{"id":64929,"date":"2021-04-27T11:11:43","date_gmt":"2021-04-27T10:11:43","guid":{"rendered":"https:\/\/datascientest.com\/?p=64929"},"modified":"2023-11-09T14:07:31","modified_gmt":"2023-11-09T14:07:31","slug":"quest-ce-que-le-modele-vgg","status":"publish","type":"post","link":"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg","title":{"rendered":"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-post\" data-elementor-id=\"64929\" class=\"elementor elementor-64929\" data-elementor-post-type=\"post\">\n\t\t\t\t\t\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-ab17bb8 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"ab17bb8\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-cc7ddc3\" data-id=\"cc7ddc3\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-147c850 elementor-widget elementor-widget-heading\" data-id=\"147c850\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.15.0 - 02-08-2023 *\/\n.elementor-heading-title{padding:0;margin:0;line-height:1}.elementor-widget-heading .elementor-heading-title[class*=elementor-size-]>a{color:inherit;font-size:inherit;line-height:inherit}.elementor-widget-heading .elementor-heading-title.elementor-size-small{font-size:15px}.elementor-widget-heading .elementor-heading-title.elementor-size-medium{font-size:19px}.elementor-widget-heading .elementor-heading-title.elementor-size-large{font-size:29px}.elementor-widget-heading .elementor-heading-title.elementor-size-xl{font-size:39px}.elementor-widget-heading .elementor-heading-title.elementor-size-xxl{font-size:59px}<\/style><h2 class=\"elementor-heading-title elementor-size-default\">Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu'\u00e0 leur dipl\u00f4me. Aujourd'hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG.<\/h2>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-cd001b4 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"cd001b4\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-5bee5a1\" data-id=\"5bee5a1\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-f07355d elementor-widget elementor-widget-text-editor\" data-id=\"f07355d\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.15.0 - 02-08-2023 *\/\n.elementor-widget-text-editor.elementor-drop-cap-view-stacked .elementor-drop-cap{background-color:#69727d;color:#fff}.elementor-widget-text-editor.elementor-drop-cap-view-framed .elementor-drop-cap{color:#69727d;border:3px solid;background-color:transparent}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap{margin-top:8px}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap-letter{width:1em;height:1em}.elementor-widget-text-editor .elementor-drop-cap{float:left;text-align:center;line-height:1;font-size:50px}.elementor-widget-text-editor .elementor-drop-cap-letter{display:inline-block}<\/style>\t\t\t\t<p><span style=\"font-weight: 400;\">Nous avons d\u00e9j\u00e0 eu l\u2019occasion, il y a quelque temps de d\u00e9tailler la notion de <\/span><strong><a href=\"https:\/\/datascientest.com\/transfer-learning\">Transfer Learning<\/a><\/strong><span style=\"font-weight: 400;\"> ou apprentissage par transfert en fran\u00e7ais. Si vous avez rat\u00e9 notre article sur le sujet, je vous invite \u00e0 le relire <a href=\"https:\/\/datascientest.com\/transfer-learning\">ici<\/a>. Ce qu\u2019il faut avoir en t\u00eate, c\u2019est que le <\/span><i><span style=\"font-weight: 400;\">Transfer Learning <\/span><\/i><span style=\"font-weight: 400;\">correspond \u00e0 la capacit\u00e9 \u00e0 utiliser des connaissances existantes, d\u00e9velopp\u00e9es pour la r\u00e9solution de probl\u00e9matiques donn\u00e9es, pour r\u00e9soudre une nouvelle probl\u00e9matique. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-6fe5355 elementor-widget elementor-widget-heading\" data-id=\"6fe5355\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">Un peu d\u2019histoire<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7745796 elementor-widget elementor-widget-text-editor\" data-id=\"7745796\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">VGG est un r\u00e9seau de neurones convolutionnels propos\u00e9s par K. Simonyan et A. Zisserman de l\u2019universit\u00e9 d\u2019Oxford et qui a acquis une notori\u00e9t\u00e9 en gagnant la comp\u00e9tition <strong>ILSVRC<\/strong> (<\/span><i><span style=\"font-weight: 400;\">ImageNet Large Scale Visual Recognition Challenge<\/span><\/i><span style=\"font-weight: 400;\">) en 2014. Le mod\u00e8le a atteint une pr\u00e9cision de 92.7% sur <\/span><i><span style=\"font-weight: 400;\">Imagenet <\/span><\/i><span style=\"font-weight: 400;\">ce qui est un des meilleurs scores obtenus. <\/span><span style=\"font-weight: 400;\">Il a marqu\u00e9 une progression par rapport aux mod\u00e8les pr\u00e9c\u00e9dents en proposant, dans les couches de convolution, <strong>des noyaux de convolution de plus petites dimensions (3&#215;3)<\/strong> que ce qui avait \u00e9t\u00e9 fait jusque-l\u00e0.<\/span><span style=\"font-weight: 400;\"> Le mod\u00e8le a \u00e9t\u00e9 entra\u00een\u00e9 sur des semaines en utilisant des cartes graphiques de pointe. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-4733ff0 elementor-align-center button-principale elementor-widget elementor-widget-button\" data-id=\"4733ff0\" data-element_type=\"widget\" data-widget_type=\"button.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-button-wrapper\">\n\t\t\t<a class=\"elementor-button elementor-button-link elementor-size-sm elementor-animation-grow\" href=\"https:\/\/datascientest.com\/formation-data-scientist\">\n\t\t\t\t\t\t<span class=\"elementor-button-content-wrapper\">\n\t\t\t\t\t\t<span class=\"elementor-button-text\">En savoir plus sur notre formation de Data Scientist<\/span>\n\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-173b1d0 elementor-widget elementor-widget-heading\" data-id=\"173b1d0\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">ImageNet<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-1c2acab elementor-widget elementor-widget-text-editor\" data-id=\"1c2acab\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">ImageNet est une gigantesque base de donn\u00e9es de plus de <\/span><b>14 millions d\u2019images labellis\u00e9es <\/b><span style=\"font-weight: 400;\">r\u00e9parties dans plus de 1000 classes, en 2014. En 2007, une chercheuse du nom de Fei-Fei Li a commenc\u00e9 \u00e0 travailler sur l\u2019id\u00e9e de cr\u00e9er un tel jeu de donn\u00e9es. <\/span><span style=\"font-weight: 400;\">Certes la mod\u00e9lisation est un aspect tr\u00e8s important pour obtenir des bonnes performances, mais disposer de donn\u00e9es de grande qualit\u00e9 l\u2019est tout autant pour avoir un apprentissage de qualit\u00e9.<\/span><span style=\"font-weight: 400;\"> Les donn\u00e9es ont \u00e9t\u00e9 collect\u00e9es et \u00e9tiquet\u00e9es depuis le web par des humains. Elles sont donc <\/span><a href=\"https:\/\/datascientest.com\/open-source\"><i><span style=\"font-weight: 400;\"><strong>open source<\/strong> <\/span><\/i><\/a><span style=\"font-weight: 400;\">et n&rsquo;appartiennent pas \u00e0 une entreprise en particulier.\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">Depuis 2010 s\u2019organise, chaque ann\u00e9e, une comp\u00e9tition <\/span><i><span style=\"font-weight: 400;\">ImageNet Large Scale Visual Recognition Challenge <\/span><\/i><span style=\"font-weight: 400;\">dont le but est de <\/span><b>challenger des mod\u00e8les de traitement d\u2019image<\/b><span style=\"font-weight: 400;\"><strong>s<\/strong>. <\/span><span style=\"font-weight: 400;\">La comp\u00e9tition s\u2019effectue sur un sous-ensemble d\u2019<\/span><i><span style=\"font-weight: 400;\">ImageNet <\/span><\/i><span style=\"font-weight: 400;\">compos\u00e9 de : 1,2 million d\u2019images d\u2019entra\u00eenement, 50000 pour validation et 150000 pour tester le mod\u00e8le. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-eecb987 elementor-widget elementor-widget-heading\" data-id=\"eecb987\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">L\u2019architecture <\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-23b35b5 elementor-widget elementor-widget-text-editor\" data-id=\"23b35b5\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400; color: #000000;\">Dans les faits, il existe deux algorithmes disponibles : <strong>VGG16 et VGG19<\/strong>. Dans cet article, nous allons nous concentrer sur l\u2019architecture du premier. Si les deux architectures sont tr\u00e8s proches et respectent la m\u00eame logique, VGG19 pr\u00e9sente un plus grand nombre de couches de convolution.<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-621d9e6 elementor-widget elementor-widget-image\" data-id=\"621d9e6\" data-element_type=\"widget\" data-widget_type=\"image.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.15.0 - 02-08-2023 *\/\n.elementor-widget-image{text-align:center}.elementor-widget-image a{display:inline-block}.elementor-widget-image a img[src$=\".svg\"]{width:48px}.elementor-widget-image img{vertical-align:middle;display:inline-block}<\/style>\t\t\t\t\t\t\t\t\t<figure class=\"wp-caption\">\n\t\t\t\t\t\t\t\t\t\t<img decoding=\"async\" width=\"800\" height=\"439\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG_Plan-de-travail-1.png\" class=\"attachment-large size-large wp-image-68385\" alt=\"Architecture Algoritme VGG16\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG_Plan-de-travail-1.png 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG_Plan-de-travail-1-300x165.png 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG_Plan-de-travail-1-768x422.png 768w\" sizes=\"(max-width: 800px) 100vw, 800px\" \/>\t\t\t\t\t\t\t\t\t\t\t<figcaption class=\"widget-image-caption wp-caption-text\">Architecture Algoritme VGG16<\/figcaption>\n\t\t\t\t\t\t\t\t\t\t<\/figure>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-beecfeb elementor-widget elementor-widget-image\" data-id=\"beecfeb\" data-element_type=\"widget\" data-widget_type=\"image.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\t\t\t\t<figure class=\"wp-caption\">\n\t\t\t\t\t\t\t\t\t\t<img decoding=\"async\" width=\"800\" height=\"439\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-02.png\" class=\"attachment-large size-large wp-image-68383\" alt=\"Structure VGG 16\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-02.png 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-02-300x165.png 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-02-768x422.png 768w\" sizes=\"(max-width: 800px) 100vw, 800px\" \/>\t\t\t\t\t\t\t\t\t\t\t<figcaption class=\"widget-image-caption wp-caption-text\">Structure Algoritme VGG16<\/figcaption>\n\t\t\t\t\t\t\t\t\t\t<\/figure>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-60b6125 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"60b6125\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-818ff64\" data-id=\"818ff64\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-de0e3c3 elementor-align-center button-principale elementor-widget elementor-widget-button\" data-id=\"de0e3c3\" data-element_type=\"widget\" data-widget_type=\"button.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-button-wrapper\">\n\t\t\t<a class=\"elementor-button elementor-button-link elementor-size-sm elementor-animation-grow\" href=\"https:\/\/datascientest.com\/formation-data-scientist\">\n\t\t\t\t\t\t<span class=\"elementor-button-content-wrapper\">\n\t\t\t\t\t\t<span class=\"elementor-button-text\">D\u00e9couvrez notre parcours de Data Scientist<\/span>\n\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-0d48b7a elementor-widget elementor-widget-text-editor\" data-id=\"0d48b7a\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">Le mod\u00e8le ne demande qu\u2019un pr\u00e9traitement sp\u00e9cifique qui consiste \u00e0 soustraire la valeur RGB moyenne, calcul\u00e9e sur l&rsquo;ensemble d&rsquo;apprentissage, de chaque pixel.<\/span><\/p><p><span style=\"font-weight: 400;\">Durant l\u2019apprentissage du mod\u00e8le, l\u2019input de la <\/span><b>premi\u00e8re couche de convolution<\/b><span style=\"font-weight: 400;\"> est une image RGB de taille 224 x 224. Pour toutes les couches de convolution, le noyau de convolution est de taille 3&#215;3:\u00a0 la plus petite dimension pour capturer les notions de haut, bas, gauche\/droite et centre. C\u2019\u00e9tait une sp\u00e9cificit\u00e9 du mod\u00e8le au moment de sa publication. Jusqu\u2019\u00e0 VGG16 beaucoup de mod\u00e8les s\u2019orientaient vers des noyaux de convolution de plus grande dimension (de taille 11 ou bien de taille 5 par exemple). <\/span><strong>Rappelons que ces couches ont pour but de filtrer l\u2019image en ne gardant que des informations discriminantes comme des formes g\u00e9om\u00e9triques atypiques. <\/strong><span style=\"font-weight: 400;\">\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">Ces couches de convolution s\u2019accompagnent de couche de <\/span><b>Max-Pooling<\/b><span style=\"font-weight: 400;\">, chacune de taille 2&#215;2, pour r\u00e9duire la taille des filtres au cours de l\u2019apprentissage. <\/span><span style=\"font-weight: 400;\">\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">En sortie des couches de convolution et pooling, nous avons 3 couches de neurones Fully-Connected. Les deux premi\u00e8res sont compos\u00e9es de 4096 neurones et la derni\u00e8re de 1000 neurones avec une fonction d\u2019activation softmax pour d\u00e9terminer la classe de l\u2019image.<\/span><\/p><p><span style=\"font-weight: 400;\">Comme vous avez pu le constater l\u2019architecture est claire et simple \u00e0 comprendre ce qui est aussi une force de ce mod\u00e8le. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-90b6928 elementor-widget elementor-widget-heading\" data-id=\"90b6928\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">R\u00e9sultats obtenus sur ImageNet<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-422cb6c elementor-widget elementor-widget-image\" data-id=\"422cb6c\" data-element_type=\"widget\" data-widget_type=\"image.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<img decoding=\"async\" width=\"602\" height=\"259\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/image2-1.png\" class=\"attachment-large size-large wp-image-64955\" alt=\"\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/image2-1.png 602w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/image2-1-300x129.png 300w\" sizes=\"(max-width: 602px) 100vw, 602px\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-e6bc949 elementor-widget elementor-widget-text-editor\" data-id=\"e6bc949\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">La figure ci-dessus compare les r\u00e9sultats de diff\u00e9rents mod\u00e8les de 2014 ou bien des ann\u00e9es pr\u00e9c\u00e9dentes. Nous pouvons voir que VGG donne les meilleurs r\u00e9sultats aussi bien sur le jeu de validation que sur le jeu de test. Remarquons \u00e9galement que le mod\u00e8le donne de bien meilleurs r\u00e9sultats que lors des sessions 2012 et 2013.\u00a0<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7f01985 elementor-widget elementor-widget-heading\" data-id=\"7f01985\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">Et le Transfer learning dans tout \u00e7a ?<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-a0dc8d4 elementor-widget elementor-widget-text-editor\" data-id=\"a0dc8d4\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">Comme nous l\u2019avons d\u00e9j\u00e0 mentionn\u00e9 pr\u00e9c\u00e9demment, le temps d\u2019entra\u00eenement d\u2019un mod\u00e8le comme VGG peut \u00eatre tr\u00e8s long surtout si vous n\u2019avez pas beaucoup de ressources \u00e0 votre disposition. De plus \u00e9tant donn\u00e9 qu\u2019il a \u00e9t\u00e9 entra\u00een\u00e9 sur <\/span><i><span style=\"font-weight: 400;\">ImageNet <\/span><\/i><span style=\"font-weight: 400;\">il peut \u00eatre int\u00e9ressant de r\u00e9cup\u00e9rer les poids du mod\u00e8le d\u00e9j\u00e0 entra\u00een\u00e9 et notamment les filtres dans les couches de convolution issus de l\u2019apprentissage sur <\/span><i><span style=\"font-weight: 400;\">ImageNet.<\/span><\/i> <strong>C\u2019est d\u2019ailleurs ce qui est fait dans la pratique : nous r\u00e9cup\u00e9rons les poids issus des couches de convolution et nous avons simplement \u00e0 entra\u00eener les 3 couches que nous ajoutons.<\/strong> <span style=\"font-weight: 400;\">Le principe restant le m\u00eame : utiliser les connaissances acquises sur <\/span><i><span style=\"font-weight: 400;\">ImageNet <\/span><\/i><span style=\"font-weight: 400;\">pour r\u00e9soudre une probl\u00e9matique proche.\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">Il est notamment possible de r\u00e9cup\u00e9rer directement le mod\u00e8le pr\u00e9-entra\u00een\u00e9 tr\u00e8s facilement et d\u2019appliquer le pr\u00e9traitement sp\u00e9cifique demand\u00e9 par le mod\u00e8le. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-022429a elementor-widget elementor-widget-heading\" data-id=\"022429a\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">Un peu de pratique<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-54e2583 elementor-widget elementor-widget-text-editor\" data-id=\"54e2583\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">Dans les faits il existe deux algorithmes disponibles : VGG16 et VGG19.\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">Gr\u00e2ce \u00e0 la librairie keras de Tensorflow, il est simple de r\u00e9cup\u00e9rer le mod\u00e8le d\u00e9j\u00e0 entra\u00een\u00e9, par d\u00e9faut, sur <\/span><i><span style=\"font-weight: 400;\">ImageNet<\/span><\/i><span style=\"font-weight: 400;\">.\u00a0<\/span><\/p><p><span style=\"font-weight: 400;\">Dans un premier temps, nous avons besoin d\u2019appliquer le m\u00eame traitement sp\u00e9cifique que celui appliqu\u00e9 au moment de l\u2019entra\u00eenement du mod\u00e8le. En plus de cela nous ajoutons de l\u2019augmentation de donn\u00e9es sur les donn\u00e9es d\u2019apprentissage pour pr\u00e9venir un risque de sur-apprentissage. Il est aussi important de v\u00e9rifier que les images donn\u00e9es en input sont des images RGB de taille 224&#215;224.\u00a0<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-848e233 elementor-widget elementor-widget-shortcode\" data-id=\"848e233\" data-element_type=\"widget\" data-widget_type=\"shortcode.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-shortcode\"><script src=\"https:\/\/gist.github.com\/Thibault-venet\/ec44f996ba1ff384448564aeb6960735.js\"><\/script><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-5235b9c elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"5235b9c\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-7aac525\" data-id=\"7aac525\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-50691a1 elementor-widget elementor-widget-text-editor\" data-id=\"50691a1\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">Nous pouvons alors r\u00e9cup\u00e9rer les poids optimis\u00e9s issus des couches de convolution et entra\u00eener les 3 couches Dense que nous ajoutons et compiler :<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-62fa10a elementor-widget elementor-widget-shortcode\" data-id=\"62fa10a\" data-element_type=\"widget\" data-widget_type=\"shortcode.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-shortcode\"><script src=\"https:\/\/gist.github.com\/Thibault-venet\/2daa77bdb6f4ab8aacc1606908ce14fb.js\"><\/script><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-a9b7027 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"a9b7027\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-129e9c5\" data-id=\"129e9c5\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-887d696 elementor-widget elementor-widget-text-editor\" data-id=\"887d696\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">Apr\u00e8s entra\u00eenement, si les r\u00e9sultats ne sont pas assez bons, il est possible de ne plus figer les derni\u00e8res couches de convolution et de les r\u00e9-entra\u00eener pour essayer d\u2019avoir de meilleures performances. L\u00e0 aussi tout d\u00e9pend des ressources dont vous disposez, cette \u00e9tape suppl\u00e9mentaire peut \u00eatre co\u00fbteuse en temps. <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7050b83 elementor-widget elementor-widget-heading\" data-id=\"7050b83\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">Conclusion<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-5c36856 elementor-widget elementor-widget-text-editor\" data-id=\"5c36856\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400;\">VGG est un algorithme connu en <\/span><strong><a href=\"https:\/\/datascientest.com\/computer-vision\"><i>Computer Vision <\/i><\/a><\/strong><span style=\"font-weight: 400;\">tr\u00e8s souvent utilis\u00e9 par transfert d\u2019apprentissage pour \u00e9viter d\u2019avoir \u00e0 le r\u00e9entra\u00eener et r\u00e9soudre des probl\u00e9matiques proches sur lesquelles VGG a d\u00e9j\u00e0 \u00e9t\u00e9 entra\u00een\u00e9. Il existe bien d\u2019autres algorithmes du m\u00eame type que VGG comme ResNet ou bien encore Xception disponible dans la librairie Keras. Si vous souhaitez vous former sur des probl\u00e9matiques de <a href=\"https:\/\/datascientest.com\/formation-en-deep-learning-par-ou-commencer\">Deep Learning<\/a>\/Computer Vision, rejoignez-nous en Bootcamp ou bien en formation continue.\u00a0 <\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-6b078a6 elementor-align-center button-principale elementor-widget elementor-widget-button\" data-id=\"6b078a6\" data-element_type=\"widget\" data-widget_type=\"button.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-button-wrapper\">\n\t\t\t<a class=\"elementor-button elementor-button-link elementor-size-sm elementor-animation-grow\" href=\"https:\/\/datascientest.com\/formation-data-scientist\">\n\t\t\t\t\t\t<span class=\"elementor-button-content-wrapper\">\n\t\t\t\t\t\t<span class=\"elementor-button-text\">Devenez Data Scientist<\/span>\n\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu&rsquo;\u00e0 leur dipl\u00f4me. Aujourd&rsquo;hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG. Nous avons d\u00e9j\u00e0 eu l\u2019occasion, il y a quelque temps de d\u00e9tailler la notion de Transfer Learning ou apprentissage par transfert en fran\u00e7ais. Si vous [&hellip;]<\/p>\n","protected":false},"author":77,"featured_media":68381,"comment_status":"open","ping_status":"open","sticky":false,"template":"elementor_theme","format":"standard","meta":{"content-type":"","inline_featured_image":false},"categories":[2497],"tags":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v21.7 (Yoast SEO v21.7) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !<\/title>\n<meta name=\"description\" content=\"Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu&#039;\u00e0 leur dipl\u00f4me. Aujourd&#039;hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !\" \/>\n<meta property=\"og:url\" content=\"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg\" \/>\n<meta property=\"og:site_name\" content=\"Formation Data Science | DataScientest.com\" \/>\n<meta property=\"article:published_time\" content=\"2021-04-27T10:11:43+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-11-09T14:07:31+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-03.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"562\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Raphael Kassel\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"Raphael Kassel\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg\",\"url\":\"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg\",\"name\":\"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !\",\"isPartOf\":{\"@id\":\"https:\/\/datascientest.com\/#website\"},\"datePublished\":\"2021-04-27T10:11:43+00:00\",\"dateModified\":\"2023-11-09T14:07:31+00:00\",\"author\":{\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1\"},\"description\":\"Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu'\u00e0 leur dipl\u00f4me. Aujourd'hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG.\",\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg\"]}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/datascientest.com\/#website\",\"url\":\"https:\/\/datascientest.com\/\",\"name\":\"Formation Data Science | DataScientest.com\",\"description\":\"Leader de la formation en Data Science\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/datascientest.com\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1\",\"name\":\"Raphael Kassel\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg\",\"contentUrl\":\"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg\",\"caption\":\"Raphael Kassel\"},\"description\":\"Raphael Kassel est le Head of Data Programs chez DataScientest. Il supervise une \u00e9quipe Data, impl\u00e9mente de nouveaux parcours comme \\\"Chef de projet IA\\\" et \\\"Data Product Manager\\\", et contribue \u00e0 la modernisation des op\u00e9rations du groupe. En collaboration \u00e9troite avec l'\u00e9quipe de d\u00e9veloppement, il adapte l'accompagnement acad\u00e9mique aux \u00e9volutions de l'\u00e9quipe data et \u00e0 l'internationalisation.\",\"sameAs\":[\"https:\/\/www.linkedin.com\/in\/raphaelkassel\/\",\"Data Science\"],\"url\":\"https:\/\/datascientest.com\/author\/raphaelkassel\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !","description":"Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu'\u00e0 leur dipl\u00f4me. Aujourd'hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg","og_locale":"fr_FR","og_type":"article","og_title":"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !","og_url":"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg","og_site_name":"Formation Data Science | DataScientest.com","article_published_time":"2021-04-27T10:11:43+00:00","article_modified_time":"2023-11-09T14:07:31+00:00","og_image":[{"width":1024,"height":562,"url":"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/04\/illu_VGG-03.png","type":"image\/png"}],"author":"Raphael Kassel","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"Raphael Kassel","Dur\u00e9e de lecture estim\u00e9e":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg","url":"https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg","name":"VGG : en quoi consiste ce mod\u00e8le ? Daniel vous dit tout !","isPartOf":{"@id":"https:\/\/datascientest.com\/#website"},"datePublished":"2021-04-27T10:11:43+00:00","dateModified":"2023-11-09T14:07:31+00:00","author":{"@id":"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1"},"description":"Retour de Daniel, le personnage embl\u00e9matique de nos formations qui guide nos apprenants jusqu'\u00e0 leur dipl\u00f4me. Aujourd'hui il va vous pr\u00e9senter un mod\u00e8le fr\u00e9quemment utilis\u00e9 en Computer Vision : VGG.","inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/datascientest.com\/quest-ce-que-le-modele-vgg"]}]},{"@type":"WebSite","@id":"https:\/\/datascientest.com\/#website","url":"https:\/\/datascientest.com\/","name":"Formation Data Science | DataScientest.com","description":"Leader de la formation en Data Science","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/datascientest.com\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1","name":"Raphael Kassel","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/datascientest.com\/#\/schema\/person\/image\/","url":"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg","contentUrl":"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg","caption":"Raphael Kassel"},"description":"Raphael Kassel est le Head of Data Programs chez DataScientest. Il supervise une \u00e9quipe Data, impl\u00e9mente de nouveaux parcours comme \"Chef de projet IA\" et \"Data Product Manager\", et contribue \u00e0 la modernisation des op\u00e9rations du groupe. En collaboration \u00e9troite avec l'\u00e9quipe de d\u00e9veloppement, il adapte l'accompagnement acad\u00e9mique aux \u00e9volutions de l'\u00e9quipe data et \u00e0 l'internationalisation.","sameAs":["https:\/\/www.linkedin.com\/in\/raphaelkassel\/","Data Science"],"url":"https:\/\/datascientest.com\/author\/raphaelkassel"}]}},"_links":{"self":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts\/64929"}],"collection":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/users\/77"}],"replies":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/comments?post=64929"}],"version-history":[{"count":0,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts\/64929\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/media\/68381"}],"wp:attachment":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/media?parent=64929"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/categories?post=64929"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/tags?post=64929"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}