{"id":1321,"date":"2023-08-13T11:47:19","date_gmt":"2023-08-13T11:47:19","guid":{"rendered":"https:\/\/www.gptmain.news\/?p=1321"},"modified":"2023-08-13T11:47:19","modified_gmt":"2023-08-13T11:47:19","slug":"%d0%ba%d0%b0%d0%ba-%d1%80%d0%b0%d0%b1%d0%be%d1%82%d0%b0%d0%b5%d1%82-vision-transformer-vit-%d0%b7%d0%b0-10-%d0%bc%d0%b8%d0%bd%d1%83%d1%82-%d0%b8%d0%b7%d0%be%d0%b1%d1%80%d0%b0%d0%b6%d0%b5%d0%bd","status":"publish","type":"post","link":"https:\/\/gptmain.news\/?p=1321","title":{"rendered":"\u041a\u0430\u043a \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442 Vision Transformer (ViT) \u0437\u0430 10 \u043c\u0438\u043d\u0443\u0442: \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16\u044516 \u0441\u043b\u043e\u0432\n | GPTMain News"},"content":{"rendered":"<div id=\"\">\n<p>\u041d\u0430 \u044d\u0442\u043e\u0442 \u0440\u0430\u0437 \u044f \u0431\u0443\u0434\u0443 \u0440\u0435\u0437\u043a\u0438\u043c \u0438 \u043a\u0440\u0430\u0442\u043a\u0438\u043c.  \u0427\u0435\u0440\u0435\u0437 10 \u043c\u0438\u043d\u0443\u0442 \u044f \u0443\u043a\u0430\u0436\u0443 \u043c\u0435\u043b\u043a\u0438\u0435 \u043c\u043e\u0434\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u044b \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u0430 \u0434\u043b\u044f \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439.<\/p>\n<p>\u041f\u043e\u0441\u043a\u043e\u043b\u044c\u043a\u0443 \u044d\u0442\u043e \u0441\u043b\u0435\u0434\u0443\u044e\u0449\u0430\u044f \u0441\u0442\u0430\u0442\u044c\u044f, \u043d\u0435 \u0441\u0442\u0435\u0441\u043d\u044f\u0439\u0442\u0435\u0441\u044c \u0441\u043e\u0432\u0435\u0442\u043e\u0432\u0430\u0442\u044c \u043c\u043e\u0438 \u043f\u0440\u0435\u0434\u044b\u0434\u0443\u0449\u0438\u0435 \u0441\u0442\u0430\u0442\u044c\u0438 \u043e Transformer \u0438 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u0435\u0441\u043b\u0438 \u0432\u044b \u043d\u0435 \u0447\u0443\u0432\u0441\u0442\u0432\u0443\u0435\u0442\u0435 \u0441\u0435\u0431\u044f \u043a\u043e\u043c\u0444\u043e\u0440\u0442\u043d\u043e \u0441 \u0442\u0435\u0440\u043c\u0438\u043d\u0430\u043c\u0438.<\/p>\n<p>\u0422\u0435\u043f\u0435\u0440\u044c, \u0434\u0430\u043c\u044b \u0438 \u0433\u043e\u0441\u043f\u043e\u0434\u0430, \u043c\u043e\u0436\u0435\u0442\u0435 \u0437\u0430\u043f\u0443\u0441\u043a\u0430\u0442\u044c \u0447\u0430\u0441\u044b!<\/p>\n<p>\u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0430\u0442\u043e\u0440\u0430\u043c \u043d\u0435 \u0445\u0432\u0430\u0442\u0430\u0435\u0442 \u0438\u043d\u0434\u0443\u043a\u0442\u0438\u0432\u043d\u044b\u0445 \u0441\u043c\u0435\u0449\u0435\u043d\u0438\u0439 \u0441\u0432\u0435\u0440\u0442\u043e\u0447\u043d\u044b\u0445 \u043d\u0435\u0439\u0440\u043e\u043d\u043d\u044b\u0445 \u0441\u0435\u0442\u0435\u0439 (CNN), \u0442\u0430\u043a\u0438\u0445 \u043a\u0430\u043a \u0438\u043d\u0432\u0430\u0440\u0438\u0430\u043d\u0442\u043d\u043e\u0441\u0442\u044c \u043a \u0442\u0440\u0430\u043d\u0441\u043b\u044f\u0446\u0438\u0438 \u0438 \u043b\u043e\u043a\u0430\u043b\u044c\u043d\u043e \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u043d\u043e\u0435 \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u043e\u0435 \u043f\u043e\u043b\u0435.  \u0412\u044b, \u043d\u0430\u0432\u0435\u0440\u043d\u043e\u0435, \u0441\u043b\u044b\u0448\u0430\u043b\u0438 \u044d\u0442\u043e \u0440\u0430\u043d\u044c\u0448\u0435.<\/p>\n<p>\u041d\u043e \u0447\u0442\u043e \u044d\u0442\u043e \u043d\u0430 \u0441\u0430\u043c\u043e\u043c \u0434\u0435\u043b\u0435 \u043e\u0437\u043d\u0430\u0447\u0430\u0435\u0442?<\/p>\n<p>\u0425\u043e\u0440\u043e\u0448\u043e, <strong>\u043d\u0435\u0438\u0437\u043c\u0435\u043d\u043d\u043e\u0441\u0442\u044c<\/strong> \u043e\u0437\u043d\u0430\u0447\u0430\u0435\u0442, \u0447\u0442\u043e \u0432\u044b \u043c\u043e\u0436\u0435\u0442\u0435 \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0442\u044c \u043e\u0431\u044a\u0435\u043a\u0442 (\u0442.\u0435. \u043e\u0431\u044a\u0435\u043a\u0442) \u043d\u0430 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0438, \u0434\u0430\u0436\u0435 \u0435\u0441\u043b\u0438 \u0435\u0433\u043e \u0432\u043d\u0435\u0448\u043d\u0438\u0439 \u0432\u0438\u0434 \u0438\u043b\u0438 \u043f\u043e\u043b\u043e\u0436\u0435\u043d\u0438\u0435 \u0440\u0430\u0437\u043b\u0438\u0447\u0430\u044e\u0442\u0441\u044f. <strong>\u041f\u0435\u0440\u0435\u0432\u043e\u0434<\/strong> \u0432 \u043a\u043e\u043c\u043f\u044c\u044e\u0442\u0435\u0440\u043d\u043e\u043c \u0437\u0440\u0435\u043d\u0438\u0438 \u043f\u043e\u0434\u0440\u0430\u0437\u0443\u043c\u0435\u0432\u0430\u0435\u0442, \u0447\u0442\u043e \u043a\u0430\u0436\u0434\u044b\u0439 \u043f\u0438\u043a\u0441\u0435\u043b\u044c \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f \u0431\u044b\u043b \u043f\u0435\u0440\u0435\u043c\u0435\u0449\u0435\u043d \u043d\u0430 \u0444\u0438\u043a\u0441\u0438\u0440\u043e\u0432\u0430\u043d\u043d\u0443\u044e \u0432\u0435\u043b\u0438\u0447\u0438\u043d\u0443 \u0432 \u043e\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u043d\u043e\u043c \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0438.<\/p>\n<p>\u0411\u043e\u043b\u0435\u0435 \u0442\u043e\u0433\u043e, \u043f\u043e\u043c\u043d\u0438\u0442\u0435, \u0447\u0442\u043e \u0441\u0432\u0435\u0440\u0442\u043a\u0430 \u2014 \u044d\u0442\u043e \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u0439 \u043b\u043e\u043a\u0430\u043b\u044c\u043d\u044b\u0439 \u043e\u043f\u0435\u0440\u0430\u0442\u043e\u0440.  \u041c\u044b \u0432\u0438\u0434\u0438\u043c \u0442\u043e\u043b\u044c\u043a\u043e \u0441\u043e\u0441\u0435\u0434\u043d\u0438\u0435 \u0437\u043d\u0430\u0447\u0435\u043d\u0438\u044f, \u0443\u043a\u0430\u0437\u0430\u043d\u043d\u044b\u0435 \u044f\u0434\u0440\u043e\u043c.<\/p>\n<p>\u0421 \u0434\u0440\u0443\u0433\u043e\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u044b, \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0430\u0442\u043e\u0440 \u043f\u043e \u0441\u0432\u043e\u0435\u0439 \u043a\u043e\u043d\u0441\u0442\u0440\u0443\u043a\u0446\u0438\u0438 <strong>\u0438\u043d\u0432\u0430\u0440\u0438\u0430\u043d\u0442 \u043f\u0435\u0440\u0435\u0441\u0442\u0430\u043d\u043e\u0432\u043a\u0438<\/strong>.  \u041f\u043b\u043e\u0445\u0430\u044f \u043d\u043e\u0432\u043e\u0441\u0442\u044c \u0437\u0430\u043a\u043b\u044e\u0447\u0430\u0435\u0442\u0441\u044f \u0432 \u0442\u043e\u043c, \u0447\u0442\u043e \u043e\u043d \u043d\u0435 \u043c\u043e\u0436\u0435\u0442 \u043e\u0431\u0440\u0430\u0431\u0430\u0442\u044b\u0432\u0430\u0442\u044c \u0434\u0430\u043d\u043d\u044b\u0435 \u0432 \u0432\u0438\u0434\u0435 \u0441\u0435\u0442\u043a\u0438.  \u041d\u0430\u043c \u043d\u0443\u0436\u043d\u044b \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438!  \u0421 \u044d\u0442\u043e\u0439 \u0446\u0435\u043b\u044c\u044e \u043c\u044b \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u0443\u0435\u043c \u043f\u0440\u043e\u0441\u0442\u0440\u0430\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0439 \u043d\u0435\u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u044b\u0439 \u0441\u0438\u0433\u043d\u0430\u043b \u0432 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u044c!<\/p>\n<p>\u041f\u043e\u0441\u043c\u043e\u0442\u0440\u0438\u043c, \u043a\u0430\u043a.<\/p>\n<h2 id=\"how-the-vision-transformer-works-in-a-nutshell\">\u041a\u0440\u0430\u0442\u043a\u043e \u043e \u0442\u043e\u043c, \u043a\u0430\u043a \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442 Vision Transformer<\/h2>\n<p>\u041e\u0431\u0449\u0430\u044f \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u0430 \u043d\u0430\u0437\u044b\u0432\u0430\u0435\u0442\u0441\u044f Vision Transformer (\u0441\u043e\u043a\u0440\u0430\u0449\u0435\u043d\u043d\u043e ViT).  \u0414\u0430\u0432\u0430\u0439\u0442\u0435 \u0440\u0430\u0441\u0441\u043c\u043e\u0442\u0440\u0438\u043c \u0435\u0433\u043e \u0448\u0430\u0433 \u0437\u0430 \u0448\u0430\u0433\u043e\u043c.<\/p>\n<ol>\n<li>\n<p>\u0420\u0430\u0437\u0431\u0438\u0442\u044c \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u043d\u0430 \u043f\u0430\u0442\u0447\u0438<\/p>\n<\/li>\n<li>\n<p>\u0421\u0433\u043b\u0430\u0434\u0438\u0442\u044c \u043f\u0430\u0442\u0447\u0438<\/p>\n<\/li>\n<li>\n<p>\u0421\u043e\u0437\u0434\u0430\u0432\u0430\u0439\u0442\u0435 \u043d\u0438\u0437\u043a\u043e\u0440\u0430\u0437\u043c\u0435\u0440\u043d\u044b\u0435 \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u0435 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u044f \u0438\u0437 \u0441\u0433\u043b\u0430\u0436\u0435\u043d\u043d\u044b\u0445 \u0443\u0447\u0430\u0441\u0442\u043a\u043e\u0432.<\/p>\n<\/li>\n<li>\n<p>\u0414\u043e\u0431\u0430\u0432\u0438\u0442\u044c \u043f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u044b\u0435 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u044f<\/p>\n<\/li>\n<li>\n<p>\u041f\u043e\u0434\u0430\u0439\u0442\u0435 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u044c \u0432 \u043a\u0430\u0447\u0435\u0441\u0442\u0432\u0435 \u0432\u0445\u043e\u0434\u043d\u044b\u0445 \u0434\u0430\u043d\u043d\u044b\u0445 \u0434\u043b\u044f \u0441\u0442\u0430\u043d\u0434\u0430\u0440\u0442\u043d\u043e\u0433\u043e \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u043e\u0433\u043e \u044d\u043d\u043a\u043e\u0434\u0435\u0440\u0430.<\/p>\n<\/li>\n<li>\n<p>\u041f\u0440\u0435\u0434\u0432\u0430\u0440\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u0443\u0447\u0438\u0442\u0435 \u043c\u043e\u0434\u0435\u043b\u044c \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u043c\u0435\u0442\u043e\u043a \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 (\u043f\u043e\u043b\u043d\u043e\u0441\u0442\u044c\u044e \u043a\u043e\u043d\u0442\u0440\u043e\u043b\u0438\u0440\u0443\u0435\u0442\u0441\u044f \u043d\u0430 \u043e\u0433\u0440\u043e\u043c\u043d\u043e\u043c \u043d\u0430\u0431\u043e\u0440\u0435 \u0434\u0430\u043d\u043d\u044b\u0445)<\/p>\n<\/li>\n<li>\n<p>\u0422\u043e\u043d\u043a\u0430\u044f \u043d\u0430\u0441\u0442\u0440\u043e\u0439\u043a\u0430 \u043d\u0438\u0441\u0445\u043e\u0434\u044f\u0449\u0435\u0433\u043e \u043d\u0430\u0431\u043e\u0440\u0430 \u0434\u0430\u043d\u043d\u044b\u0445 \u0434\u043b\u044f \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439<\/p>\n<\/li>\n<\/ol>\n<p><img decoding=\"async\" src=\"https:\/\/media.giphy.com\/media\/ATsWtUsuuFRfq8OhZ7\/source.gif\" alt=\"\u0432\u0438\u0434\u0435\u043d\u0438\u0435-\u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440-gif\"\/><br \/>\n<em>\u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0431\u043b\u043e\u0433 Google AI.<\/em><\/p>\n<p>\u041f\u0430\u0442\u0447\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u2014 \u044d\u0442\u043e \u0432 \u043e\u0441\u043d\u043e\u0432\u043d\u043e\u043c \u0442\u043e\u043a\u0435\u043d\u044b \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438 (\u043d\u0430\u043f\u0440\u0438\u043c\u0435\u0440, \u0441\u043b\u043e\u0432\u0430).  \u0424\u0430\u043a\u0442\u0438\u0447\u0435\u0441\u043a\u0438 \u0431\u043b\u043e\u043a \u044d\u043d\u043a\u043e\u0434\u0435\u0440\u0430 \u0438\u0434\u0435\u043d\u0442\u0438\u0447\u0435\u043d \u043e\u0440\u0438\u0433\u0438\u043d\u0430\u043b\u044c\u043d\u043e\u043c\u0443 \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u043e\u0432\u0430\u0442\u0435\u043b\u044e, \u043f\u0440\u0435\u0434\u043b\u043e\u0436\u0435\u043d\u043d\u043e\u043c\u0443 Vaswani et al.  (2017), \u043a\u0430\u043a \u043c\u044b \u043f\u043e\u0434\u0440\u043e\u0431\u043d\u043e \u043e\u043f\u0438\u0441\u0430\u043b\u0438:<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:216px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:199.53703703703704%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,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');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0431\u043b\u043e\u043a-\u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440-\u0432\u0438\u0442\" title=\"\u0431\u043b\u043e\u043a-\u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440-\u0432\u0438\u0442\" src=\"https:\/\/theaisummer.com\/static\/aa65d942973255da238052d8cdfa4fcd\/7d4ec\/the-transformer-block-vit.png 216w\" sizes=\"(max-width: 216px) 100vw, 216px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0412\u0441\u0435\u043c \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u044b\u0439 \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0430\u0442\u043e\u0440\u043d\u044b\u0439 \u0431\u043b\u043e\u043a.  \u0418\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0410\u043b\u0435\u043a\u0441\u0435\u044f \u0414\u043e\u0441\u043e\u0432\u0438\u0446\u043a\u043e\u0433\u043e \u0438 \u0434\u0440. 2020. \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0418\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16&#215;16 \u0441\u043b\u043e\u0432: \u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u044b \u0434\u043b\u044f \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0435<\/em><\/p>\n<p>\u0415\u0434\u0438\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0435, \u0447\u0442\u043e \u043c\u0435\u043d\u044f\u0435\u0442\u0441\u044f, \u044d\u0442\u043e \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u043e \u044d\u0442\u0438\u0445 \u0431\u043b\u043e\u043a\u043e\u0432.  \u0421 \u044d\u0442\u043e\u0439 \u0446\u0435\u043b\u044c\u044e \u0438 \u0434\u043b\u044f \u0434\u0430\u043b\u044c\u043d\u0435\u0439\u0448\u0435\u0433\u043e \u0434\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u0442\u043e\u0433\u043e, \u0447\u0442\u043e \u0441 \u0431\u043e\u043b\u044c\u0448\u0438\u043c \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u043e\u043c \u0434\u0430\u043d\u043d\u044b\u0445 \u043e\u043d\u0438 \u043c\u043e\u0433\u0443\u0442 \u043e\u0431\u0443\u0447\u0430\u0442\u044c \u0431\u043e\u043b\u0435\u0435 \u043a\u0440\u0443\u043f\u043d\u044b\u0435 \u0432\u0430\u0440\u0438\u0430\u043d\u0442\u044b ViT, \u0431\u044b\u043b\u0438 \u043f\u0440\u0435\u0434\u043b\u043e\u0436\u0435\u043d\u044b 3 \u043c\u043e\u0434\u0435\u043b\u0438:<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:718px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:20.666666666666668%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAAECAIAAAABPYjBAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAt0lEQVQI1yVOOQ6DMBDk\/+9KulBgHIOEZYx8RwYhDnMVySTZYjWrOXaycRyLopBSzvOMXde1c44QwqoKWCndti0piRDCWVuWRMquFYJSOgxDtu875xzctm3TNHnvj+NIadXaIAUYAvhDCOd5eufWde1jRDLODCxjLM8fIbyu67LWvX\/Dm0Z23R8\/KVVKASAOnr7vb\/f715xSAgGpdx7N8WFZFsRrra21YHEaY9AIGCzaxRibhkPzAewm2jDop0dbAAAAAElFTkSuQmCC');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0432\u0438\u0442-\u043c\u043e\u0434\u0435\u043b\u0438-\u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435-\u0442\u0430\u0431\u043b\u0438\u0446\u0430\" title=\"\u0432\u0438\u0442-\u043c\u043e\u0434\u0435\u043b\u0438-\u043e\u043f\u0438\u0441\u0430\u043d\u0438\u0435-\u0442\u0430\u0431\u043b\u0438\u0446\u0430\" src=\"https:\/\/theaisummer.com\/static\/156f6be8c232bb0b03d4793b2be2fbad\/5a46d\/vit-models-description-table.png 300w,\/static\/156f6be8c232bb0b03d4793b2be2fbad\/0a47e\/vit-models-description-table.png 600w,\/static\/156f6be8c232bb0b03d4793b2be2fbad\/57dc1\/vit-models-description-table.png 718w\" sizes=\"(max-width: 718px) 100vw, 718px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u0414\u043e\u0441\u043e\u0432\u0438\u0446\u043a\u0438\u0439 \u0438 \u0434\u0440. 2020. \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0418\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16&#215;16 \u0441\u043b\u043e\u0432: \u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u044b \u0434\u043b\u044f \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0435<\/em><\/p>\n<p>\u0413\u043e\u043b\u043e\u0432\u043a\u0438 \u043e\u0442\u043d\u043e\u0441\u044f\u0442\u0441\u044f \u043a \u043c\u043d\u043e\u0433\u043e\u0433\u043e\u043b\u043e\u0432\u043e\u043c\u0443 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044e, \u0430 \u0440\u0430\u0437\u043c\u0435\u0440 MLP \u043e\u0442\u043d\u043e\u0441\u0438\u0442\u0441\u044f \u043a \u0441\u0438\u043d\u0435\u043c\u0443 \u043c\u043e\u0434\u0443\u043b\u044e \u043d\u0430 \u0440\u0438\u0441\u0443\u043d\u043a\u0435.  MLP \u0440\u0430\u0441\u0448\u0438\u0444\u0440\u043e\u0432\u044b\u0432\u0430\u0435\u0442\u0441\u044f \u043a\u0430\u043a \u043c\u043d\u043e\u0433\u043e\u0441\u043b\u043e\u0439\u043d\u044b\u0439 \u043f\u0435\u0440\u0441\u0435\u043f\u0442\u0440\u043e\u043d, \u043d\u043e \u043d\u0430 \u0441\u0430\u043c\u043e\u043c \u0434\u0435\u043b\u0435 \u044d\u0442\u043e \u043d\u0430\u0431\u043e\u0440 \u0441\u043b\u043e\u0435\u0432 \u043b\u0438\u043d\u0435\u0439\u043d\u043e\u0433\u043e \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u043e\u0432\u0430\u043d\u0438\u044f.<\/p>\n<p>\u0421\u043a\u0440\u044b\u0442\u044b\u0439 \u0440\u0430\u0437\u043c\u0435\u0440 <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u0414<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u0414<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.02778em\">\u0414<\/span><\/span><\/span><\/span><\/span>  &#8211; \u0440\u0430\u0437\u043c\u0435\u0440 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u044f, \u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u043e\u0441\u0442\u0430\u0435\u0442\u0441\u044f \u0444\u0438\u043a\u0441\u0438\u0440\u043e\u0432\u0430\u043d\u043d\u044b\u043c \u0432\u043e \u0432\u0441\u0435\u0445 \u0441\u043b\u043e\u044f\u0445.  \u0417\u0430\u0447\u0435\u043c \u0434\u0435\u0440\u0436\u0430\u0442\u044c \u044d\u0442\u043e \u0444\u0438\u043a\u0441\u0438\u0440\u043e\u0432\u0430\u043d\u043d\u044b\u043c?  \u0422\u0430\u043a \u0447\u0442\u043e \u043c\u044b \u043c\u043e\u0436\u0435\u043c \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u0442\u044c \u043a\u043e\u0440\u043e\u0442\u043a\u0438\u0435 \u0441\u043e\u0435\u0434\u0438\u043d\u0435\u043d\u0438\u044f \u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e\u0433\u043e \u043f\u0440\u043e\u043f\u0443\u0441\u043a\u0430.<\/p>\n<p>\u0415\u0441\u043b\u0438 \u0432\u044b \u043f\u0440\u043e\u043f\u0443\u0441\u0442\u0438\u043b\u0438 \u044d\u0442\u043e, \u0435\u0441\u0442\u044c <strong>\u043d\u0435\u0442<\/strong> \u0434\u0435\u043a\u043e\u0434\u0435\u0440 \u0432 \u0438\u0433\u0440\u0435.  \u041f\u0440\u043e\u0441\u0442\u043e \u0434\u043e\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0439 \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u0439 \u0441\u043b\u043e\u0439 \u0434\u043b\u044f \u043e\u043a\u043e\u043d\u0447\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438, \u043d\u0430\u0437\u044b\u0432\u0430\u0435\u043c\u044b\u0439 \u0433\u043e\u043b\u043e\u0432\u043a\u043e\u0439 MLP.<\/p>\n<p>\u041d\u043e \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u043b\u0438 \u044d\u0442\u043e\u0433\u043e?<\/p>\n<p>\u0414\u0430 \u0438 \u043d\u0435\u0442.  \u041d\u0430 \u0441\u0430\u043c\u043e\u043c \u0434\u0435\u043b\u0435 \u043d\u0430\u043c \u043d\u0443\u0436\u0435\u043d \u043e\u0433\u0440\u043e\u043c\u043d\u044b\u0439 \u043e\u0431\u044a\u0435\u043c \u0434\u0430\u043d\u043d\u044b\u0445 \u0438, \u043a\u0430\u043a \u0441\u043b\u0435\u0434\u0441\u0442\u0432\u0438\u0435, \u0432\u044b\u0447\u0438\u0441\u043b\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0435 \u0440\u0435\u0441\u0443\u0440\u0441\u044b.<\/p>\n<h2 id=\"important-details\">\u0412\u0430\u0436\u043d\u044b\u0435 \u0434\u0435\u0442\u0430\u043b\u0438<\/h2>\n<p>\u0412 \u0447\u0430\u0441\u0442\u043d\u043e\u0441\u0442\u0438, \u0435\u0441\u043b\u0438 ViT \u043e\u0431\u0443\u0447\u0435\u043d \u043d\u0430 \u043d\u0430\u0431\u043e\u0440\u0430\u0445 \u0434\u0430\u043d\u043d\u044b\u0445 \u0441 \u0431\u043e\u043b\u0435\u0435 \u0447\u0435\u043c 14 \u043c\u0438\u043b\u043b\u0438\u043e\u043d\u0430\u043c\u0438 (\u043a\u0430\u043a \u043c\u0438\u043d\u0438\u043c\u0443\u043c: P) \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439, \u043e\u043d \u043c\u043e\u0436\u0435\u0442 \u043f\u0440\u0438\u0431\u043b\u0438\u0437\u0438\u0442\u044c\u0441\u044f \u0438\u043b\u0438 \u043f\u0440\u0435\u0432\u0437\u043e\u0439\u0442\u0438 \u0441\u043e\u0432\u0440\u0435\u043c\u0435\u043d\u043d\u044b\u0435 CNN.<\/p>\n<p>\u0415\u0441\u043b\u0438 \u043d\u0435\u0442, \u0432\u0430\u043c \u043b\u0443\u0447\u0448\u0435 \u043f\u0440\u0438\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u0442\u044c\u0441\u044f ResNets \u0438\u043b\u0438 EfficientNets.<\/p>\n<p>ViT \u043f\u0440\u0435\u0434\u0432\u0430\u0440\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u0443\u0447\u0430\u0435\u0442\u0441\u044f \u043d\u0430 \u0431\u043e\u043b\u044c\u0448\u043e\u043c \u043d\u0430\u0431\u043e\u0440\u0435 \u0434\u0430\u043d\u043d\u044b\u0445, \u0430 \u0437\u0430\u0442\u0435\u043c \u043d\u0430\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043d\u0430 \u043c\u0430\u043b\u044b\u0435.  \u0415\u0434\u0438\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u0430\u044f \u043c\u043e\u0434\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u044f \u0441\u043e\u0441\u0442\u043e\u0438\u0442 \u0432 \u0442\u043e\u043c, \u0447\u0442\u043e\u0431\u044b \u043e\u0442\u043a\u0430\u0437\u0430\u0442\u044c\u0441\u044f \u043e\u0442 \u043f\u0440\u043e\u0433\u043d\u043e\u0437\u0438\u0440\u0443\u044e\u0449\u0435\u0439 \u0433\u043e\u043b\u043e\u0432\u043a\u0438 (MLP-\u0433\u043e\u043b\u043e\u0432\u043a\u0438) \u0438 \u043f\u0440\u0438\u0441\u043e\u0435\u0434\u0438\u043d\u0438\u0442\u044c \u043d\u043e\u0432\u0443\u044e. <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u0414<\/mi><mo>\u00d7<\/mo><mi>\u041a<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u0414 \\ \u0440\u0430\u0437 \u041a<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.76666em;vertical-align:-0.08333em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.02778em\">\u0414<\/span><span class=\"mspace\" style=\"margin-right:0.2222222222222222em\"\/><span class=\"mbin\">\u00d7<\/span><span class=\"mspace\" style=\"margin-right:0.2222222222222222em\"\/><\/span><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.07153em\">\u041a<\/span><\/span><\/span><\/span><\/span>  \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u0439 \u0441\u043b\u043e\u0439, \u0433\u0434\u0435 K \u2014 \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u043e \u043a\u043b\u0430\u0441\u0441\u043e\u0432 \u043d\u0435\u0431\u043e\u043b\u044c\u0448\u043e\u0433\u043e \u043d\u0430\u0431\u043e\u0440\u0430 \u0434\u0430\u043d\u043d\u044b\u0445.<\/p>\n<blockquote>\n<p>\u041c\u043d\u0435 \u043f\u043e\u043a\u0430\u0437\u0430\u043b\u043e\u0441\u044c \u0438\u043d\u0442\u0435\u0440\u0435\u0441\u043d\u044b\u043c, \u0447\u0442\u043e \u0430\u0432\u0442\u043e\u0440\u044b \u0443\u0442\u0432\u0435\u0440\u0436\u0434\u0430\u044e\u0442, \u0447\u0442\u043e \u043b\u0443\u0447\u0448\u0435 \u043f\u0440\u043e\u0432\u043e\u0434\u0438\u0442\u044c \u0442\u043e\u043d\u043a\u0443\u044e \u043d\u0430\u0441\u0442\u0440\u043e\u0439\u043a\u0443 \u043f\u0440\u0438 \u0431\u043e\u043b\u0435\u0435 \u0432\u044b\u0441\u043e\u043a\u0438\u0445 \u0440\u0430\u0437\u0440\u0435\u0448\u0435\u043d\u0438\u044f\u0445, \u0447\u0435\u043c \u043f\u0440\u0435\u0434\u0432\u0430\u0440\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435.<\/p>\n<\/blockquote>\n<p>\u0414\u043b\u044f \u0442\u043e\u0447\u043d\u043e\u0439 \u043d\u0430\u0441\u0442\u0440\u043e\u0439\u043a\u0438 \u0432 \u0431\u043e\u043b\u0435\u0435 \u0432\u044b\u0441\u043e\u043a\u0438\u0445 \u0440\u0430\u0437\u0440\u0435\u0448\u0435\u043d\u0438\u044f\u0445 \u0432\u044b\u043f\u043e\u043b\u043d\u044f\u0435\u0442\u0441\u044f \u0434\u0432\u0443\u043c\u0435\u0440\u043d\u0430\u044f \u0438\u043d\u0442\u0435\u0440\u043f\u043e\u043b\u044f\u0446\u0438\u044f \u043f\u0440\u0435\u0434\u0432\u0430\u0440\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u043d\u044b\u0445 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u0439 \u043f\u043e\u0437\u0438\u0446\u0438\u0439.  \u041f\u0440\u0438\u0447\u0438\u043d\u0430 \u0432 \u0442\u043e\u043c, \u0447\u0442\u043e \u043e\u043d\u0438 \u043c\u043e\u0434\u0435\u043b\u0438\u0440\u0443\u044e\u0442 \u043f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u044b\u0435 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u044f \u0441 \u043e\u0431\u0443\u0447\u0430\u0435\u043c\u044b\u043c\u0438 \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u043c\u0438 \u0441\u043b\u043e\u044f\u043c\u0438.  \u041f\u0440\u0438 \u044d\u0442\u043e\u043c \u043a\u043b\u044e\u0447\u0435\u0432\u0430\u044f \u0438\u043d\u0436\u0435\u043d\u0435\u0440\u043d\u0430\u044f \u0447\u0430\u0441\u0442\u044c \u044d\u0442\u043e\u0439 \u0441\u0442\u0430\u0442\u044c\u0438 \u043f\u043e\u0441\u0432\u044f\u0449\u0435\u043d\u0430 \u043f\u0435\u0440\u0435\u0434\u0430\u0447\u0435 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f \u0432 \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u043e\u0432\u0430\u0442\u0435\u043b\u044c.<\/p>\n<h2 id=\"representing-an-image-as-a-sequence-of-patches\">\u041f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u0435 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f \u0432 \u0432\u0438\u0434\u0435 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438 \u043f\u0430\u0442\u0447\u0435\u0439<\/h2>\n<p>\u041c\u043d\u0435 \u0442\u0430\u043a\u0436\u0435 \u0431\u044b\u043b\u043e \u043e\u0447\u0435\u043d\u044c \u043b\u044e\u0431\u043e\u043f\u044b\u0442\u043d\u043e, \u043a\u0430\u043a \u0432\u044b \u043c\u043e\u0436\u0435\u0442\u0435 \u044d\u043b\u0435\u0433\u0430\u043d\u0442\u043d\u043e \u0438\u0437\u043c\u0435\u043d\u0438\u0442\u044c \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0432 \u043f\u0430\u0442\u0447\u0430\u0445.  \u0414\u043b\u044f \u0432\u0445\u043e\u0434\u043d\u043e\u0433\u043e \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mtext mathvariant=\"bold\">(<\/mtext><mi>\u0418\u043a\u0441<\/mi><mo stretchy=\"false\">)<\/mo><mo>\u0435<\/mo><msup><mi>\u0440<\/mi><mrow><mi>\u0427\u0410\u0421<\/mi><mo>\u00d7<\/mo><mi>\u0412\u0442<\/mi><mo>\u00d7<\/mo><mi>\u0421<\/mi><\/mrow><\/msup><\/mrow><annotation encoding=\"application\/x-tex\">\\textbf(x) \\in R^{H \\times W \\times C}<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:1em;vertical-align:-0.25em\"\/><span class=\"mord text\"><span class=\"mord textbf\">(<\/span><\/span><span class=\"mord mathnormal\">\u0418\u043a\u0441<\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><span class=\"mrel\">\u0435<\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><\/span><span class=\"base\"><span class=\"strut\" style=\"height:0.8413309999999999em;vertical-align:0em\"\/><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right:0.00773em\">\u0440<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.8413309999999999em\"><span style=\"top:-3.063em;margin-right:0.05em\"><span class=\"pstrut\" style=\"height:2.7em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right:0.08125em\">\u0427\u0410\u0421<\/span><span class=\"mbin mtight\">\u00d7<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right:0.13889em\">\u0412\u0442<\/span><span class=\"mbin mtight\">\u00d7<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right:0.07153em\">\u0421<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>  \u0438 \u0440\u0430\u0437\u043c\u0435\u0440 \u043f\u0430\u0442\u0447\u0430 <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u043f<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u043f<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.625em;vertical-align:-0.19444em\"\/><span class=\"mord mathnormal\">\u043f<\/span><\/span><\/span><\/span><\/span>\u043c\u044b \u0445\u043e\u0442\u0438\u043c \u0441\u043e\u0437\u0434\u0430\u0442\u044c <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u041d<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u041d<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.10903em\">\u041d<\/span><\/span><\/span><\/span><\/span>  \u043f\u0430\u0442\u0447\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f, \u043e\u0431\u043e\u0437\u043d\u0430\u0447\u0430\u0435\u043c\u044b\u0435 \u043a\u0430\u043a <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mtext mathvariant=\"bold\">(<\/mtext><mi>\u0418\u043a\u0441<\/mi><msub><mo stretchy=\"false\">)<\/mo><mi>\u043f<\/mi><\/msub><mo>\u0435<\/mo><msup><mi>\u0440<\/mi><mrow><mi>\u041d<\/mi><mo>\u00d7<\/mo><mo stretchy=\"false\">(<\/mo><msup><mi>\u043f<\/mi><mn>2<\/mn><\/msup><mi>\u0421<\/mi><mo stretchy=\"false\">)<\/mo><\/mrow><\/msup><\/mrow><annotation encoding=\"application\/x-tex\">\\textbf(x)_p \\in R^{N \\times (P^{2} C)}<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:1.036108em;vertical-align:-0.286108em\"\/><span class=\"mord text\"><span class=\"mord textbf\">(<\/span><\/span><span class=\"mord mathnormal\">\u0418\u043a\u0441<\/span><span class=\"mclose\"><span class=\"mclose\">)<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.15139200000000003em\"><span style=\"top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em\"><span class=\"pstrut\" style=\"height:2.7em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">\u043f<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.286108em\"><span\/><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><span class=\"mrel\">\u0435<\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><\/span><span class=\"base\"><span class=\"strut\" style=\"height:0.9869199999999998em;vertical-align:0em\"\/><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right:0.00773em\">\u0440<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.9869199999999998em\"><span style=\"top:-3.063em;margin-right:0.05em\"><span class=\"pstrut\" style=\"height:2.7em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right:0.10903em\">\u041d<\/span><span class=\"mbin mtight\">\u00d7<\/span><span class=\"mopen mtight\">(<\/span><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right:0.13889em\">\u043f<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.8913142857142857em\"><span style=\"top:-2.931em;margin-right:0.07142857142857144em\"><span class=\"pstrut\" style=\"height:2.5em\"\/><span class=\"sizing reset-size3 size1 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal mtight\" style=\"margin-right:0.07153em\">\u0421<\/span><span class=\"mclose mtight\">)<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u0433\u0434\u0435 <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u041d<\/mi><mo>&#8220;=&#8221;<\/mo><mfrac><mrow><mi>\u0427\u0410\u0421<\/mi><mi>\u0412\u0442<\/mi><\/mrow><msup><mi>\u043f<\/mi><mn>2<\/mn><\/msup><\/mfrac><\/mrow><annotation encoding=\"application\/x-tex\">N = \\ \u0433\u0438\u0434\u0440\u043e\u0440\u0430\u0437\u0440\u044b\u0432 {HW} {P ^ {2}}<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.10903em\">\u041d<\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><span class=\"mrel\">&#8220;=&#8221;<\/span><span class=\"mspace\" style=\"margin-right:0.2777777777777778em\"\/><\/span><span class=\"base\"><span class=\"strut\" style=\"height:1.217331em;vertical-align:-0.345em\"\/><span class=\"mord\"><span class=\"mopen nulldelimiter\"\/><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.872331em\"><span style=\"top:-2.6550000000000002em\"><span class=\"pstrut\" style=\"height:3em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right:0.13889em\">\u043f<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.7463142857142857em\"><span style=\"top:-2.786em;margin-right:0.07142857142857144em\"><span class=\"pstrut\" style=\"height:2.5em\"\/><span class=\"sizing reset-size3 size1 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span style=\"top:-3.23em\"><span class=\"pstrut\" style=\"height:3em\"\/><span class=\"frac-line\" style=\"border-bottom-width:0.04em\"\/><\/span><span style=\"top:-3.394em\"><span class=\"pstrut\" style=\"height:3em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right:0.08125em\">\u0427\u0410\u0421<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right:0.13889em\">\u0412\u0442<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.345em\"><span\/><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"\/><\/span><\/span><\/span><\/span><\/span>. <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u041d<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u041d<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.10903em\">\u041d<\/span><\/span><\/span><\/span><\/span>  \u0434\u043b\u0438\u043d\u0430 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438 \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u0430 \u0441\u043b\u043e\u0432\u0430\u043c \u0432 \u043f\u0440\u0435\u0434\u043b\u043e\u0436\u0435\u043d\u0438\u0438.<\/p>\n<p>\u0415\u0441\u043b\u0438 \u0432\u044b \u043d\u0435 \u0437\u0430\u043c\u0435\u0442\u0438\u043b\u0438 \u043f\u0430\u0442\u0447 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f, \u0442.\u0435. <!-- -->[16,16,3]<!-- -->  \u0441\u043f\u043b\u044e\u0449\u0435\u043d \u0434\u043e 16x16x3.  \u041d\u0430\u0434\u0435\u044e\u0441\u044c, \u043d\u0430\u0437\u0432\u0430\u043d\u0438\u0435 \u0443\u0436\u0435 \u0438\u043c\u0435\u0435\u0442 \u0441\u043c\u044b\u0441\u043b \ud83d\ude09<\/p>\n<p>\u042f \u0431\u0443\u0434\u0443 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u0442\u044c \u0431\u0438\u0431\u043b\u0438\u043e\u0442\u0435\u043a\u0443 einops, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442 \u043f\u043e\u0432\u0435\u0440\u0445 PyTorch.  \u0412\u044b \u043c\u043e\u0436\u0435\u0442\u0435 \u0443\u0441\u0442\u0430\u043d\u043e\u0432\u0438\u0442\u044c \u0435\u0433\u043e \u0447\u0435\u0440\u0435\u0437 pip:<\/p>\n<pre class=\"prism-code language-\" style=\"color:#F8F8F2;background-color:#282A36\"><p><span class=\"token plain\">$ pip install einops<\/span><\/p><\/pre>\n<p>\u0410 \u0437\u0430\u0442\u0435\u043c \u043d\u0435\u043c\u043d\u043e\u0433\u043e \u043a\u043e\u043c\u043f\u0430\u043a\u0442\u043d\u043e\u0433\u043e \u043a\u043e\u0434\u0430 Pytorch:<\/p>\n<pre class=\"prism-code language-python\" style=\"color:#F8F8F2;background-color:#282A36\"><p><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">from<\/span><span class=\"token plain\"> einops <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">import<\/span><span class=\"token plain\"> rearrange<\/span><\/p><p><span class=\"token plain\">p <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> patch_size <\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">x_p <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> rearrange<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">img<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token string\" style=\"color:rgb(255, 121, 198)\">'b c (h p1) (w p2) -&gt; b (h w) (p1 p2 c)'<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> p1 <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> p<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> p2 <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> p<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><\/p><\/pre>\n<p>\u041a\u043e\u0440\u043e\u0447\u0435 \u0433\u043e\u0432\u043e\u0440\u044f, \u043a\u0430\u0436\u0434\u044b\u0439 \u0441\u0438\u043c\u0432\u043e\u043b \u0438\u043b\u0438 \u043a\u0430\u0436\u0434\u0430\u044f \u0441\u043a\u043e\u0431\u043a\u0430 \u0443\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u043d\u0430 \u0438\u0437\u043c\u0435\u0440\u0435\u043d\u0438\u0435.  \u0414\u043b\u044f \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u0438\u044f \u0434\u043e\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u043e\u0431 \u043e\u043f\u0435\u0440\u0430\u0446\u0438\u044f\u0445 einsum \u043e\u0437\u043d\u0430\u043a\u043e\u043c\u044c\u0442\u0435\u0441\u044c \u0441 \u043d\u0430\u0448\u0438\u043c \u0441\u043e\u043e\u0431\u0449\u0435\u043d\u0438\u0435\u043c \u0432 \u0431\u043b\u043e\u0433\u0435 \u043e\u0431 \u043e\u043f\u0435\u0440\u0430\u0446\u0438\u044f\u0445 einsum.<\/p>\n<p>\u041e\u0431\u0440\u0430\u0442\u0438\u0442\u0435 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u0447\u0442\u043e \u0434\u043b\u044f \u043f\u0440\u043e\u0441\u0442\u043e\u0442\u044b \u043f\u0430\u0442\u0447\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f \u0432\u0441\u0435\u0433\u0434\u0430 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u044b\u0435.<\/p>\n<p>\u0410 \u043a\u0430\u043a \u043d\u0430\u0441\u0447\u0435\u0442 \u043f\u0435\u0440\u0435\u0445\u043e\u0434\u0430 \u043e\u0442 \u043f\u0430\u0442\u0447\u0430 \u043a \u0432\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u043d\u0438\u044e?  \u042d\u0442\u043e \u043f\u0440\u043e\u0441\u0442\u043e \u0441\u043b\u043e\u0439 \u043b\u0438\u043d\u0435\u0439\u043d\u043e\u0433\u043e \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u043e\u0432\u0430\u043d\u0438\u044f, \u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0435\u0442 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u044c <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><msup><mi>\u043f<\/mi><mn>2<\/mn><\/msup><mi>\u0421<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u041f^{2} \u0421<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.8141079999999999em;vertical-align:0em\"\/><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right:0.13889em\">\u043f<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.8141079999999999em\"><span style=\"top:-3.063em;margin-right:0.05em\"><span class=\"pstrut\" style=\"height:2.7em\"\/><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\" style=\"margin-right:0.07153em\">\u0421<\/span><\/span><\/span><\/span><\/span>  \u044d\u043b\u0435\u043c\u0435\u043d\u0442\u044b \u0438 \u0432\u044b\u0445\u043e\u0434\u044b <span class=\"inlineMath\"><span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http:\/\/www.w3.org\/1998\/Math\/MathML\"><semantics><mrow><mi>\u0414<\/mi><\/mrow><annotation encoding=\"application\/x-tex\">\u0414<\/annotation><\/semantics><\/math><\/span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.68333em;vertical-align:0em\"\/><span class=\"mord mathnormal\" style=\"margin-right:0.02778em\">\u0414<\/span><\/span><\/span><\/span><\/span>.<\/p>\n<pre class=\"prism-code language-python\" style=\"color:#F8F8F2;background-color:#282A36\"><p><span class=\"token plain\">patch_dim <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">patch_size<\/span><span class=\"token operator\">**<\/span><span class=\"token number\">2<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"> <\/span><span class=\"token operator\">*<\/span><span class=\"token plain\"> channels <\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">patch_to_embedding <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Linear<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">patch_dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><\/p><\/pre>\n<p>\u0412\u044b \u0432\u0438\u0434\u0438\u0442\u0435, \u0447\u0435\u0433\u043e \u043d\u0435 \u0445\u0432\u0430\u0442\u0430\u0435\u0442?<\/p>\n<p>\u0411\u044c\u044e\u0441\u044c \u043e\u0431 \u0437\u0430\u043a\u043b\u0430\u0434, \u0432\u044b!  \u041d\u0430\u043c \u043d\u0443\u0436\u043d\u043e \u043e\u0431\u0435\u0441\u043f\u0435\u0447\u0438\u0442\u044c \u043a\u0430\u043a\u043e\u0439-\u0442\u043e \u043f\u043e\u0440\u044f\u0434\u043e\u043a.<\/p>\n<h2 id=\"positional-embeddings\">\u041f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u044b\u0435 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u044f<\/h2>\n<p>\u041d\u0435\u0441\u043c\u043e\u0442\u0440\u044f \u043d\u0430 \u0442\u043e, \u0447\u0442\u043e \u0431\u044b\u043b\u043e \u043f\u0440\u0438\u043c\u0435\u043d\u0435\u043d\u043e \u043c\u043d\u043e\u0436\u0435\u0441\u0442\u0432\u043e \u0441\u0445\u0435\u043c \u043f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u043e\u0433\u043e \u0432\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u043d\u0438\u044f, \u0441\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0439 \u0440\u0430\u0437\u043d\u0438\u0446\u044b \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u043e \u043d\u0435 \u0431\u044b\u043b\u043e.  \u0412\u0435\u0440\u043e\u044f\u0442\u043d\u043e, \u044d\u0442\u043e \u0441\u0432\u044f\u0437\u0430\u043d\u043e \u0441 \u0442\u0435\u043c, \u0447\u0442\u043e \u043a\u043e\u0434\u0435\u0440-\u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440 \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442 \u043d\u0430 \u0443\u0440\u043e\u0432\u043d\u0435 \u043f\u0430\u0442\u0447\u0435\u0439.  \u0418\u0437\u0443\u0447\u0435\u043d\u0438\u0435 \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u0439, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0444\u0438\u043a\u0441\u0438\u0440\u0443\u044e\u0442 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u044f \u043f\u043e\u0440\u044f\u0434\u043a\u0430 \u043c\u0435\u0436\u0434\u0443 \u043f\u0430\u0442\u0447\u0430\u043c\u0438 (\u043f\u0440\u043e\u0441\u0442\u0440\u0430\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u0430\u044f \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u044f), \u043d\u0435 \u0442\u0430\u043a \u0432\u0430\u0436\u043d\u043e.  \u041e\u0442\u043d\u043e\u0441\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u043b\u0435\u0433\u0447\u0435 \u043f\u043e\u043d\u044f\u0442\u044c \u0432\u0437\u0430\u0438\u043c\u043e\u0441\u0432\u044f\u0437\u044c \u043c\u0435\u0436\u0434\u0443 \u0444\u0440\u0430\u0433\u043c\u0435\u043d\u0442\u0430\u043c\u0438 P x P, \u0447\u0435\u043c \u043c\u0435\u0436\u0434\u0443 \u0432\u044b\u0441\u043e\u0442\u043e\u0439 x \u0448\u0438\u0440\u0438\u043d\u043e\u0439 \u043f\u043e\u043b\u043d\u043e\u0433\u043e \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f.<\/p>\n<blockquote>\n<p>\u0418\u043d\u0442\u0443\u0438\u0442\u0438\u0432\u043d\u043e \u0432\u044b \u043c\u043e\u0436\u0435\u0442\u0435 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u0438\u0442\u044c \u0440\u0435\u0448\u0435\u043d\u0438\u0435 \u0433\u043e\u043b\u043e\u0432\u043e\u043b\u043e\u043c\u043a\u0438 \u0438\u0437 100 \u0447\u0430\u0441\u0442\u0435\u0439 (\u043f\u0430\u0442\u0447\u0435\u0439) \u0432\u043c\u0435\u0441\u0442\u043e 5000 \u0447\u0430\u0441\u0442\u0435\u0439 (\u043f\u0438\u043a\u0441\u0435\u043b\u0435\u0439).<\/p>\n<\/blockquote>\n<p>\u0421\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e, \u043f\u043e\u0441\u043b\u0435 \u043d\u0438\u0437\u043a\u043e\u0440\u0430\u0437\u043c\u0435\u0440\u043d\u043e\u0439 \u043b\u0438\u043d\u0435\u0439\u043d\u043e\u0439 \u043f\u0440\u043e\u0435\u043a\u0446\u0438\u0438 <strong>\u043e\u0431\u0443\u0447\u0430\u0435\u043c\u044b\u0439<\/strong> \u0432\u043b\u043e\u0436\u0435\u043d\u0438\u0435 \u043f\u043e\u0437\u0438\u0446\u0438\u0438 \u0434\u043e\u0431\u0430\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u043a \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f\u043c \u043f\u0430\u0442\u0447\u0435\u0439.  \u0418\u043d\u0442\u0435\u0440\u0435\u0441\u043d\u043e \u043f\u043e\u0441\u043c\u043e\u0442\u0440\u0435\u0442\u044c, \u043a\u0430\u043a \u0432\u044b\u0433\u043b\u044f\u0434\u044f\u0442 \u044d\u0442\u0438 \u044d\u043c\u0431\u0435\u0434\u0434\u0438\u043d\u0433\u0438 \u043f\u043e\u0437\u0438\u0446\u0438\u0439 \u043f\u043e\u0441\u043b\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f:<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:459px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:89.99999999999999%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,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');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u043f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u043e\u0435-\u043a\u043e\u0434\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435-vit\" title=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u043f\u043e\u0437\u0438\u0446\u0438\u043e\u043d\u043d\u043e\u0435-\u043a\u043e\u0434\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435-vit\" src=\"https:\/\/theaisummer.com\/static\/327309a892d82b4f59eb07818ee28ee6\/5a46d\/visualizing-positional-encodings-vit.png 300w,\/static\/327309a892d82b4f59eb07818ee28ee6\/48711\/visualizing-positional-encodings-vit.png 459w\" sizes=\"(max-width: 459px) 100vw, 459px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u0414\u043e\u0441\u043e\u0432\u0438\u0446\u043a\u0438\u0439 \u0438 \u0434\u0440. 2020. \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0418\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16&#215;16 \u0441\u043b\u043e\u0432: \u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u044b \u0434\u043b\u044f \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0435<\/em><\/p>\n<p>\u0412\u043e-\u043f\u0435\u0440\u0432\u044b\u0445, \u0435\u0441\u0442\u044c \u043a\u0430\u043a\u0430\u044f-\u0442\u043e \u0434\u0432\u0443\u043c\u0435\u0440\u043d\u0430\u044f \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u0430.  \u0412\u043e-\u0432\u0442\u043e\u0440\u044b\u0445, \u0448\u0430\u0431\u043b\u043e\u043d\u044b \u0432 \u0441\u0442\u0440\u043e\u043a\u0430\u0445 (\u0438 \u0441\u0442\u043e\u043b\u0431\u0446\u0430\u0445) \u0438\u043c\u0435\u044e\u0442 \u0441\u0445\u043e\u0436\u0438\u0435 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f.  \u0414\u043b\u044f \u0432\u044b\u0441\u043e\u043a\u0438\u0445 \u0440\u0430\u0437\u0440\u0435\u0448\u0435\u043d\u0438\u0439 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043b\u0430\u0441\u044c \u0441\u0438\u043d\u0443\u0441\u043e\u0438\u0434\u0430\u043b\u044c\u043d\u0430\u044f \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u0430.<\/p>\n<h2 id=\"key-findings\">\u041a\u043b\u044e\u0447\u0435\u0432\u044b\u0435 \u0440\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442\u044b<\/h2>\n<p>\u0412 \u043f\u0435\u0440\u0432\u044b\u0435 \u0434\u043d\u0438 \u043a\u043e\u043d\u0432\u0435\u0440\u0441\u0438\u0438 \u043c\u044b \u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0438\u0440\u043e\u0432\u0430\u043b\u0438 \u0440\u0430\u043d\u043d\u0438\u0435 \u0441\u043b\u043e\u0438.<\/p>\n<p>\u041f\u043e\u0447\u0435\u043c\u0443?<\/p>\n<p>\u041f\u043e\u0441\u043a\u043e\u043b\u044c\u043a\u0443 \u043c\u044b \u0441\u0447\u0438\u0442\u0430\u0435\u043c, \u0447\u0442\u043e \u0445\u043e\u0440\u043e\u0448\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u043d\u044b\u0435 \u0441\u0435\u0442\u0438 \u0447\u0430\u0441\u0442\u043e \u043f\u043e\u043a\u0430\u0437\u044b\u0432\u0430\u044e\u0442 \u0445\u043e\u0440\u043e\u0448\u0438\u0435 \u0438 <strong>\u0433\u043b\u0430\u0434\u043a\u0438\u0439<\/strong> \u0444\u0438\u043b\u044c\u0442\u0440\u044b.<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:859px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:46.333333333333336%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,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');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u043a\u043e\u043d\u0432-\u0444\u0438\u043b\u044c\u0442\u0440\u044b-VS-\u0432\u0438\u0442\" title=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u043a\u043e\u043d\u0432-\u0444\u0438\u043b\u044c\u0442\u0440\u044b-VS-\u0432\u0438\u0442\" src=\"https:\/\/theaisummer.com\/static\/1a825502d30ae8a5fde3386057ed1ef5\/5a46d\/visualizing-conv-filters-vs-vit.png 300w,\/static\/1a825502d30ae8a5fde3386057ed1ef5\/0a47e\/visualizing-conv-filters-vs-vit.png 600w,\/static\/1a825502d30ae8a5fde3386057ed1ef5\/39a20\/visualizing-conv-filters-vs-vit.png 859w\" sizes=\"(max-width: 859px) 100vw, 859px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0421\u043b\u0435\u0432\u0430: \u0412\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f \u0444\u0438\u043b\u044c\u0442\u0440\u043e\u0432 Alexnet.  \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0421\u0442\u044d\u043d\u0434\u0444\u043e\u0440\u0434\u0441\u043a\u0438\u0439 \u043a\u0443\u0440\u0441 CS231n. \u0421\u043f\u0440\u0430\u0432\u0430: ViT \u0432\u044b\u0443\u0447\u0438\u043b \u0444\u0438\u043b\u044c\u0442\u0440\u044b.  \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16&#215;16 \u0441\u043b\u043e\u0432: \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u044b \u0434\u043b\u044f \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0435<\/em><\/p>\n<p>\u042f \u043f\u043e\u0437\u0430\u0438\u043c\u0441\u0442\u0432\u043e\u0432\u0430\u043b \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0438\u0437 \u0421\u0442\u044d\u043d\u0444\u043e\u0440\u0434\u0441\u043a\u043e\u0433\u043e \u043a\u0443\u0440\u0441\u0430 CS231n: \u0441\u0432\u0435\u0440\u0442\u043e\u0447\u043d\u044b\u0435 \u043d\u0435\u0439\u0440\u043e\u043d\u043d\u044b\u0435 \u0441\u0435\u0442\u0438 \u0434\u043b\u044f \u0432\u0438\u0437\u0443\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f.<\/p>\n<p>\u041a\u0430\u043a \u043f\u0440\u0435\u043a\u0440\u0430\u0441\u043d\u043e \u0441\u043a\u0430\u0437\u0430\u043d\u043e \u0432 CS231n:<\/p>\n<blockquote>\n<p>\u00ab\u041e\u0431\u0440\u0430\u0442\u0438\u0442\u0435 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u0447\u0442\u043e \u0432\u0435\u0441\u0430 \u043f\u0435\u0440\u0432\u043e\u0433\u043e \u0441\u043b\u043e\u044f \u043e\u0447\u0435\u043d\u044c \u0445\u043e\u0440\u043e\u0448\u0438\u0435 \u0438 \u0433\u043b\u0430\u0434\u043a\u0438\u0435, \u0447\u0442\u043e \u0443\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u043d\u0430 \u0445\u043e\u0440\u043e\u0448\u043e \u0441\u0445\u043e\u0434\u044f\u0449\u0443\u044e\u0441\u044f \u0441\u0435\u0442\u044c.  \u0424\u0443\u043d\u043a\u0446\u0438\u0438 \u0446\u0432\u0435\u0442\u0430\/\u043e\u0442\u0442\u0435\u043d\u043a\u043e\u0432 \u0441\u0435\u0440\u043e\u0433\u043e \u0441\u0433\u0440\u0443\u043f\u043f\u0438\u0440\u043e\u0432\u0430\u043d\u044b, \u043f\u043e\u0441\u043a\u043e\u043b\u044c\u043a\u0443 AlexNet \u0441\u043e\u0434\u0435\u0440\u0436\u0438\u0442 <strong>\u0434\u0432\u0430 \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u044b\u0445 \u043f\u043e\u0442\u043e\u043a\u0430 \u043e\u0431\u0440\u0430\u0431\u043e\u0442\u043a\u0438<\/strong>, \u0438 \u043e\u0447\u0435\u0432\u0438\u0434\u043d\u044b\u043c \u0441\u043b\u0435\u0434\u0441\u0442\u0432\u0438\u0435\u043c \u044d\u0442\u043e\u0439 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u044b \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0442\u043e, \u0447\u0442\u043e \u043e\u0434\u0438\u043d \u043f\u043e\u0442\u043e\u043a \u0440\u0430\u0437\u0432\u0438\u0432\u0430\u0435\u0442 \u0432\u044b\u0441\u043e\u043a\u043e\u0447\u0430\u0441\u0442\u043e\u0442\u043d\u044b\u0435 \u0444\u0443\u043d\u043a\u0446\u0438\u0438 \u043e\u0442\u0442\u0435\u043d\u043a\u043e\u0432 \u0441\u0435\u0440\u043e\u0433\u043e, \u0430 \u0434\u0440\u0443\u0433\u043e\u0439 \u2014 \u043d\u0438\u0437\u043a\u043e\u0447\u0430\u0441\u0442\u043e\u0442\u043d\u044b\u0435 \u0446\u0432\u0435\u0442\u043e\u0432\u044b\u0435 \u0444\u0443\u043d\u043a\u0446\u0438\u0438\u00bb.  ~ \u0421\u0442\u044d\u043d\u0444\u043e\u0440\u0434\u0441\u043a\u0438\u0439 \u043a\u0443\u0440\u0441 CS231: \u0412\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f \u0442\u043e\u0433\u043e, \u0447\u0435\u043c\u0443 \u0443\u0447\u0430\u0442 ConvNets<\/p>\n<\/blockquote>\n<p>\u0414\u043b\u044f \u0442\u0430\u043a\u0438\u0445 \u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u0439 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u0435\u0442\u0441\u044f PCA.  \u0422\u0430\u043a\u0438\u043c \u043e\u0431\u0440\u0430\u0437\u043e\u043c, \u0430\u0432\u0442\u043e\u0440 \u043f\u043e\u043a\u0430\u0437\u0430\u043b, \u0447\u0442\u043e \u0440\u0430\u043d\u043d\u0438\u0435 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0441\u043b\u043e\u0435\u0432 \u043c\u043e\u0433\u0443\u0442 <strong>\u0438\u043c\u0435\u0442\u044c \u0441\u0445\u043e\u0436\u0438\u0435 \u0447\u0435\u0440\u0442\u044b<\/strong>.<\/p>\n<p>\u0421\u043b\u0435\u0434\u0443\u044e\u0449\u0438\u0439 \u0432\u043e\u043f\u0440\u043e\u0441, \u043f\u043e\u0436\u0430\u043b\u0443\u0439\u0441\u0442\u0430.<\/p>\n<h2 id=\"how-far-aways-are-the-learned-non-local-interactions\">\u041a\u0430\u043a \u0434\u0430\u043b\u0435\u043a\u043e \u043d\u0430\u0445\u043e\u0434\u044f\u0442\u0441\u044f \u0438\u0437\u0443\u0447\u0435\u043d\u043d\u044b\u0435 \u043d\u0435\u043b\u043e\u043a\u0430\u043b\u044c\u043d\u044b\u0435 \u0432\u0437\u0430\u0438\u043c\u043e\u0434\u0435\u0439\u0441\u0442\u0432\u0438\u044f?<\/h2>\n<p><strong>\u041a\u043e\u0440\u043e\u0442\u043a\u0438\u0439 \u043e\u0442\u0432\u0435\u0442<\/strong>: \u0414\u043b\u044f \u0440\u0430\u0437\u043c\u0435\u0440\u0430 \u043f\u0430\u0442\u0447\u0430 P, \u043c\u0430\u043a\u0441\u0438\u043c\u0443\u043c P<!-- -->*<!-- -->P, \u0447\u0442\u043e \u0432 \u043d\u0430\u0448\u0435\u043c \u0441\u043b\u0443\u0447\u0430\u0435 128, \u0434\u0430\u0436\u0435 \u0441 1-\u0433\u043e \u0441\u043b\u043e\u044f!<\/p>\n<p>\u041d\u0430\u043c \u043d\u0435 \u043d\u0443\u0436\u043d\u044b \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u044b\u0435 \u043a\u043e\u043d\u0432.  \u0441\u043b\u043e\u0438, \u0447\u0442\u043e\u0431\u044b \u0434\u043e\u0431\u0440\u0430\u0442\u044c\u0441\u044f \u0434\u043e 128 \u043f\u0438\u043a\u0441\u0435\u043b\u0435\u0439 \u0431\u043e\u043b\u044c\u0448\u0435.  \u041f\u0440\u0438 \u0438\u0437\u0432\u0438\u043b\u0438\u043d\u0430\u0445 \u0431\u0435\u0437 \u0434\u0438\u043b\u0430\u0442\u0430\u0446\u0438\u0438 \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u043e\u0435 \u043f\u043e\u043b\u0435 \u0443\u0432\u0435\u043b\u0438\u0447\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043b\u0438\u043d\u0435\u0439\u043d\u043e.  \u0418\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u044f \u0441\u0430\u043c\u043e-\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u043c\u044b \u0438\u043c\u0435\u0435\u043c \u0432\u0437\u0430\u0438\u043c\u043e\u0434\u0435\u0439\u0441\u0442\u0432\u0438\u0435 \u043c\u0435\u0436\u0434\u0443 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f\u043c\u0438 \u043f\u0438\u043a\u0441\u0435\u043b\u0435\u0439 \u0432 1-\u043c \u0441\u043b\u043e\u0435 \u0438 \u043f\u0430\u0440\u0430\u043c\u0438 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u0439 \u0432\u043e 2-\u043c \u0441\u043b\u043e\u0435 \u0438 \u0442\u0430\u043a \u0434\u0430\u043b\u0435\u0435.<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:1200px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:37.666666666666664%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,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');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0432\u0438\u0442-\u0433\u043e\u043b\u043e\u0432\u044b-\u0441\u0440\u0435\u0434\u043d\u0435\u0435-\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435-\u0440\u0430\u0441\u0441\u0442\u043e\u044f\u043d\u0438\u0435-\u043f\u0440\u043e\u0442\u0438\u0432-\u0438\u0437\u0432\u0438\u043b\u0438\u043d\" title=\"\u0432\u0438\u0442-\u0433\u043e\u043b\u043e\u0432\u044b-\u0441\u0440\u0435\u0434\u043d\u0435\u0435-\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435-\u0440\u0430\u0441\u0441\u0442\u043e\u044f\u043d\u0438\u0435-\u043f\u0440\u043e\u0442\u0438\u0432-\u0438\u0437\u0432\u0438\u043b\u0438\u043d\" src=\"https:\/\/theaisummer.com\/static\/1298903e0744cbb786ae7313bf6e42c2\/5a46d\/vit-heads-mean-attention-distance-vs-convolutions.png 300w,\/static\/1298903e0744cbb786ae7313bf6e42c2\/0a47e\/vit-heads-mean-attention-distance-vs-convolutions.png 600w,\/static\/1298903e0744cbb786ae7313bf6e42c2\/c1b63\/vit-heads-mean-attention-distance-vs-convolutions.png 1200w,\/static\/1298903e0744cbb786ae7313bf6e42c2\/d61c2\/vit-heads-mean-attention-distance-vs-convolutions.png 1800w,\/static\/1298903e0744cbb786ae7313bf6e42c2\/10b63\/vit-heads-mean-attention-distance-vs-convolutions.png 1897w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0421\u043f\u0440\u0430\u0432\u0430: \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435, \u0441\u043e\u0437\u0434\u0430\u043d\u043d\u043e\u0435 \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u043a\u0430\u043b\u044c\u043a\u0443\u043b\u044f\u0442\u043e\u0440\u0430 \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430 Fomoro. \u0421\u043b\u0435\u0432\u0430: \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0410\u043b\u0435\u043a\u0441\u0435\u044f \u0414\u043e\u0441\u043e\u0432\u0438\u0446\u043a\u043e\u0433\u043e \u0438 \u0434\u0440. 2020.<\/em><\/p>\n<p>\u0418\u0441\u0445\u043e\u0434\u044f \u0438\u0437 \u0434\u0438\u0430\u0433\u0440\u0430\u043c\u043c\u044b \u0441\u043b\u0435\u0432\u0430 \u043e\u0442 ViT, \u043c\u043e\u0436\u043d\u043e \u0443\u0442\u0432\u0435\u0440\u0436\u0434\u0430\u0442\u044c, \u0447\u0442\u043e:<\/p>\n<ul>\n<li>\n<p>\u0414\u0435\u0439\u0441\u0442\u0432\u0438\u0442\u0435\u043b\u044c\u043d\u043e, \u0435\u0441\u0442\u044c \u0433\u043e\u043b\u043e\u0432\u044b, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0437\u0430\u043d\u0438\u043c\u0430\u044e\u0442\u0441\u044f \u0432\u0441\u0435\u043c \u043f\u0430\u0442\u0447\u0435\u043c \u0443\u0436\u0435 \u0432 \u0440\u0430\u043d\u043d\u0438\u0445 \u0441\u043b\u043e\u044f\u0445.<\/p>\n<\/li>\n<li>\n<p>\u041c\u043e\u0436\u043d\u043e \u043e\u043f\u0440\u0430\u0432\u0434\u0430\u0442\u044c \u043f\u0440\u0438\u0440\u043e\u0441\u0442 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438 \u043d\u0430 \u043e\u0441\u043d\u043e\u0432\u0435 \u0432\u0437\u0430\u0438\u043c\u043e\u0434\u0435\u0439\u0441\u0442\u0432\u0438\u044f \u043f\u0438\u043a\u0441\u0435\u043b\u0435\u0439 \u0440\u0430\u043d\u043d\u0435\u0433\u043e \u0434\u043e\u0441\u0442\u0443\u043f\u0430.  \u041a\u0430\u0436\u0435\u0442\u0441\u044f \u0431\u043e\u043b\u0435\u0435 \u0432\u0430\u0436\u043d\u044b\u043c, \u0447\u0442\u043e\u0431\u044b \u0440\u0430\u043d\u043d\u0438\u0435 \u0441\u043b\u043e\u0438 \u0438\u043c\u0435\u043b\u0438 \u0434\u043e\u0441\u0442\u0443\u043f \u043a\u043e \u0432\u0441\u0435\u043c\u0443 \u043f\u0430\u0442\u0447\u0443 (\u0433\u043b\u043e\u0431\u0430\u043b\u044c\u043d\u0430\u044f \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u044f).  \u0414\u0440\u0443\u0433\u0438\u043c\u0438 \u0441\u043b\u043e\u0432\u0430\u043c\u0438, \u0433\u043e\u043b\u043e\u0432\u044b, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u043f\u0440\u0438\u043d\u0430\u0434\u043b\u0435\u0436\u0430\u0442 \u0432\u0435\u0440\u0445\u043d\u0435\u0439 \u043b\u0435\u0432\u043e\u0439 \u0447\u0430\u0441\u0442\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f, \u043c\u043e\u0433\u0443\u0442 \u0431\u044b\u0442\u044c \u043e\u0441\u043d\u043e\u0432\u043d\u043e\u0439 \u043f\u0440\u0438\u0447\u0438\u043d\u043e\u0439 \u043f\u0440\u0435\u0432\u043e\u0441\u0445\u043e\u0434\u043d\u043e\u0439 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438.<\/p>\n<\/li>\n<li>\n<p>\u0418\u043d\u0442\u0435\u0440\u0435\u0441\u043d\u043e, \u0447\u0442\u043e \u0434\u0438\u0441\u0442\u0430\u043d\u0446\u0438\u044f \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0443\u0432\u0435\u043b\u0438\u0447\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u0441 \u0433\u043b\u0443\u0431\u0438\u043d\u043e\u0439 \u0441\u0435\u0442\u0438, \u043f\u043e\u0434\u043e\u0431\u043d\u043e \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u043e\u043c\u0443 \u043f\u043e\u043b\u044e \u043b\u043e\u043a\u0430\u043b\u044c\u043d\u044b\u0445 \u043e\u043f\u0435\u0440\u0430\u0446\u0438\u0439.<\/p>\n<\/li>\n<li>\n<p>\u0412 \u043d\u0438\u0436\u043d\u0438\u0445 \u0441\u043b\u043e\u044f\u0445 \u0442\u0430\u043a\u0436\u0435 \u0435\u0441\u0442\u044c \u0433\u043e\u043b\u043e\u0432\u044b \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0441 \u043f\u043e\u0441\u0442\u043e\u044f\u043d\u043d\u043e \u043d\u0435\u0431\u043e\u043b\u044c\u0448\u043e\u0439 \u0434\u0438\u0441\u0442\u0430\u043d\u0446\u0438\u0435\u0439 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f.  \u0421\u043f\u0440\u0430\u0432\u0430 24-\u0441\u043b\u043e\u0439\u043d\u044b\u0439 \u0441\u043b\u043e\u0439 \u0441\u043e \u0441\u0442\u0430\u043d\u0434\u0430\u0440\u0442\u043d\u044b\u043c\u0438 \u0438\u0437\u0432\u0438\u043b\u0438\u043d\u0430\u043c\u0438 3&#215;3 \u0438\u043c\u0435\u0435\u0442 \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u043e\u0435 \u043f\u043e\u043b\u0435 \u043c\u0435\u043d\u0435\u0435 50. \u041d\u0430\u043c \u043f\u043e\u0442\u0440\u0435\u0431\u0443\u0435\u0442\u0441\u044f \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e 50 \u0441\u0432\u0451\u0440\u043d\u0443\u0442\u044b\u0445 \u0441\u043b\u043e\u0435\u0432, \u0447\u0442\u043e\u0431\u044b \u043e\u0431\u0441\u043b\u0443\u0436\u0438\u0432\u0430\u0442\u044c \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e 100 \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u044b\u0445 \u043f\u043e\u043b\u0435\u0439, \u0431\u0435\u0437 \u0440\u0430\u0441\u0448\u0438\u0440\u0435\u043d\u0438\u044f \u0438\u043b\u0438 \u043e\u0431\u044a\u0435\u0434\u0438\u043d\u0435\u043d\u0438\u044f \u0441\u043b\u043e\u0435\u0432.<\/p>\n<\/li>\n<li>\n<p>\u0427\u0442\u043e\u0431\u044b \u0440\u0435\u0430\u043b\u0438\u0437\u043e\u0432\u0430\u0442\u044c \u044d\u0442\u0443 \u0438\u0434\u0435\u044e \u0432\u044b\u0441\u043e\u043a\u043e \u043b\u043e\u043a\u0430\u043b\u0438\u0437\u043e\u0432\u0430\u043d\u043d\u044b\u0445 \u0433\u043e\u043b\u043e\u0432\u043e\u043a \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f, \u0430\u0432\u0442\u043e\u0440\u044b \u044d\u043a\u0441\u043f\u0435\u0440\u0438\u043c\u0435\u043d\u0442\u0438\u0440\u043e\u0432\u0430\u043b\u0438 \u0441 \u0433\u0438\u0431\u0440\u0438\u0434\u043d\u044b\u043c\u0438 \u043c\u043e\u0434\u0435\u043b\u044f\u043c\u0438, \u0432 \u043a\u043e\u0442\u043e\u0440\u044b\u0445 ResNet \u043f\u0440\u0438\u043c\u0435\u043d\u044f\u0435\u0442\u0441\u044f \u0434\u043e Transformer.  \u041a\u0430\u043a \u0438 \u043e\u0436\u0438\u0434\u0430\u043b\u043e\u0441\u044c, \u043e\u043d\u0438 \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0438\u043b\u0438 \u043c\u0435\u043d\u0435\u0435 \u0441\u0438\u043b\u044c\u043d\u043e \u043b\u043e\u043a\u0430\u043b\u0438\u0437\u043e\u0432\u0430\u043d\u043d\u044b\u0435 \u0433\u043e\u043b\u043e\u0432\u044b.  \u041d\u0430\u0440\u044f\u0434\u0443 \u0441 \u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u0435\u0439 \u0444\u0438\u043b\u044c\u0442\u0440\u0430 \u043f\u0440\u0435\u0434\u043f\u043e\u043b\u0430\u0433\u0430\u0435\u0442\u0441\u044f, \u0447\u0442\u043e \u043e\u043d \u043c\u043e\u0436\u0435\u0442 \u0432\u044b\u043f\u043e\u043b\u043d\u044f\u0442\u044c \u0442\u0443 \u0436\u0435 \u0444\u0443\u043d\u043a\u0446\u0438\u044e, \u0447\u0442\u043e \u0438 \u0440\u0430\u043d\u043d\u0438\u0435 \u0441\u0432\u0435\u0440\u0442\u043e\u0447\u043d\u044b\u0435 \u0441\u043b\u043e\u0438 \u0432 CNN.<\/p>\n<\/li>\n<\/ul>\n<h2 id=\"attention-distance-and-visualization\">\u0414\u0438\u0441\u0442\u0430\u043d\u0446\u0438\u044f \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0438 \u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f<\/h2>\n<p>\u0422\u0435\u043c \u043d\u0435 \u043c\u0435\u043d\u0435\u0435, \u044f \u0441\u0447\u0438\u0442\u0430\u044e \u0432\u0430\u0436\u043d\u044b\u043c \u043f\u043e\u043d\u044f\u0442\u044c, \u043a\u0430\u043a \u043e\u043d\u0438 \u0438\u0437\u043c\u0435\u0440\u044f\u043b\u0438 \u0441\u0440\u0435\u0434\u043d\u044e\u044e \u0434\u0438\u0441\u0442\u0430\u043d\u0446\u0438\u044e \u200b\u200b\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f.  \u041e\u043d\u043e \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e \u0440\u0435\u0446\u0435\u043f\u0442\u0438\u0432\u043d\u043e\u043c\u0443 \u043f\u043e\u043b\u044e, \u043d\u043e \u043d\u0435 \u0441\u043e\u0432\u0441\u0435\u043c \u0442\u043e \u0436\u0435 \u0441\u0430\u043c\u043e\u0435.<\/p>\n<p>\u0414\u0438\u0441\u0442\u0430\u043d\u0446\u0438\u044f \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0440\u0430\u0441\u0441\u0447\u0438\u0442\u044b\u0432\u0430\u043b\u0430\u0441\u044c \u043a\u0430\u043a <strong>\u0441\u0440\u0435\u0434\u043d\u0435\u0435 \u0440\u0430\u0441\u0441\u0442\u043e\u044f\u043d\u0438\u0435 \u043c\u0435\u0436\u0434\u0443 \u043f\u0438\u043a\u0441\u0435\u043b\u0435\u043c \u0437\u0430\u043f\u0440\u043e\u0441\u0430 \u0438 \u043e\u0441\u0442\u0430\u043b\u044c\u043d\u043e\u0439 \u0447\u0430\u0441\u0442\u044c\u044e \u043f\u0430\u0442\u0447\u0430<\/strong>, \u0443\u043c\u043d\u043e\u0436\u0435\u043d\u043d\u044b\u0439 \u043d\u0430 \u0432\u0435\u0441 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f.  \u041e\u043d\u0438 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043b\u0438 128 \u043f\u0440\u0438\u043c\u0435\u0440\u043e\u0432 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0438 \u0443\u0441\u0440\u0435\u0434\u043d\u0438\u043b\u0438 \u0438\u0445 \u0440\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442\u044b.<\/p>\n<p>\u041f\u0440\u0438\u043c\u0435\u0440: \u0435\u0441\u043b\u0438 \u043f\u0438\u043a\u0441\u0435\u043b\u044c \u043d\u0430\u0445\u043e\u0434\u0438\u0442\u0441\u044f \u043d\u0430 \u0440\u0430\u0441\u0441\u0442\u043e\u044f\u043d\u0438\u0438 20 \u043f\u0438\u043a\u0441\u0435\u043b\u0435\u0439, \u0430 \u0432\u0435\u0441 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0440\u0430\u0432\u0435\u043d 0,5, \u0440\u0430\u0441\u0441\u0442\u043e\u044f\u043d\u0438\u0435 \u0440\u0430\u0432\u043d\u043e 10.<\/p>\n<p>\u041d\u0430\u043a\u043e\u043d\u0435\u0446, \u043c\u043e\u0434\u0435\u043b\u044c \u043e\u0431\u0440\u0430\u0449\u0430\u0435\u0442 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435 \u043d\u0430 \u043e\u0431\u043b\u0430\u0441\u0442\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u044f, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0441\u0435\u043c\u0430\u043d\u0442\u0438\u0447\u0435\u0441\u043a\u0438 \u0440\u0435\u043b\u0435\u0432\u0430\u043d\u0442\u043d\u044b \u0434\u043b\u044f \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438, \u043a\u0430\u043a \u043f\u043e\u043a\u0430\u0437\u0430\u043d\u043e \u043d\u0438\u0436\u0435:<\/p>\n<p><span class=\"gatsby-resp-image-wrapper\" style=\"position:relative;display:block;margin-left:auto;margin-right:auto;max-width:1200px\"><\/p>\n<p>    <span class=\"gatsby-resp-image-background-image\" style=\"padding-bottom:9%;position:relative;bottom:0;left:0;background-image:url('data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAACCAIAAADXZGvcAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAhUlEQVQI1wF6AIX\/AJaDdJeSjCsmJIWNgnaCcYWKgikpJ2ZoYqKnf6u3o1haUUdEPKu7xrzO24SNkg8QFH2DdYebOpOgbzxCJQBUPjR5c28oIR18gXttaV2Vkos1NjVoaWWFo1yLqGJVWE4uLiW\/wcK2ubyYmJg3OTyPkoKGlEuMl3RHSzOTSzSXzkK8QgAAAABJRU5ErkJggg==');background-size:cover;display:block\"\/><br \/>\n  <img decoding=\"async\" class=\"gatsby-resp-image-image\" alt=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435-vit\" title=\"\u0432\u0438\u0437\u0443\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044f-\u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435-vit\" src=\"https:\/\/theaisummer.com\/static\/f7115f622470f12aac4bbb7a20dea366\/5a46d\/visualizing-attention-vit.png 300w,\/static\/f7115f622470f12aac4bbb7a20dea366\/0a47e\/visualizing-attention-vit.png 600w,\/static\/f7115f622470f12aac4bbb7a20dea366\/c1b63\/visualizing-attention-vit.png 1200w,\/static\/f7115f622470f12aac4bbb7a20dea366\/cd78c\/visualizing-attention-vit.png 1236w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" style=\"width:100%;height:100%;margin:0;vertical-align:middle;position:absolute;top:0;left:0\" loading=\"lazy\"\/><\/p>\n<p>    <\/span><br \/>\n<em>\u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u0414\u043e\u0441\u043e\u0432\u0438\u0446\u043a\u0438\u0439 \u0438 \u0434\u0440. 2020. \u0418\u0441\u0442\u043e\u0447\u043d\u0438\u043a: \u0418\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0441\u0442\u043e\u0438\u0442 16&#215;16 \u0441\u043b\u043e\u0432: \u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u044b \u0434\u043b\u044f \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u044f \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0435<\/em><\/p>\n<h2 id=\"implementation\">\u0412\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u0435<\/h2>\n<p>\u041f\u043e\u0441\u0435\u0442\u0438\u0442\u0435 \u043d\u0430\u0448 \u0440\u0435\u043f\u043e\u0437\u0438\u0442\u043e\u0440\u0438\u0439, \u0447\u0442\u043e\u0431\u044b \u043d\u0430\u0439\u0442\u0438 \u043c\u043e\u0434\u0443\u043b\u0438 \u0441\u0430\u043c\u043e\u043a\u043e\u043d\u0442\u0440\u043e\u043b\u044f \u0434\u043b\u044f \u043a\u043e\u043c\u043f\u044c\u044e\u0442\u0435\u0440\u043d\u043e\u0433\u043e \u0437\u0440\u0435\u043d\u0438\u044f.  \u0423\u0447\u0438\u0442\u044b\u0432\u0430\u044f \u0440\u0435\u0430\u043b\u0438\u0437\u0430\u0446\u0438\u044e \u0432\u0430\u043d\u0438\u043b\u044c\u043d\u043e\u0433\u043e Transformer Encoder, ViT \u0432\u044b\u0433\u043b\u044f\u0434\u0438\u0442 \u0442\u0430\u043a \u043f\u0440\u043e\u0441\u0442\u043e:<\/p>\n<pre class=\"prism-code language-python\" style=\"color:#F8F8F2;background-color:#282A36\"><p><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">import<\/span><span class=\"token plain\"> torch<\/span><\/p><p><span class=\"token plain\"\/><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">import<\/span><span class=\"token plain\"> torch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">nn <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">as<\/span><span class=\"token plain\"> nn<\/span><\/p><p><span class=\"token plain\"\/><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">from<\/span><span class=\"token plain\"> einops <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">import<\/span><span class=\"token plain\"> rearrange<\/span><\/p><p><span class=\"token plain\"\/><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">from<\/span><span class=\"token plain\"> self_attention_cv <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">import<\/span><span class=\"token plain\"> TransformerEncoder<\/span><\/p><p><span class=\"token plain\"\/><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">class<\/span><span class=\"token plain\"> <\/span><span class=\"token class-name\">ViT<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Module<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">    <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">def<\/span><span class=\"token plain\"> <\/span><span class=\"token function\" style=\"color:rgb(80, 250, 123)\">__init__<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token operator\">*<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 img_dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 in_channels<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">3<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 patch_dim<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">16<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 num_classes<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">10<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 dim<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">512<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 blocks<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">6<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 heads<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">4<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 dim_linear_block<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">1024<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 dim_head<\/span><span class=\"token operator\">=<\/span><span class=\"token boolean\">None<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                 dropout<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> transformer<\/span><span class=\"token operator\">=<\/span><span class=\"token boolean\">None<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> classification<\/span><span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">\"\"\"<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">        Args:<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            img_dim: the spatial image size<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            in_channels: number of img channels<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            patch_dim: desired patch dim<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            num_classes: classification task classes<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            dim: the linear layer's dim to project the patches for MHSA<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            blocks: number of transformer blocks<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            heads: number of heads<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            dim_linear_block: inner dim of the transformer linear block<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            dim_head: dim head in case you want to define it. defaults to dim\/heads<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            dropout: for pos emb and transformer<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            transformer: in case you want to provide another transformer implementation<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            classification: creates an extra CLS token<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">        \"\"\"<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token builtin\" style=\"color:rgb(189, 147, 249)\">super<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">__init__<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">assert<\/span><span class=\"token plain\"> img_dim <\/span><span class=\"token operator\">%<\/span><span class=\"token plain\"> patch_dim <\/span><span class=\"token operator\">==<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token string-interpolation string\" style=\"color:rgb(255, 121, 198)\">f'patch size <\/span><span class=\"token string-interpolation interpolation punctuation\" style=\"color:rgb(248, 248, 242)\">{<\/span><span class=\"token string-interpolation interpolation\">patch_dim<\/span><span class=\"token string-interpolation interpolation punctuation\" style=\"color:rgb(248, 248, 242)\">}<\/span><span class=\"token string-interpolation string\" style=\"color:rgb(255, 121, 198)\"> not divisible'<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">p <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> patch_dim<\/span><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">classification <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> classification<\/span><\/p><p><span class=\"token plain\">        tokens <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">img_dim <\/span><span class=\"token operator\">\/\/<\/span><span class=\"token plain\"> patch_dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"> <\/span><span class=\"token operator\">**<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">2<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">token_dim <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> in_channels <\/span><span class=\"token operator\">*<\/span><span class=\"token plain\"> <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">patch_dim <\/span><span class=\"token operator\">**<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">2<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">dim <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> dim<\/span><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">dim_head <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token builtin\" style=\"color:rgb(189, 147, 249)\">int<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">dim <\/span><span class=\"token operator\">\/<\/span><span class=\"token plain\"> heads<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"> <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">if<\/span><span class=\"token plain\"> dim_head <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">is<\/span><span class=\"token plain\"> <\/span><span class=\"token boolean\">None<\/span><span class=\"token plain\"> <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">else<\/span><span class=\"token plain\"> dim_head<\/span><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">project_patches <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Linear<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">token_dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">emb_dropout <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Dropout<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">dropout<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">if<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">classification<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">cls_token <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Parameter<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">torch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">randn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">pos_emb1D <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Parameter<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">torch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">randn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">tokens <\/span><span class=\"token operator\">+<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">mlp_head <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Linear<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> num_classes<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">else<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">pos_emb1D <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> nn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">Parameter<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">torch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">randn<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">tokens<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">if<\/span><span class=\"token plain\"> transformer <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">is<\/span><span class=\"token plain\"> <\/span><span class=\"token boolean\">None<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">transformer <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> TransformerEncoder<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">dim<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> blocks<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">blocks<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> heads<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">heads<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                                                  dim_head<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">dim_head<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                                                  dim_linear_block<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">dim_linear_block<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                                                  dropout<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">dropout<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">else<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">transformer <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> transformer<\/span><\/p><p><span class=\"token plain\">    <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">def<\/span><span class=\"token plain\"> <\/span><span class=\"token function\" style=\"color:rgb(80, 250, 123)\">expand_cls_to_batch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> batch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">\"\"\"<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">        Args:<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">            batch: batch size<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">        Returns: cls token expanded to the batch size<\/span><\/p><p><span class=\"token triple-quoted-string string\" style=\"color:rgb(255, 121, 198)\">        \"\"\"<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">return<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">cls_token<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">expand<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">[<\/span><span class=\"token plain\">batch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token operator\">-<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token operator\">-<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">]<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">    <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">def<\/span><span class=\"token plain\"> <\/span><span class=\"token function\" style=\"color:rgb(80, 250, 123)\">forward<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> img<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> mask<\/span><span class=\"token operator\">=<\/span><span class=\"token boolean\">None<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        batch_size <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> img<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">shape<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">]<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        img_patches <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> rearrange<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            img<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token string\" style=\"color:rgb(255, 121, 198)\">'b c (patch_x x) (patch_y y) -&gt; b (x y) (patch_x patch_y c)'<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                                patch_x<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">p<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> patch_y<\/span><span class=\"token operator\">=<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">p<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        img_patches <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">project_patches<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">img_patches<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">if<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">classification<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            img_patches <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> torch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">cat<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">                <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">expand_cls_to_batch<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">batch_size<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> img_patches<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> dim<\/span><span class=\"token operator\">=<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        patch_embeddings <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">emb_dropout<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">img_patches <\/span><span class=\"token operator\">+<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">pos_emb1D<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        y <\/span><span class=\"token operator\">=<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">transformer<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">patch_embeddings<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> mask<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">if<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">classification<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">return<\/span><span class=\"token plain\"> self<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">.<\/span><span class=\"token plain\">mlp_head<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">(<\/span><span class=\"token plain\">y<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">[<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">,<\/span><span class=\"token plain\"> <\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">]<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">)<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">        <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">else<\/span><span class=\"token punctuation\" style=\"color:rgb(248, 248, 242)\">:<\/span><span class=\"token plain\"\/><\/p><p><span class=\"token plain\">            <\/span><span class=\"token keyword\" style=\"color:rgb(189, 147, 249);font-style:italic\">return<\/span><span class=\"token plain\"> y<\/span><\/p><\/pre>\n<h2 id=\"conclusion\">\u0417\u0430\u043a\u043b\u044e\u0447\u0435\u043d\u0438\u0435<\/h2>\n<p>\u041a\u043b\u044e\u0447\u0435\u0432\u043e\u0439 \u0438\u043d\u0436\u0435\u043d\u0435\u0440\u043d\u043e\u0439 \u0447\u0430\u0441\u0442\u044c\u044e \u044d\u0442\u043e\u0439 \u0440\u0430\u0431\u043e\u0442\u044b \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u043f\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043a\u0430 \u0437\u0430\u0434\u0430\u0447\u0438 \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u0432\u0438\u0434\u0435 \u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0437\u0430\u0434\u0430\u0447\u0438 \u0441 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435\u043c \u0444\u0440\u0430\u0433\u043c\u0435\u043d\u0442\u043e\u0432 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439 \u0432 \u043a\u0430\u0447\u0435\u0441\u0442\u0432\u0435 \u0442\u043e\u043a\u0435\u043d\u043e\u0432 \u0438 \u0438\u0445 \u043e\u0431\u0440\u0430\u0431\u043e\u0442\u043a\u0430 \u043f\u0440\u0435\u043e\u0431\u0440\u0430\u0437\u043e\u0432\u0430\u0442\u0435\u043b\u0435\u043c.  \u042d\u0442\u043e \u0437\u0432\u0443\u0447\u0438\u0442 \u0445\u043e\u0440\u043e\u0448\u043e \u0438 \u043f\u0440\u043e\u0441\u0442\u043e, \u043d\u043e \u0434\u043b\u044f \u044d\u0442\u043e\u0433\u043e \u043d\u0443\u0436\u043d\u044b \u043c\u0430\u0441\u0441\u0438\u0432\u043d\u044b\u0435 \u0434\u0430\u043d\u043d\u044b\u0435.  \u041a \u0441\u043e\u0436\u0430\u043b\u0435\u043d\u0438\u044e, \u043f\u0440\u0435\u0434\u0432\u0430\u0440\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u043d\u044b\u0439 \u043d\u0430\u0431\u043e\u0440 \u0434\u0430\u043d\u043d\u044b\u0445 \u043f\u0440\u0438\u043d\u0430\u0434\u043b\u0435\u0436\u0438\u0442 Google, \u043f\u043e\u044d\u0442\u043e\u043c\u0443 \u0440\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442\u044b \u043d\u0435\u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e \u0432\u043e\u0441\u043f\u0440\u043e\u0438\u0437\u0432\u0435\u0441\u0442\u0438.  \u0418 \u0434\u0430\u0436\u0435 \u0435\u0441\u043b\u0438 \u0431\u044b \u043e\u043d\u0438 \u0431\u044b\u043b\u0438, \u0432\u0430\u043c \u043d\u0443\u0436\u043d\u043e \u0431\u044b\u043b\u043e \u0431\u044b \u0438\u043c\u0435\u0442\u044c \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u0443\u044e \u0432\u044b\u0447\u0438\u0441\u043b\u0438\u0442\u0435\u043b\u044c\u043d\u0443\u044e \u043c\u043e\u0449\u043d\u043e\u0441\u0442\u044c.<\/p>\n<div class=\"dl-prod-book-inline-banner\">\n<div class=\"dl-prod-book-inline-banner__image gatsby-image-wrapper\" style=\"position:relative;overflow:hidden\"><img decoding=\"async\" aria-hidden=\"true\" src=\"data:image\/png;base64,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\" alt=\"\u041a\u043d\u0438\u0433\u0430 \u00ab\u0413\u043b\u0443\u0431\u043e\u043a\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0432 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u0435\u00bb\" style=\"position:absolute;top:0;left:0;width:100%;height:100%;object-fit:contain;object-position:center;opacity:1;transition-delay:500ms\"\/><noscript><picture><source srcset=\"https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/69585\/deep-learning-book-cover.png 200w,&#10;https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/497c6\/deep-learning-book-cover.png 400w,&#10;https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/3c17d\/deep-learning-book-cover.png 720w\" sizes=\"(max-width: 720px) 100vw, 720px\"\/><img loading=\"lazy\" sizes=\"(max-width: 720px) 100vw, 720px\" srcset=\"https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/69585\/deep-learning-book-cover.png 200w,&#10;https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/497c6\/deep-learning-book-cover.png 400w,&#10;https:\/\/theaisummer.com\/static\/502e7c498dd9d981ac44c1dcd10f9276\/3c17d\/deep-learning-book-cover.png 720w\" alt=\"\u041a\u043d\u0438\u0433\u0430 \u00ab\u0413\u043b\u0443\u0431\u043e\u043a\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0432 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u0435\u00bb\" style=\"position:absolute;top:0;left:0;opacity:1;width:100%;height:100%;object-fit:cover;object-position:center\"\/><\/picture><\/noscript><\/div>\n<div class=\"dl-prod-book-inline-banner__text\">\n<h2>\u041a\u043d\u0438\u0433\u0430 \u00ab\u0413\u043b\u0443\u0431\u043e\u043a\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0432 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u0435\u00bb \ud83d\udcd6<\/h2>\n<h4>\u0423\u0437\u043d\u0430\u0439\u0442\u0435, \u043a\u0430\u043a \u0441\u043e\u0437\u0434\u0430\u0432\u0430\u0442\u044c, \u043e\u0431\u0443\u0447\u0430\u0442\u044c, \u0440\u0430\u0437\u0432\u0435\u0440\u0442\u044b\u0432\u0430\u0442\u044c, \u043c\u0430\u0441\u0448\u0442\u0430\u0431\u0438\u0440\u043e\u0432\u0430\u0442\u044c \u0438 \u043f\u043e\u0434\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u0442\u044c \u043c\u043e\u0434\u0435\u043b\u0438 \u0433\u043b\u0443\u0431\u043e\u043a\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f.  \u0418\u0437\u0443\u0447\u0438\u0442\u0435 \u0438\u043d\u0444\u0440\u0430\u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u0443 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0438 MLOps \u043d\u0430 \u043f\u0440\u0430\u043a\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u043f\u0440\u0438\u043c\u0435\u0440\u0430\u0445.<\/h4>\n<p>\u0423\u0437\u043d\u0430\u0442\u044c \u0431\u043e\u043b\u044c\u0448\u0435<\/p><\/div>\n<\/div>\n<p><em class=\"affiliate-disclosure\">* \u0420\u0430\u0441\u043a\u0440\u044b\u0442\u0438\u0435 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438: \u043e\u0431\u0440\u0430\u0442\u0438\u0442\u0435 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u0447\u0442\u043e \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0438\u0437 \u043f\u0440\u0438\u0432\u0435\u0434\u0435\u043d\u043d\u044b\u0445 \u0432\u044b\u0448\u0435 \u0441\u0441\u044b\u043b\u043e\u043a \u043c\u043e\u0433\u0443\u0442 \u0431\u044b\u0442\u044c \u043f\u0430\u0440\u0442\u043d\u0435\u0440\u0441\u043a\u0438\u043c\u0438 \u0441\u0441\u044b\u043b\u043a\u0430\u043c\u0438, \u0438 \u043c\u044b \u0431\u0435\u0437 \u0434\u043e\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0445 \u0437\u0430\u0442\u0440\u0430\u0442 \u0434\u043b\u044f \u0432\u0430\u0441 \u043f\u043e\u043b\u0443\u0447\u0438\u043c \u043a\u043e\u043c\u0438\u0441\u0441\u0438\u044e, \u0435\u0441\u043b\u0438 \u0432\u044b \u0440\u0435\u0448\u0438\u0442\u0435 \u0441\u043e\u0432\u0435\u0440\u0448\u0438\u0442\u044c \u043f\u043e\u043a\u0443\u043f\u043a\u0443 \u043f\u043e\u0441\u043b\u0435 \u043f\u0435\u0440\u0435\u0445\u043e\u0434\u0430 \u043f\u043e \u0441\u0441\u044b\u043b\u043a\u0435.<\/em><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u041d\u0430 \u044d\u0442\u043e\u0442 \u0440\u0430\u0437 \u044f \u0431\u0443\u0434\u0443 \u0440\u0435\u0437\u043a\u0438\u043c \u0438 \u043a\u0440\u0430\u0442\u043a\u0438\u043c. \u0427\u0435\u0440\u0435\u0437 10 \u043c\u0438\u043d\u0443\u0442 \u044f \u0443\u043a\u0430\u0436\u0443 \u043c\u0435\u043b\u043a\u0438\u0435 \u043c\u043e\u0434\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0430\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u044b \u0442\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0435\u0440\u0430 \u0434\u043b\u044f \u043a\u043b\u0430\u0441\u0441\u0438\u0444\u0438\u043a\u0430\u0446\u0438\u0438 \u0438\u0437\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0439. \u041f\u043e\u0441\u043a\u043e\u043b\u044c\u043a\u0443 \u044d\u0442\u043e \u0441\u043b\u0435\u0434\u0443\u044e\u0449\u0430\u044f \u0441\u0442\u0430\u0442\u044c\u044f, \u043d\u0435 \u0441\u0442\u0435\u0441\u043d\u044f\u0439\u0442\u0435\u0441\u044c \u0441\u043e\u0432\u0435\u0442\u043e\u0432\u0430\u0442\u044c \u043c\u043e\u0438 \u043f\u0440\u0435\u0434\u044b\u0434\u0443\u0449\u0438\u0435 \u0441\u0442\u0430\u0442\u044c\u0438 \u043e Transformer \u0438 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435, \u0435\u0441\u043b\u0438 \u0432\u044b \u043d\u0435 \u0447\u0443\u0432\u0441\u0442\u0432\u0443\u0435\u0442\u0435 \u0441\u0435\u0431\u044f \u043a\u043e\u043c\u0444\u043e\u0440\u0442\u043d\u043e \u0441 \u0442\u0435\u0440\u043c\u0438\u043d\u0430\u043c\u0438. \u0422\u0435\u043f\u0435\u0440\u044c, \u0434\u0430\u043c\u044b \u0438 \u0433\u043e\u0441\u043f\u043e\u0434\u0430, \u043c\u043e\u0436\u0435\u0442\u0435 \u0437\u0430\u043f\u0443\u0441\u043a\u0430\u0442\u044c \u0447\u0430\u0441\u044b! \u0422\u0440\u0430\u043d\u0441\u0444\u043e\u0440\u043c\u0430\u0442\u043e\u0440\u0430\u043c \u043d\u0435 \u0445\u0432\u0430\u0442\u0430\u0435\u0442 \u0438\u043d\u0434\u0443\u043a\u0442\u0438\u0432\u043d\u044b\u0445 \u0441\u043c\u0435\u0449\u0435\u043d\u0438\u0439 \u0441\u0432\u0435\u0440\u0442\u043e\u0447\u043d\u044b\u0445 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":1322,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":{"0":"post-1321","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-ai-research-and-news"},"_links":{"self":[{"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/posts\/1321","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/gptmain.news\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1321"}],"version-history":[{"count":0,"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/posts\/1321\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/gptmain.news\/index.php?rest_route=\/wp\/v2\/media\/1322"}],"wp:attachment":[{"href":"https:\/\/gptmain.news\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1321"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/gptmain.news\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1321"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/gptmain.news\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1321"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}