{"id":4084,"date":"2024-07-12T18:12:23","date_gmt":"2024-07-12T10:12:23","guid":{"rendered":"https:\/\/nullthought.net\/?p=4084"},"modified":"2024-12-25T14:30:37","modified_gmt":"2024-12-25T06:30:37","slug":"paligemma%ef%bc%9a%e4%b8%80%e7%a7%8d%e5%8f%af%e6%9c%89%e6%95%88%e8%bf%9b%e8%a1%8c%e4%bb%bb%e5%8a%a1%e8%bf%81%e7%a7%bb%e8%a7%86%e8%a7%89%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8bvlm","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=4084","title":{"rendered":"PaliGemma\uff1a\u4e00\u79cd\u53ef\u6709\u6548\u8fdb\u884c\u4efb\u52a1\u8fc1\u79fb\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b(VLM)"},"content":{"rendered":"\n<p><strong>PaliGemma<\/strong> \u662f\u4e00\u4e2a\u5f00\u653e\u7684<strong><a href=\"https:\/\/nullthought.net\/?p=4920\" target=\"_blank\" rel=\"noreferrer noopener\">\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09<\/a><\/strong>\uff0c\u5b83\u7ed3\u5408\u4e86 SigLIP-So400m \u89c6\u89c9\u7f16\u7801\u5668\u548c Gemma-2B \u8bed\u8a00\u6a21\u578b\u3002\u5176\u76ee\u6807\u662f\u63d0\u4f9b\u4e00\u4e2a\u591a\u529f\u80fd\u7684\u57fa\u7840\u6a21\u578b\uff0c\u80fd\u591f\u6709\u6548\u8fc1\u79fb\u5230\u5404\u79cd\u4efb\u52a1\uff0c\u5728\u8fd140\u4e2a\u4e0d\u540c\u7684\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u5305\u62ec\u6807\u51c6\u7684 VLM \u57fa\u51c6\u6d4b\u8bd5\u4ee5\u53ca\u66f4\u4e13\u4e1a\u7684\u4efb\u52a1\u5982\u9065\u611f\u548c\u5206\u5272\u3002<\/p>\n\n\n\n<p>PaliGemma\u662f\u6765\u81eaGoogle Deepmind\u56e2\u961f\u7684\u8fd1\u671f\u7814\u7a76\u6210\u679c\uff0c\u7814\u7a76\u5185\u5bb9\u53d1\u8868\u5728\u8bba\u6587<strong><em><a href=\"https:\/\/arxiv.org\/pdf\/2407.07726\" target=\"_blank\" rel=\"noreferrer noopener\">PaliGemma: A versatile 3B VLM for transfer<\/a><\/em><\/strong>\u4e2d\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u5185\u5bb9\u7b80\u4ecb\u5982\u4e0b\uff1a<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e00\u3001PaliGemma\u4e3b\u8981\u7ec4\u6210\u90e8\u5206<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u89c6\u89c9\u7f16\u7801\u5668\uff08SigLIP-So400m\uff09<\/strong>\uff1a\u4e00\u4e2a\u4f18\u5316\u5f62\u72b6\u7684\u89c6\u89c9\u7f16\u7801\u5668\uff0c\u4f7f\u7528\u5bf9\u6bd4\u635f\u5931\u65b9\u6cd5\u8fdb\u884c\u9884\u8bad\u7ec3\u3002\u5b83\u5c55\u793a\u4e86\u6700\u5148\u8fdb\u7684\u6027\u80fd\uff0c\u5c24\u5176\u662f\u8003\u8651\u5230\u5176\u76f8\u5bf9\u8f83\u5c0f\u7684\u5c3a\u5bf8\u3002<\/li>\n\n\n\n<li><strong>\u8bed\u8a00\u6a21\u578b\uff08Gemma-2B\uff09<\/strong>\uff1a\u6765\u81ea Gemma \u7cfb\u5217\u7684\u81ea\u56de\u5f52\u89e3\u7801\u5668\uff0c\u4ec5\u7528\u4e8e\u8bed\u8a00\u6a21\u578b\uff0c\u5728\u6027\u80fd\u548c\u5927\u5c0f\u4e4b\u95f4\u53d6\u5f97\u4e86\u826f\u597d\u7684\u5e73\u8861\u3002PaliGemma \u4f7f\u7528\u4e86 2B \u53d8\u4f53\u3002<\/li>\n\n\n\n<li><strong>\u7ebf\u6027\u6295\u5f71\u5c42<\/strong>\uff1a\u5c06 SigLIP \u7684\u8f93\u51fa\u6807\u8bb0\u6295\u5f71\u5230\u4e0e Gemma-2B \u7684\u8bcd\u6c47\u6807\u8bb0\u76f8\u540c\u7684\u7ef4\u5ea6\uff0c\u4ee5\u4fbf\u8fdb\u884c\u62fc\u63a5\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e8c\u3001\u8bad\u7ec3\u9636\u6bb5<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5355\u6a21\u6001\u9884\u8bad\u7ec3\uff08Stage0\uff09<\/strong>\uff1a\u4f7f\u7528\u73b0\u6709\u7684\u9884\u8bad\u7ec3\u68c0\u67e5\u70b9\uff0c\u5206\u522b\u7528\u4e8e\u89c6\u89c9\u548c\u8bed\u8a00\u7ec4\u4ef6\u3002<\/li>\n\n\n\n<li><strong>\u591a\u6a21\u6001\u9884\u8bad\u7ec3\uff08Stage1\uff09<\/strong>\uff1a\u7ed3\u5408\u89c6\u89c9\u548c\u8bed\u8a00\u6a21\u578b\uff0c\u5e76\u5728\u591a\u6a21\u6001\u4efb\u52a1\u6df7\u5408\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u786e\u4fdd\u4e0d\u51bb\u7ed3\u56fe\u50cf\u7f16\u7801\u5668\u3002\u4f7f\u7528\u7ebf\u6027\u7f13\u6162\u7684\u9884\u70ed\u6765\u8c03\u6574\u56fe\u50cf\u7f16\u7801\u5668\u7684\u5b66\u4e60\u7387\u3002<\/li>\n\n\n\n<li><strong>\u5206\u8fa8\u7387\u63d0\u5347\uff08Stage2\uff09<\/strong>\uff1a\u5728\u66f4\u9ad8\u5206\u8fa8\u7387\uff08448px \u548c 896px\uff09\u4e0b\u8bad\u7ec3\u6a21\u578b\uff0c\u4ee5\u63d0\u9ad8\u9700\u8981\u9ad8\u5206\u8fa8\u7387\u8f93\u5165\u7684\u4efb\u52a1\uff08\u5982\u68c0\u6d4b\u548c\u5206\u5272\uff09\u7684\u6027\u80fd\u3002<\/li>\n\n\n\n<li><strong>\u8fc1\u79fb\uff08Stage3\uff09<\/strong>\uff1a\u9488\u5bf9\u7279\u5b9a\u4efb\u52a1\u5fae\u8c03\u57fa\u7840\u6a21\u578b\uff0c\u8c03\u6574\u8d85\u53c2\u6570\u5982\u5206\u8fa8\u7387\u3001\u5b66\u4e60\u7387\u3001\u8bad\u7ec3\u5468\u671f\u548c\u5176\u4ed6\u4efb\u52a1\u7279\u5b9a\u8bbe\u7f6e\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e09\u3001\u67b6\u6784\u7ec6\u8282<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u56fe\u50cf+\u6587\u672c\u8f93\u5165\uff0c\u6587\u672c\u8f93\u51fa<\/strong>\uff1a\u6a21\u578b\u63a5\u53d7\u4e00\u5f20\u6216\u591a\u5f20\u56fe\u50cf\u548c\u4e00\u4e2a\u6587\u672c\u63cf\u8ff0\u4f5c\u4e3a\u8f93\u5165\uff0c\u5e76\u751f\u6210\u4e00\u4e2a\u6587\u672c\u5b57\u7b26\u4e32\u4f5c\u4e3a\u8f93\u51fa\u3002<\/li>\n\n\n\n<li><strong>\u6ce8\u610f\u529b\u673a\u5236<\/strong>\uff1a\u5bf9\u56fe\u50cf\u548c\u524d\u7f00\u6807\u8bb0\u4f7f\u7528\u5168\u6ce8\u610f\u529b\uff0c\u5bf9\u540e\u7f00\u6807\u8bb0\u4f7f\u7528\u81ea\u56de\u5f52\u6ce8\u610f\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">\u56db\u3001\u9884\u8bad\u7ec3\u76ee\u6807<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u56fe\u50cf\u63cf\u8ff0<\/strong>\uff1a\u4f7f\u7528 WebLI \u548c CC3M-35L \u6570\u636e\u96c6\u8fdb\u884c\u591a\u8bed\u8a00\u56fe\u50cf\u63cf\u8ff0\u3002<\/li>\n\n\n\n<li><strong>\u5149\u5b66\u5b57\u7b26\u8bc6\u522b\uff08OCR\uff09<\/strong>\uff1a\u8fde\u63a5\u56fe\u50cf\u4e2d\u8f6c\u5f55\u7684\u6240\u6709\u6587\u672c\u3002<\/li>\n\n\n\n<li><strong>\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09<\/strong>\uff1a\u751f\u6210\u591a\u8bed\u8a00\u95ee\u9898\u5e76\u7ed9\u51fa\u82f1\u6587\u7b54\u6848\u3002<\/li>\n\n\n\n<li><strong>\u68c0\u6d4b\u548c\u5206\u5272<\/strong>\uff1a\u4f7f\u7528\u4f2a\u6807\u7b7e\u6570\u636e\u8fdb\u884c\u591a\u76ee\u6807\u68c0\u6d4b\u548c\u5206\u5272\u3002<\/li>\n\n\n\n<li><strong>\u5b9a\u4f4d\u63cf\u8ff0<\/strong>\uff1a\u751f\u6210\u56fe\u50cf\u4e2d\u7279\u5b9a\u533a\u57df\u7684\u63cf\u8ff0\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e94\u3001\u8bc4\u4f30\u548c\u7ed3\u679c<\/h4>\n\n\n\n<p>PaliGemma \u5728\u901a\u8fc7\u5fae\u8c03\u768430\u591a\u4e2a\u5b66\u672f\u57fa\u51c6\u4e0a\u8fdb\u884c\u4e86\u8bc4\u4f30\u3002\u5173\u952e\u6027\u80fd\u6307\u6807\u6db5\u76d6\u4e86\u56fe\u50cf\u63cf\u8ff0\u3001\u89c6\u89c9\u95ee\u7b54\u548c\u56fe\u50cf\u5206\u5272\u7b49\u4efb\u52a1\u3002\u7ed3\u679c\u663e\u793a\uff0c\u5728\u8fc1\u79fb\u8fc7\u7a0b\u4e2d\u589e\u52a0\u56fe\u50cf\u5206\u8fa8\u7387\u80fd\u663e\u8457\u63d0\u9ad8\u6027\u80fd\uff0c\u5c24\u5176\u662f\u5bf9\u4e8e\u9700\u8981\u9ad8\u7ec6\u8282\u7684\u4efb\u52a1\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">\u516d\u3001\u6d88\u878d\u7814\u7a76\u548c\u53d1\u73b0<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u9884\u8bad\u7ec3\u65f6\u957f<\/strong>\uff1a\u8f83\u957f\u7684\u9884\u8bad\u7ec3\u901a\u5e38\u80fd\u5e26\u6765\u66f4\u597d\u7684\u6027\u80fd\uff0c\u4f46100M\u793a\u4f8b\u7684\u9884\u8bad\u7ec3\u65f6\u957f\u5bf9\u4e8e\u5b9e\u9a8c\u6765\u8bf4\u662f\u4e00\u4e2a\u4e0d\u9519\u7684\u6298\u8877\u3002<\/li>\n\n\n\n<li><strong>\u63a9\u7801\u548c\u5b66\u4e60\u76ee\u6807<\/strong>\uff1a\u524d\u7f00-LM \u63a9\u7801\u4ec5\u5bf9\u540e\u7f00\u6807\u8bb0\u8fdb\u884c\u76d1\u7763\u6700\u4e3a\u6709\u6548\u3002<\/li>\n\n\n\n<li><strong>\u6807\u8bb0\u521d\u59cb\u5316<\/strong>\uff1a\u4f7f\u7528\u5c0f\u9ad8\u65af\u566a\u58f0\u521d\u59cb\u5316\u65b0\u6807\u8bb0\u6bd4\u5e73\u5747\u9884\u8bad\u7ec3\u5d4c\u5165\u6548\u679c\u66f4\u597d\u3002<\/li>\n\n\n\n<li><strong>\u7ec4\u4ef6\u51bb\u7ed3<\/strong>\uff1a\u5728\u9884\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4e0d\u51bb\u7ed3\u4efb\u4f55\u90e8\u5206\u7684\u6a21\u578b\u6548\u679c\u66f4\u4f73\u3002<\/li>\n\n\n\n<li><strong>\u8fde\u63a5\u5668\u8bbe\u8ba1<\/strong>\uff1a\u89c6\u89c9\u548c\u8bed\u8a00\u6a21\u578b\u4e4b\u95f4\u7684\u7ebf\u6027\u8fde\u63a5\u5668\u7684\u6027\u80fd\u4e0e\u66f4\u590d\u6742\u7684\u66ff\u4ee3\u65b9\u6848\u5982 MLP \u76f8\u5f53\u3002<\/li>\n\n\n\n<li><strong>\u5206\u8fa8\u7387<\/strong>\uff1a\u66f4\u9ad8\u7684\u56fe\u50cf\u5206\u8fa8\u7387\u6709\u5229\u4e8e\u9700\u8981\u8be6\u7ec6\u89c6\u89c9\u4fe1\u606f\u7684\u4efb\u52a1\u3002\u5efa\u8bae\u4e3a\u4e0d\u540c\u5206\u8fa8\u7387\u63d0\u4f9b\u5355\u72ec\u7684\u68c0\u67e5\u70b9\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e03\u3001\u7ed3\u8bba<\/h4>\n\n\n\n<p><strong>PaliGemma<\/strong> \u662f\u4e00\u4e2a\u591a\u529f\u80fd\u4e14\u9ad8\u6027\u80fd\u7684 VLM\uff0c\u5728\u5e7f\u6cdb\u7684\u4efb\u52a1\u4e2d\u7279\u522b\u6709\u6548\u3002\u5176\u67b6\u6784\u3001\u9884\u8bad\u7ec3\u7b56\u7565\u548c\u5fae\u8c03\u65b9\u6cd5\u4f7f\u5176\u6210\u4e3a\u4e00\u4e2a\u5f3a\u5927\u7684\u5de5\u5177\uff0c\u9002\u7528\u4e8e\u5b66\u672f\u57fa\u51c6\u6d4b\u8bd5\u548c\u5b9e\u9645\u5e94\u7528\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>Page for the paper on Hugging Face: <a href=\"https:\/\/huggingface.co\/papers\/2407.07726\" target=\"_blank\" rel=\"noreferrer noopener\">PaliGemma: A versatile 3B VLM for transfer<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>PaliGemma \u662f\u4e00\u4e2a\u5f00\u653e\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\uff0c\u5b83\u7ed3\u5408\u4e86 SigLIP-So400m \u89c6\u89c9\u7f16\u7801\u5668\u548c G [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8,36],"tags":[39,90,80,66],"class_list":["post-4084","post","type-post","status-publish","format-standard","hentry","category-tech","category-36","tag-ai","tag-multimodal","tag-cv","tag-google"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a> <a href=\"https:\/\/nullthought.net\/?cat=36\" rel=\"category\">\u79d1\u5b66<\/a>","rttpg_excerpt":"PaliGemma \u662f\u4e00\u4e2a\u5f00\u653e\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\uff0c\u5b83\u7ed3\u5408\u4e86 SigLIP-So400m \u89c6\u89c9\u7f16\u7801\u5668\u548c G&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4084","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4084"}],"version-history":[{"count":3,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4084\/revisions"}],"predecessor-version":[{"id":5188,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4084\/revisions\/5188"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4084"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4084"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4084"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}