiz\u00e1ci\u00ed bez extern\u00fdch kni\u017en\u00edc.<\/p>\n<\/li>\n\nAktualizovan\u00e9 3D a kalibr\u00e1cia kamier.<\/strong>
\ncalib3d bol rozdelen\u00fd na geometry, calib, stereo a ptcloud. USAC sa stal \u0161tandardn\u00fdm backendom pre stabiln\u00e9 vyhodnocovanie homografie, essential\/fundamental matice, PnP a \u010fal\u0161\u00edch \u00faloh. Pridan\u00fd pipeline na kalibr\u00e1ciu viacer\u00fdch kamier, po\u010diato\u010dn\u00e1 podpora spracovania siet\u00ed a mrakov bodov vr\u00e1tane TSDF, ICP, importu a exportu .ply a .obj. To je d\u00f4le\u017eit\u00fd krok pre robotiku, SLAM, AR\/VR a viac kamier syst\u00e9my, kde u\u017e d\u00e1vno nie je dostato\u010dn\u00e9 len 2D spracovanie.<\/p>\n<\/li>\n\nPrepracovan\u00e9 pr\u00edklady a dokument\u00e1cia.<\/strong>
\nZastaral\u00e9 pr\u00edklady boli odstr\u00e1nen\u00e9, zost\u00e1vaj\u00face aktualizovan\u00e9. Pre DNN pr\u00edklady sa pou\u017e\u00edva spolo\u010dn\u00e1 zbierka modelov, ktor\u00fa je mo\u017en\u00e9 stiahnu\u0165 skriptom download_models.py. Pridan\u00e9 experiment\u00e1lne pr\u00edklady pre VLM a LDM, a dokument\u00e1cia dostala modern\u00fa t\u00e9mu, r\u00fdchle vyh\u013ead\u00e1vanie, svetl\u00fd\/tmav\u00fd re\u017eim a vylep\u0161en\u00fa navig\u00e1ciu. Pre tak\u00fa ve\u013ek\u00fa kni\u017enicu to nie je kozmetick\u00e1 zmena: norm\u00e1lna dokument\u00e1cia a funguj\u00face pr\u00edklady priamo ovplyv\u0148uj\u00fa to, \u010di sa v\u00fdvoj\u00e1ri skuto\u010dne m\u00f4\u017eu presun\u00fa\u0165 na OpenCV 5.x.<\/p>\n<\/li>\n<\/ul>\nCelkovo OpenCV 5.0 nie je len \u010fal\u0161ia verzia s optimaliz\u00e1ciami, ale v\u00fdznamn\u00e1 aktualiz\u00e1cia v\u0161etk\u00e9ho: menej dedi\u010dstva z obdobia OpenCV 1.x a viac orient\u00e1cie na ONNX, neur\u00f3nov\u00fd inferen\u010dn\u00fd syst\u00e9m, modern\u00e9 rysy, 3D spracovanie a heterog\u00e9nne ur\u00fdchlenie. Pre star\u00e9 projekty m\u00f4\u017ee by\u0165 prechod n\u00e1ro\u010dn\u00fd vzh\u013eadom na odstr\u00e1nenie C API a presun niektor\u00fdch modulov do opencv_contrib.<\/p>\n
Zdroj: linux.org.ru<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"\u041e\u043f\u0443\u0431\u043b\u0438\u043a\u043e\u0432\u0430\u043d OpenCV 5.0.0 \u2014 \u043a\u0440\u0443\u043f\u043d\u044b\u0439 \u0432\u044b\u043f\u0443\u0441\u043a \u043e\u0442\u043a\u0440\u044b\u0442\u043e\u0439 \u0431\u0438\u0431\u043b\u0438\u043e\u0442\u0435\u043a\u0438 \u043a\u043e\u043c\u043f\u044c\u044e\u0442\u0435\u0440\u043d\u043e\u0433\u043e \u0437\u0440\u0435\u043d\u0438\u044f \u0438 \u043e\u0431\u0440\u0430\u0431\u043e\u0442\u043a\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439. \u0420\u0435\u043b\u0438\u0437 \u0441\u043e\u0441\u0442\u043e\u044f\u043b\u0441\u044f 6 \u0438\u044e\u043d\u044f 2026 \u0433\u043e\u0434\u0430 \u0438 \u0443\u0436\u0435 \u043e\u0442\u043c\u0435\u0447\u0435\u043d \u043a\u0430\u043a \u0430\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u044b\u0439 \u0432 \u0440\u0435\u043f\u043e\u0437\u0438\u0442\u043e\u0440\u0438\u0438 \u043f\u0440\u043e\u0435\u043a\u0442\u0430 \u043d\u0430 GitHub. OpenCV \u0440\u0430\u0441\u043f\u0440\u043e\u0441\u0442\u0440\u0430\u043d\u044f\u0435\u0442\u0441\u044f \u043f\u043e\u0434 \u043b\u0438\u0446\u0435\u043d\u0437\u0438\u0435\u0439 Apache-2.0. \u0421\u0443\u0442\u044c \u0440\u0435\u043b\u0438\u0437\u0430 \u2014 \u043f\u0435\u0440\u0435\u0445\u043e\u0434 OpenCV \u043e\u0442 \u00ab\u043a\u043b\u0430\u0441\u0441\u0438\u0447\u0435\u0441\u043a\u043e\u0439 \u0431\u0438\u0431\u043b\u0438\u043e\u0442\u0435\u043a\u0438 \u043a\u043e\u043c\u043f\u044c\u044e\u0442\u0435\u0440\u043d\u043e\u0433\u043e \u0437\u0440\u0435\u043d\u0438\u044f\u00bb \u043a \u0431\u043e\u043b\u0435\u0435 \u0441\u043e\u0432\u0440\u0435\u043c\u0435\u043d\u043d\u043e\u0439 \u043f\u043b\u0430\u0442\u0444\u043e\u0440\u043c\u0435 \u0434\u043b\u044f CV\/AI-\u0437\u0430\u0434\u0430\u0447: \u043f\u0440\u043e\u0435\u043a\u0442 \u0432\u044b\u0447\u0438\u0441\u0442\u0438\u043b \u0441\u0442\u0430\u0440\u043e\u0435 API, \u043f\u0435\u0440\u0435\u0440\u0430\u0431\u043e\u0442\u0430\u043b […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":8,"featured_media":181945,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-181944","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd47OpenCV 5.0.0 | ProHoster","description":"Zverejnen\u00e9","canonical_url":"https:\/\/prohoster.info\/sk\/blog\/news\/opencv-5-0-0","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"sk_SK","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47OpenCV 5.0.0 | ProHoster","og:description":"\u041e\u043f\u0443\u0431\u043b\u0438\u043a\u043e\u0432\u0430\u043d","og:url":"https:\/\/prohoster.info\/sk\/blog\/news\/opencv-5-0-0","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2026-06-08T00:48:46+00:00","article:modified_time":"2026-06-08T00:48:46+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":[],"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/181944","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/comments?post=181944"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/181944\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/media\/181945"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/media?parent=181944"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/categories?post=181944"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/tags?post=181944"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![](\"\/wp-content\/uploads\/2026\/06\/0827c35d7663a25a36859f66c23893bc.jpg\")
<\/p>\n