{"id":3161,"date":"2020-03-08T21:48:27","date_gmt":"2020-03-08T13:48:27","guid":{"rendered":"http:\/\/www.sniper97.cn\/?p=3161"},"modified":"2020-03-08T21:48:27","modified_gmt":"2020-03-08T13:48:27","slug":"%e3%80%90%e5%90%b4%e6%81%a9%e8%be%be%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e3%80%91%e5%8d%b7%e7%a7%af%e7%a5%9e%e7%bb%8f%e7%bd%91%e7%bb%9c","status":"publish","type":"post","link":"http:\/\/www.sniper97.cn\/index.php\/note\/deep-learning\/3161\/","title":{"rendered":"\u3010\u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u3011\u5377\u79ef\u795e\u7ecf\u7f51\u7edc"},"content":{"rendered":"\n<p>\u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u7b2c\u56db\u8bfe \u7b2c\u4e00\u5468 \u5377\u79ef\u795e\u7ecf\u7f51\u7edc<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"280\" height=\"346\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-72.png\" alt=\"\" class=\"wp-image-3213\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-72.png 280w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-72-243x300.png 243w\" sizes=\"(max-width: 280px) 100vw, 280px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">1.\u8ba1\u7b97\u673a\u89c6\u89c9<\/h2>\n\n\n<p> \u6df1\u5ea6\u5b66\u4e60\u4e0e\u8ba1\u7b97\u673a\u89c6\u89c9\u53ef\u4ee5\u5e2e\u52a9\u6c7d\u8f66\uff0c\u67e5\u660e\u5468\u56f4\u7684\u884c\u4eba\u548c\u6c7d\u8f66\uff0c\u5e76\u5e2e\u52a9\u6c7d\u8f66\u907f\u5f00\u5b83\u4eec\u3002\u8fd8\u4f7f\u5f97\u4eba\u8138\u8bc6\u522b\u6280\u672f\u53d8\u5f97\u66f4\u52a0\u6548\u7387\u548c\u7cbe\u51c6\uff0c\u4f60\u4eec\u5373\u5c06\u80fd\u591f\u4f53\u9a8c\u5230\u6216\u65e9\u5df2\u4f53\u9a8c\u8fc7\u4ec5\u4ec5\u901a\u8fc7\u5237\u8138\u5c31\u80fd\u89e3\u9501\u624b\u673a\u6216\u8005\u95e8\u9501\u3002\u5f53\u4f60\u89e3\u9501\u4e86\u624b\u673a\uff0c\u6211\u731c\u624b\u673a\u4e0a\u4e00\u5b9a\u6709\u5f88\u591a\u5206\u4eab\u56fe\u7247\u7684\u5e94\u7528\u3002\u5728\u4e0a\u9762\uff0c\u4f60\u80fd\u770b\u5230\u7f8e\u98df\uff0c\u9152\u5e97\u6216\u7f8e\u4e3d\u98ce\u666f\u7684\u56fe\u7247\u3002\u6709\u4e9b\u516c\u53f8\u5728\u8fd9\u4e9b\u5e94\u7528\u4e0a\u4f7f\u7528\u4e86\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\u6765\u5411\u4f60\u5c55\u793a\u6700\u4e3a\u751f\u52a8\u7f8e\u4e3d\u4ee5\u53ca\u4e0e\u4f60\u6700\u4e3a\u76f8\u5173\u7684\u56fe\u7247\u3002 <\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"370\" height=\"308\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-31.png\" alt=\"\" class=\"wp-image-3162\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-31.png 370w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-31-300x250.png 300w\" sizes=\"(max-width: 370px) 100vw, 370px\" \/><\/figure><\/div>\n\n\n<p>\u6211\u4eec\u73b0\u5728\u5df2\u7ecf\u53ef\u4ee5\u505a64*64\u5927\u5c0f\u7684\u732b\u8bc6\u522b\uff0c\u7b97\u4e0aRGB\u4e5f\u5c31\u662f64*64*3\u4e2a\u6570\u636e\uff0c\u4e5f\u5c31\u662f12288\u4e2a\uff0c\u90a3\u4e48\u5982\u679c\u6211\u4eec\u8981\u8ba1\u7b971000*1000\u5927\u5c0f\u7684\u56fe\u7247\uff0c\u5c31\u9700\u8981\u8d85\u7ea7\u591a\u7684\u53c2\u6570\uff0c\u5927\u6982300w\u4e2a\u3002<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"357\" height=\"269\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-32.png\" alt=\"\" class=\"wp-image-3163\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-32.png 357w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-32-300x226.png 300w\" sizes=\"(max-width: 357px) 100vw, 357px\" \/><\/figure><\/div>\n\n\n<p>\u56e0\u6b64\u5f15\u5165\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u6765\u89e3\u51b3\u6bd4\u8f83\u5927\u7684\u56fe\u7247\u8bad\u7ec3\u95ee\u9898\u3002<\/p>\n\n\n<h2 class=\"wp-block-heading\">2.\u8fb9\u7f18\u68c0\u6d4b\u7b97\u6cd5<\/h2>\n\n\n<p>\u5229\u7528\u8fb9\u7f18\u76d1\u6d4b\u7b97\u6cd5\u6211\u4eec\u53ef\u4ee5\u68c0\u6d4b\u51fa\u56fe\u7247\u7684\u6a2a\u8fb9\u548c\u7ad6\u8fb9<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"600\" height=\"296\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-33.png\" alt=\"\" class=\"wp-image-3164\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-33.png 600w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-33-300x148.png 300w\" sizes=\"(max-width: 600px) 100vw, 600px\" \/><\/figure><\/div>\n\n\n<p>\u5b83\u7684\u539f\u7406\u5b9e\u9645\u4e0a\u662f\u901a\u8fc7\u5377\u79ef\u64cd\u4f5c\uff0c\u91cd\u65b0\u8ba1\u7b97\u503c\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"580\" height=\"267\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-34.png\" alt=\"\" class=\"wp-image-3165\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-34.png 580w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-34-300x138.png 300w\" sizes=\"(max-width: 580px) 100vw, 580px\" \/><\/figure><\/div>\n\n\n<p>\u5982\u56fe\uff0c\u5bf9\u56fe\u7247\u5de6\u4e0a\u76843*3\u8fdb\u884c\u8ba1\u7b97\u7136\u540e\u586b\u5165\u8f93\u51fa\u7684\u6700\u5de6\u4e0a\u7684\u4f4d\u7f6e\uff0c\u7136\u540e\u4ee5\u6b64\u7c7b\u63a8<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"581\" height=\"272\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-35.png\" alt=\"\" class=\"wp-image-3166\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-35.png 581w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-35-300x140.png 300w\" sizes=\"(max-width: 581px) 100vw, 581px\" \/><\/figure><\/div>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"567\" height=\"259\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-36.png\" alt=\"\" class=\"wp-image-3167\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-36.png 567w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-36-300x137.png 300w\" sizes=\"(max-width: 567px) 100vw, 567px\" \/><\/figure><\/div>\n\n\n<p>\u8fd9\u4e2a\u64cd\u4f5c\u53eb\u5377\u79ef\u64cd\u4f5c\uff0c\u8fd9\u91cc\u53ea\u662f\u7528\u2018*\u2019\u6765\u8fdb\u884c\u8868\u793a\uff0c\u5728TensorFlow\u4e2d\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528tf.conv2d\u6765\u8fdb\u884c\u8fd9\u9879\u64cd\u4f5c\u3002<\/p>\n\n\n<p>\u4ed6\u6709\u4ec0\u4e48\u6548\u679c\u5462\uff1f\u4e3a\u4ec0\u4e48\u80fd\u591f\u8fdb\u884c\u8fb9\u754c\u5212\u5206\u5462\uff1f<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"577\" height=\"316\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-37.png\" alt=\"\" class=\"wp-image-3168\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-37.png 577w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-37-300x164.png 300w\" sizes=\"(max-width: 577px) 100vw, 577px\" \/><\/figure><\/div>\n\n\n<p>\u6211\u4eec\u53ef\u4ee5\u770b\u5230\u786e\u5b9e\u5728\u8fb9\u754c\u4e0a\u505a\u51fa\u4e86\u533a\u5206\uff0c\u540c\u65f6\u8fd8\u53ef\u4ee5\u6709\u660e\u6697\u6216\u8005\u6697\u660e\u533a\u5206\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"575\" height=\"158\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-38.png\" alt=\"\" class=\"wp-image-3169\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-38.png 575w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-38-300x82.png 300w\" sizes=\"(max-width: 575px) 100vw, 575px\" \/><\/figure><\/div>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"648\" height=\"207\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-39.png\" alt=\"\" class=\"wp-image-3170\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-39.png 648w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-39-300x96.png 300w\" sizes=\"(max-width: 648px) 100vw, 648px\" \/><\/figure><\/div>\n\n\n<p>\u540c\u7406\uff0c\u5982\u679c\u6211\u4eec\u4f7f\u7528\u6a2a\u5411\u7684\u5377\u79ef\u77e9\u9635\uff0c\u6211\u4eec\u4e5f\u53ef\u4ee5\u83b7\u5f97\u6a2a\u5411\u7684\u8fb9\u754c<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"622\" height=\"197\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-40.png\" alt=\"\" class=\"wp-image-3171\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-40.png 622w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-40-300x95.png 300w\" sizes=\"(max-width: 622px) 100vw, 622px\" \/><\/figure><\/div>\n\n\n<p>\u800c\u5377\u79ef\u77e9\u9635\u4e2d\u7684\u53c2\u6570\u6211\u4eec\u4e5f\u53ef\u4ee5\u53d8\uff08\u4e0b\u56fe\u5206\u522b\u662f Sobel\u8fc7\u6ee4\u5668\u548cScharr\u8fc7\u6ee4\u5668 \uff09\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"631\" height=\"157\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-41.png\" alt=\"\" class=\"wp-image-3172\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-41.png 631w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-41-300x75.png 300w\" sizes=\"(max-width: 631px) 100vw, 631px\" \/><\/figure><\/div>\n\n\n<p>\u751a\u81f3\u53ef\u4ee5\u628a\u4ed6\u4eec\u5f53\u505a\u53c2\u6570\u6765\u4f7f\u7528\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"622\" height=\"198\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-42.png\" alt=\"\" class=\"wp-image-3173\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-42.png 622w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-42-300x95.png 300w\" sizes=\"(max-width: 622px) 100vw, 622px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">3.Padding<\/h2>\n\n\n<p>\u6211\u4eec\u53d1\u73b0\u6bcf\u7ecf\u8fc7\u5377\u79ef\u56fe\u7247\u90fd\u53d8\u5c0f\u4e86\uff0c\u4e3a\u4e86\u907f\u514d\u8fd9\u79cd\u60c5\u51b5\uff0c\u6211\u4eec\u5c06\u539f\u56fe\u7247\u6269\u5927\uff0c\u8fd9\u6837\u5c31\u4e0d\u4f1a\u53d8\u5c0f\u4e86<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"613\" height=\"290\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-59.png\" alt=\"\" class=\"wp-image-3199\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-59.png 613w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-59-300x142.png 300w\" sizes=\"(max-width: 613px) 100vw, 613px\" \/><\/figure><\/div>\n\n\n<p>\u6211\u4eec\u79f0\u6ca1\u6709\u4f7f\u7528padding\u7684\u53eb\u505aValid\u5377\u79ef\uff08\u5377\u79ef\u540e\u56fe\u7247\u53d8\u5c0f\uff09\uff0c\u4f7f\u7528Padding\u7684\u79f0\u4f5cSame\u5377\u79ef\uff08\u5377\u79ef\u540e\u56fe\u7247\u4e0d\u53d8\u5c0f\uff09\u3002<\/p>\n\n\n<h2 class=\"wp-block-heading\">4.\u5377\u79ef\u6b65\u957f<\/h2>\n\n\n<p>\u8fd9\u4e2a\u4e5f\u633a\u597d\u89e3\u91ca\uff0c\u4e4b\u524d\u6211\u4eec\u662f\u4e00\u6b65\u4e00\u6b65\u79fb\u52a8\uff0c\u8fd9\u65f6\u5019\u5377\u79ef\u6b65\u957f\u5c31\u4e3a1.<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"522\" height=\"195\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-60.png\" alt=\"\" class=\"wp-image-3200\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-60.png 522w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-60-300x112.png 300w\" sizes=\"(max-width: 522px) 100vw, 522px\" \/><\/figure><\/div>\n\n\n<p>\u8fd9\u662f\u6b65\u957f\u4e3a2\u65f6\u7684\u60c5\u51b5<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"522\" height=\"225\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-61.png\" alt=\"\" class=\"wp-image-3201\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-61.png 522w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-61-300x129.png 300w\" sizes=\"(max-width: 522px) 100vw, 522px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">5.\u4e09\u7ef4\u5377\u79ef<\/h2>\n\n\n<p>\u6211\u4eec\u4e86\u89e3\u4e86\u5377\u79ef\u7684\u64cd\u4f5c\uff0c\u90a3\u4e48\u5982\u4f55\u5904\u7406\u56fe\u7247\u7684RGB\u6570\u636e\u5462\uff1f\u5c31\u662f\u4f7f\u7528\u4e09\u7ef4\u5377\u79ef\uff0c\u4e5f\u5c31\u662f\u53ef\u4ee5\u7406\u89e3\u6210\u8fdb\u884c\u4e09\u6b21\u5377\u79ef\u64cd\u4f5c<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"568\" height=\"247\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-63.png\" alt=\"\" class=\"wp-image-3203\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-63.png 568w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-63-300x130.png 300w\" sizes=\"(max-width: 568px) 100vw, 568px\" \/><\/figure><\/div>\n\n\n<p>\u76f8\u5f53\u4e8e\u53ea\u7528\u6b63\u65b9\u4f53\u7684\u5757\u8fdb\u884c\u68c0\u67e5<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"633\" height=\"349\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-64.png\" alt=\"\" class=\"wp-image-3204\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-64.png 633w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-64-300x165.png 300w\" sizes=\"(max-width: 633px) 100vw, 633px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">6.\u7b80\u5355\u5377\u79ef\u7f51\u7edc\u793a\u4f8b<\/h2>\n\n\n<p>\u901a\u8fc7\u5377\u79ef\u7f51\u7edc\uff0c\u5c06\u4e00\u4e2a\u56fe\u7247\u53d8\u5c0f\uff0c\u7279\u5f81\u53d8\u5927\uff08\u76f8\u5f53\u4e8e\u68c0\u6d4b\u4e8620\u4e2a\u7279\u5f81\uff09<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"619\" height=\"343\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-65.png\" alt=\"\" class=\"wp-image-3205\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-65.png 619w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-65-300x166.png 300w\" sizes=\"(max-width: 619px) 100vw, 619px\" \/><\/figure><\/div>\n\n\n<p>\u6211\u4eec\u5f97\u5230\u6700\u540e\u4e00\u4e2a\u957f\u65b9\u4f53\u4e4b\u540e\uff0c\u4f7f\u7528sigmoid\u6216\u8005softmax\uff08\u8fd9\u53d6\u51b3\u4e8e\u662f\u8981\u8fdb\u884c\u4e8c\u5206\u7c7b\u8fd8\u662f\u591a\u5206\u7c7b\uff09\u5c31\u53ef\u4ee5\u5f97\u5230\u7ed3\u679c\u3002<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"434\" height=\"208\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-66.png\" alt=\"\" class=\"wp-image-3206\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-66.png 434w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-66-300x144.png 300w\" sizes=\"(max-width: 434px) 100vw, 434px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">7.\u6c60\u5316\u5c42<\/h2>\n\n\n<p><strong>\u6700\u5927\u6c60\u5316<\/strong><\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"453\" height=\"206\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-67.png\" alt=\"\" class=\"wp-image-3207\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-67.png 453w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-67-300x136.png 300w\" sizes=\"(max-width: 453px) 100vw, 453px\" \/><\/figure><\/div>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"313\" height=\"146\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-68.png\" alt=\"\" class=\"wp-image-3208\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-68.png 313w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-68-300x140.png 300w\" sizes=\"(max-width: 313px) 100vw, 313px\" \/><\/figure><\/div>\n\n\n<p>\u5c31\u662f\u5728\u4e00\u4e2a\u5927\u5c0f\u5185\u627e\u6700\u5927\u503c\uff0c\u7528\u9014\u5462\u53ef\u80fd\u5c31\u662f\u9632\u6b62\u56fe\u7247\u592a\u5927\u4e00\u4e2a\u7279\u5f81\u7684\u63d0\u53d6\uff0c\u56e0\u4e3a\u6ca1\u6709\u53c2\u6570\uff0c\u53ea\u662f\u4e00\u4e2a\u56fa\u5b9a\u7684\u6b65\u9aa4\uff0c\u6240\u4ee5\u4ed6\u7684\u8d85\u53c2\u5c31\u53ea\u662f\u7a97\u53e3\u5927\u5c0f\u548c\u6b65\u6570\u3002<\/p>\n\n\n<p><strong>\u8fd8\u6709\u5e73\u5747\u6c60\u5316<\/strong><\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"442\" height=\"247\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-69.png\" alt=\"\" class=\"wp-image-3209\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-69.png 442w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-69-300x168.png 300w\" sizes=\"(max-width: 442px) 100vw, 442px\" \/><\/figure><\/div>\n\n\n<p>\u76ee\u524d\u6765\u8bf4\u6700\u5927\u6c60\u5316\u6bd4\u5e73\u5747\u6c60\u5316\u66f4\u5e38\u7528\u3002<\/p>\n\n\n<h2 class=\"wp-block-heading\">8.\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u793a\u4f8b<\/h2>\n\n\n<p>\u5982\u4e0b\u56fe\uff0c\u5c31\u662f\u4e00\u4e2a\u8f83\u4e3a\u5b8c\u6574\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7684\u7ed3\u6784\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"590\" height=\"315\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-70.png\" alt=\"\" class=\"wp-image-3210\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-70.png 590w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-70-300x160.png 300w\" sizes=\"(max-width: 590px) 100vw, 590px\" \/><\/figure><\/div>\n\n\n<p>\u9996\u5148\u5c31\u662f\u4e00\u4e2a32*32\u7684RGB\u56fe\u7247\uff0c\u7136\u540e\u7ecf\u8fc7\u7b2c\u4e00\u6b21\u5377\u79ef\uff08\u56e0\u4e3a\u6c60\u5316\u6ca1\u6709\u53c2\u6570\uff0c\u6240\u4ee5\u5f88\u591a\u4eba\u8ba4\u4e3a\u5377\u79ef\u548c\u6c60\u5316\u5c5e\u4e8e\u4e00\u5c42\uff09\uff0c\u53d8\u6210\u4e8610*10*16\uff0c\u76f8\u5f53\u4e8e\u6211\u4eec\u63d0\u53d6\u51fa\u6765\u4e8616\u4e2a\u7279\u5f81\uff0c\u7ee7\u7eed\u5411\u4e0b\uff0c\u7ecf\u8fc7\u7b2c\u4e8c\u5c42\u6211\u4eec\u7ee7\u7eed\u53d8\u62105*5*16\uff0c\u6c34\u5e73\u5c55\u5f00\u5c31\u662f400\u4e2a\u53c2\u6570\uff0c\u7136\u540e\u4f7f\u7528\u5168\u8fde\u63a5\u7f51\u7edc\uff0c\u7ee7\u7eed\u5411\u4e0b\u8bad\u7ec3\uff0c120\u4e2a\u7ed3\u70b9-&gt;-&gt;\u76f4\u5230\u51fa\u6765\u6211\u4eec\u60f3\u8981\u7684\u7ed3\u679c<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"696\" height=\"346\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-71.png\" alt=\"\" class=\"wp-image-3211\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-71.png 696w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-71-300x149.png 300w\" sizes=\"(max-width: 696px) 100vw, 696px\" \/><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading\">9.\u4e3a\u4ec0\u4e48\u4f7f\u7528\u5377\u79ef\uff1f<\/h2>\n\n\n<p>\u548c\u5168\u8fde\u63a5\u5c42\u76f8\u6bd4\uff0c\u5377\u79ef\u7684\u4e24\u4e2a\u4f18\u52bf\u4e3b\u8981\u5728\u53c2\u6570\u5171\u4eab\u548c\u7a00\u758f\u8fde\u63a5\u3002<\/p>\n\n\n<h2 class=\"wp-block-heading\">\u6d4b\u9a8c<\/h2>\n\n\n<p><strong>1. \u4f60\u8ba4\u4e3a\u628a\u4e0b\u9762\u8fd9\u4e2a\u8fc7\u6ee4\u5668\u5e94\u7528\u5230\u7070\u5ea6\u56fe\u50cf\u4f1a\u600e\u4e48\u6837\uff1f  <\/strong><\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"171\" height=\"114\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-73.png\" alt=\"\" class=\"wp-image-3214\"\/><\/figure><\/div>\n\n\n<ol><li> \u4f1a\u68c0\u6d4b45\u5ea6\u8fb9\u7f18 <\/li><li> \u4f1a\u68c0\u6d4b\u5782\u76f4\u8fb9\u7f18 <\/li><li> \u4f1a\u68c0\u6d4b\u6c34\u5e73\u8fb9\u7f18 <\/li><li> \u4f1a\u68c0\u6d4b\u56fe\u50cf\u5bf9\u6bd4\u5ea6 <\/li><\/ol>\n\n\n<p>2\u3002<\/p>\n\n\n<p><strong>2. \u5047\u8bbe\u4f60\u7684\u8f93\u5165\u662f\u4e00\u4e2a300\u00d7300\u7684\u5f69\u8272\uff08RGB\uff09\u56fe\u50cf\uff0c\u800c\u4f60\u6ca1\u6709\u4f7f\u7528\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u3002 \u5982\u679c\u7b2c\u4e00\u4e2a\u9690\u85cf\u5c42\u6709100\u4e2a\u795e\u7ecf\u5143\uff0c\u6bcf\u4e2a\u795e\u7ecf\u5143\u4e0e\u8f93\u5165\u5c42\u8fdb\u884c\u5168\u8fde\u63a5\uff0c\u90a3\u4e48\u8fd9\u4e2a\u9690\u85cf\u5c42\u6709\u591a\u5c11\u4e2a\u53c2\u6570\uff08\u5305\u62ec\u504f\u7f6e\u53c2\u6570\uff09\uff1f <\/strong><\/p>\n\n\n<p> 27,000,100 \uff08300*300*3*100+100\uff08\u504f\u7f6e\uff09\uff09<\/p>\n\n\n<p><strong>3. \u5047\u8bbe\u4f60\u7684\u8f93\u5165\u662f300\u00d7300\u5f69\u8272\uff08RGB\uff09\u56fe\u50cf\uff0c\u5e76\u4e14\u4f60\u4f7f\u7528\u5377\u79ef\u5c42\u548c100\u4e2a\u8fc7\u6ee4\u5668\uff0c\u6bcf\u4e2a\u8fc7\u6ee4\u5668\u90fd\u662f5\u00d75\u7684\u5927\u5c0f\uff0c\u8bf7\u95ee\u8fd9\u4e2a\u9690\u85cf\u5c42\u6709\u591a\u5c11\u4e2a\u53c2\u6570\uff08\u5305\u62ec\u504f\u7f6e\u53c2\u6570\uff09\uff1f <\/strong><\/p>\n\n\n<p>7600\uff08 \uff085*5*3+1\uff09*00 \uff09<\/p>\n\n\n<p><strong>4. \u4f60\u6709\u4e00\u4e2a63x63x16\u7684\u8f93\u5165\uff0c\u5e76\u4f7f\u7528\u5927\u5c0f\u4e3a7&#215;7\u768432\u4e2a\u8fc7\u6ee4\u5668\u8fdb\u884c\u5377\u79ef\uff0c\u4f7f\u7528\u6b65\u5e45\u4e3a2\u548c\u65e0\u586b\u5145\uff0c\u8bf7\u95ee\u8f93\u51fa\u662f\u591a\u5c11\uff1f <\/strong><\/p>\n\n\n<p>29*29*32 \u3010\uff08\uff08 63+2*0-7 \uff09\/2\uff09+1\u30110\u662fpadding<\/p>\n\n\n<p><strong>5. \u4f60\u6709\u4e00\u4e2a15x15x8\u7684\u8f93\u5165\uff0c\u5e76\u4f7f\u7528\u201cpad = 2\u201d\u8fdb\u884c\u586b\u5145\uff0c\u586b\u5145\u540e\u7684\u5c3a\u5bf8\u662f\u591a\u5c11\uff1f <\/strong><\/p>\n\n\n<p>19*19*8<\/p>\n\n\n<p><strong>6. \u4f60\u6709\u4e00\u4e2a63x63x16\u7684\u8f93\u5165\uff0c\u670932\u4e2a\u8fc7\u6ee4\u5668\u8fdb\u884c\u5377\u79ef\uff0c\u6bcf\u4e2a\u8fc7\u6ee4\u5668\u7684\u5927\u5c0f\u4e3a7&#215;7\uff0c\u6b65\u5e45\u4e3a1\uff0c\u4f60\u60f3\u8981\u4f7f\u7528\u201csame\u201d\u7684\u5377\u79ef\u65b9\u5f0f\uff0c\u8bf7\u95eepad\u7684\u503c\u662f\u591a\u5c11\uff1f <\/strong><\/p>\n\n\n<p>3\u3010\uff087-1\uff09\/2\u3011<\/p>\n\n\n<p><strong>7. \u4f60\u6709\u4e00\u4e2a32x32x16\u7684\u8f93\u5165\uff0c\u5e76\u4f7f\u7528\u6b65\u5e45\u4e3a2\u3001\u8fc7\u6ee4\u5668\u5927\u5c0f\u4e3a2\u7684\u6700\u5927\u5316\u6c60\uff0c\u8bf7\u95ee\u8f93\u51fa\u662f\u591a\u5c11\uff1f <\/strong><\/p>\n\n\n<p>16*16*16 <\/p>\n\n\n<p><strong>8. \u56e0\u4e3a\u6c60\u5316\u5c42\u4e0d\u5177\u6709\u53c2\u6570\uff0c\u6240\u4ee5\u5b83\u4eec\u4e0d\u5f71\u54cd\u53cd\u5411\u4f20\u64ad\u7684\u8ba1\u7b97\u3002 <\/strong><\/p>\n\n\n<p>\u9519\u8bef\u3002<\/p>\n\n\n<p><strong>9. \u5728\u89c6\u9891\u4e2d\uff0c\u6211\u4eec\u8c08\u5230\u4e86\u201c\u53c2\u6570\u5171\u4eab\u201d\u662f\u4f7f\u7528\u5377\u79ef\u7f51\u7edc\u7684\u597d\u5904\u3002\u5173\u4e8e\u53c2\u6570\u5171\u4eab\u7684\u4e0b\u5217\u54ea\u4e2a\u9648\u8ff0\u662f\u6b63\u786e\u7684\uff1f\uff08\u68c0\u67e5\u6240\u6709\u9009\u9879\u3002\uff09 <\/strong><\/p>\n\n\n<ol><li> \u5b83\u51cf\u5c11\u4e86\u53c2\u6570\u7684\u603b\u6570\uff0c\u4ece\u800c\u51cf\u5c11\u8fc7\u62df\u5408\u3002 <\/li><li> \u5b83\u5141\u8bb8\u5728\u6574\u4e2a\u8f93\u5165\u503c\u7684\u591a\u4e2a\u4f4d\u7f6e\u4f7f\u7528\u7279\u5f81\u68c0\u6d4b\u5668\u3002<\/li><li> \u5b83\u5141\u8bb8\u4e3a\u4e00\u9879\u4efb\u52a1\u5b66\u4e60\u7684\u53c2\u6570\u5373\u4f7f\u5bf9\u4e8e\u4e0d\u540c\u7684\u4efb\u52a1\u4e5f\u53ef\u4ee5\u5171\u4eab\uff08\u8fc1\u79fb\u5b66\u4e60\uff09\u3002  <\/li><li> \u5b83\u5141\u8bb8\u68af\u5ea6\u4e0b\u964d\u5c06\u8bb8\u591a\u53c2\u6570\u8bbe\u7f6e\u4e3a\u96f6\uff0c\u4ece\u800c\u4f7f\u5f97\u8fde\u63a5\u7a00\u758f\u3002 <\/li><\/ol>\n\n\n<p>2\uff0c4\u3002<\/p>\n\n\n<p><strong>10. \u5728\u8bfe\u5802\u4e0a\uff0c\u6211\u4eec\u8ba8\u8bba\u4e86\u201c\u7a00\u758f\u8fde\u63a5\u201d\u662f\u4f7f\u7528\u5377\u79ef\u5c42\u7684\u597d\u5904\u3002\u8fd9\u662f\u4ec0\u4e48\u610f\u601d? <\/strong><\/p>\n\n\n<ol><li>  \u6b63\u5219\u5316\u5bfc\u81f4\u68af\u5ea6\u4e0b\u964d\u5c06\u8bb8\u591a\u53c2\u6570\u8bbe\u7f6e\u4e3a\u96f6\u3002 <\/li><li> \u6bcf\u4e2a\u8fc7\u6ee4\u5668\u90fd\u8fde\u63a5\u5230\u4e0a\u4e00\u5c42\u7684\u6bcf\u4e2a\u901a\u9053\u3002 <\/li><li> \u4e0b\u4e00\u5c42\u4e2d\u7684\u6bcf\u4e2a\u6fc0\u6d3b\u53ea\u4f9d\u8d56\u4e8e\u524d\u4e00\u5c42\u7684\u5c11\u91cf\u6fc0\u6d3b\u3002 <\/li><li> \u5377\u79ef\u7f51\u7edc\u4e2d\u7684\u6bcf\u4e00\u5c42\u53ea\u8fde\u63a5\u5230\u53e6\u5916\u4e24\u5c42\u3002 <\/li><\/ol>\n\n\n<p>3\u3002<\/p>\n\n\n<h2 class=\"wp-block-heading\">\u7f16\u7a0b\u4f5c\u4e1a<\/h2>\n\n\n<p>\u8fd9\u8282\u8bfe\u521a\u5f00\u59cb\u5148\u7b80\u5355\u4ecb\u7ecd\u624b\u5199\u5377\u79ef\u7f51\u7edc\u76f8\u5173\u77e5\u8bc6\uff0c\u7136\u540e\u8bd5\u56fe\u4f7f\u7528\u5377\u79ef\u7f51\u7edc\u8fdb\u884c\u4e0a\u8282\u8bfe\u7684\u624b\u52bf\u8bc6\u522b\u3002<\/p>\n\n\n<p>\u9996\u5148\u624b\u5199\u5377\u79ef\u76f8\u5173\u6280\u672f\uff1a<\/p>\n\n\n<p>\u5148\u5bfc\u5305\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">import numpy as np<br \/>import h5py<br \/>import matplotlib.pyplot as plt<\/pre>\n\n\n<p>\u9996\u5148\u662fpadding\uff0c\u6211\u4eec\u4f7f\u7528numpy\u63d0\u4f9b\u7684pad\u53c2\u6570\u6765\u8fdb\u884cpadding\u64cd\u4f5c\uff0c\u9700\u8981\u4f20\u7684\u51e0\u4e2a\u53c2\u6570\u5206\u522b\u662farray,\u7136\u540e\u5c31\u662f\u4e09\u4e2a\u7ef4\u5ea6\u4e0a\u5206\u522b\u6269\u5c55\u591a\u5c11\u4f4d\uff08arr3D\u4e3a\u4e00\u4e2a\u4e09\u7ef4\u6570\u7ec4\uff0c\u6211\u4eec\u4e0d\u5e0c\u671b\u53d8\u62104\u7ef4\u6240\u4ee5\u7b2c\u4e00\u4e2a\u5747\u4e3a0\uff0c\u7136\u540e\u7b2c\u4e8c\u7ef4\u7b2c\u4e09\u7ef4\u524d\u540e\u5206\u522b\u52a0\u4e0a\u51e0\u4e2a0\uff09\uff0c\u7136\u540econstant\u8868\u793a\u586b\u5145\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\"># padding<br \/>arr3D = np.array([[[1, 1, 2, 2, 3, 4],<br \/>                   [1, 1, 2, 2, 3, 4],<br \/>                   [1, 1, 2, 2, 3, 4]],<br \/><br \/>                  [[0, 1, 2, 3, 4, 5],<br \/>                   [0, 1, 2, 3, 4, 5],<br \/>                   [0, 1, 2, 3, 4, 5]],<br \/><br \/>                  [[1, 1, 2, 2, 3, 4],<br \/>                   [1, 1, 2, 2, 3, 4],<br \/>                   [1, 1, 2, 2, 3, 4]]])<br \/><br \/>print('constant:  \\n' + str(np.pad(arr3D, ((0, 0), (1, 1), (2, 2)), 'constant')))<\/pre>\n\n\n<p>\u8f93\u51fa\uff1a<\/p>\n\n\n<pre class=\"wp-block-code\"><code>constant:\n[[[0 0 0 0 0 0 0 0 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 0 0 0 0 0 0 0 0]]\n [[0 0 0 0 0 0 0 0 0 0]\n  [0 0 0 1 2 3 4 5 0 0]\n  [0 0 0 1 2 3 4 5 0 0]\n  [0 0 0 1 2 3 4 5 0 0]\n  [0 0 0 0 0 0 0 0 0 0]]\n [[0 0 0 0 0 0 0 0 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 1 1 2 2 3 4 0 0]\n  [0 0 0 0 0 0 0 0 0 0]]]<\/code><\/pre>\n\n\n<p>\u6211\u4eec\u8bd5\u7740\u53ef\u89c6\u5316\u4e00\u4e0b\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def zero_pad(X, pad):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5bf9\u56fe\u7247<\/em><em>X<\/em><em>\u8fdb\u884c\u586b\u5145<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> pad:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>X_paded = np.pad(X, (<br \/>        (0, 0),  # \u6837\u672c\u6570\uff0c\u4e0d\u586b\u5145<br \/>        (pad, pad),  # \u56fe\u50cf\u9ad8\u5ea6,\u4f60\u53ef\u4ee5\u89c6\u4e3a\u4e0a\u9762\u586b\u5145x\u4e2a\uff0c\u4e0b\u9762\u586b\u5145y\u4e2a(x,y)<br \/>        (pad, pad),  # \u56fe\u50cf\u5bbd\u5ea6,\u4f60\u53ef\u4ee5\u89c6\u4e3a\u5de6\u8fb9\u586b\u5145x\u4e2a\uff0c\u53f3\u8fb9\u586b\u5145y\u4e2a(x,y)<br \/>        (0, 0)),  # \u901a\u9053\u6570\uff0c\u4e0d\u586b\u5145<br \/>                     'constant', constant_values=0)  # \u8fde\u7eed\u4e00\u6837\u7684\u503c\u586b\u5145<br \/><br \/>    return X_paded<br \/><br \/><br \/>np.random.seed(1)<br \/>x = np.random.randn(4, 3, 3, 2)<br \/>x_paded = zero_pad(x, 2)<br \/>#<br \/># # \u7ed8\u5236\u56fe<br \/>fig, axarr = plt.subplots(1, 2)  # \u4e00\u884c\u4e24\u5217<br \/>axarr[0].set_title('x')<br \/>axarr[0].imshow(x[0, :, :, 0])<br \/>axarr[1].set_title('x_paded')<br \/>axarr[1].imshow(x_paded[0, :, :, 0])<br \/>plt.show()<\/pre>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"335\" height=\"192\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-84.png\" alt=\"\" class=\"wp-image-3228\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-84.png 335w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-84-300x172.png 300w\" sizes=\"(max-width: 335px) 100vw, 335px\" \/><\/figure><\/div>\n\n\n<p>\u7136\u540e\u5c31\u662f\u5377\u79ef\u64cd\u4f5c\uff0c\u5377\u79ef\u64cd\u4f5c\u7684\u52a8\u56fe\u5982\u4e0b\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"526\" height=\"384\" src=\"\/wp-content\/uploads\/2020\/03\/Convolution_schematic.gif\" alt=\"\" class=\"wp-image-3231\"\/><\/figure><\/div>\n\n\n<p>\u8ba1\u7b97\u5377\u79ef\u540e\u56fe\u7247\u5927\u5c0f\u516c\u5f0f\u5982\u4e0b\uff1a<\/p>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"330\" height=\"170\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-86.png\" alt=\"\" class=\"wp-image-3232\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-86.png 330w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-86-300x155.png 300w\" sizes=\"(max-width: 330px) 100vw, 330px\" \/><\/figure><\/div>\n\n\n<pre class=\"wp-block-preformatted\">def conv_single_step(a_slice_prev, W, b):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u8fdb\u884c\u5377\u79ef\u64cd\u4f5c<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> a_slice_prev:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> W:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> b:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>s = np.multiply(a_slice_prev, W) + b<br \/><br \/>    Z = np.sum(s)<br \/><br \/>    return Z<br \/><br \/><br \/>def conv_forward(A_prev, W, b, hparameters):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5377\u79ef\u524d\u5411\u4f20\u64ad<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> A_prev:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> W:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> b:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> hparameters:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em># \u83b7\u53d6\u6765\u81ea\u4e0a\u4e00\u5c42\u6570\u636e\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape<br \/>    # \u83b7\u53d6\u6743\u91cd\u77e9\u9635\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (f, f, n_C_prev, n_C) = W.shape<br \/>    # \u83b7\u53d6\u8d85\u53c2\u503c<br \/>    stride = hparameters[\"stride\"]<br \/>    pad = hparameters[\"pad\"]<br \/><br \/>    # \u8ba1\u7b97\u5377\u79ef\u540e\u7684\u56fe\u50cf\u7684\u5bbd\u5ea6\u9ad8\u5ea6\uff0c\u53c2\u8003\u4e0a\u9762\u7684\u516c\u5f0f\uff0c\u4f7f\u7528int()\u6765\u8fdb\u884c\u677f\u9664<br \/>    n_H = int((n_H_prev - f + 2 * pad) \/ stride) + 1<br \/>    n_W = int((n_W_prev - f + 2 * pad) \/ stride) + 1<br \/><br \/>    # \u4f7f\u75280\u6765\u521d\u59cb\u5316\u5377\u79ef\u8f93\u51faZ<br \/>    Z = np.zeros((m, n_H, n_W, n_C))<br \/><br \/>    # \u901a\u8fc7A_prev\u521b\u5efa\u586b\u5145\u8fc7\u4e86\u7684A_prev_pad<br \/>    A_prev_pad = zero_pad(A_prev, pad)<br \/><br \/>    for i in range(m):  # \u904d\u5386\u6837\u672c<br \/>        a_prev_pad = A_prev_pad[i]  # \u9009\u62e9\u7b2ci\u4e2a\u6837\u672c\u7684\u6269\u5145\u540e\u7684\u6fc0\u6d3b\u77e9\u9635<br \/>        for h in range(n_H):  # \u5728\u8f93\u51fa\u7684\u5782\u76f4\u8f74\u4e0a\u5faa\u73af<br \/>            for w in range(n_W):  # \u5728\u8f93\u51fa\u7684\u6c34\u5e73\u8f74\u4e0a\u5faa\u73af<br \/>                for c in range(n_C):  # \u5faa\u73af\u904d\u5386\u8f93\u51fa\u7684\u901a\u9053<br \/>                    # \u5b9a\u4f4d\u5f53\u524d\u7684\u5207\u7247\u4f4d\u7f6e<br \/>                    vert_start = h * stride  # \u7ad6\u5411\uff0c\u5f00\u59cb\u7684\u4f4d\u7f6e<br \/>                    vert_end = vert_start + f  # \u7ad6\u5411\uff0c\u7ed3\u675f\u7684\u4f4d\u7f6e<br \/>                    horiz_start = w * stride  # \u6a2a\u5411\uff0c\u5f00\u59cb\u7684\u4f4d\u7f6e<br \/>                    horiz_end = horiz_start + f  # \u6a2a\u5411\uff0c\u7ed3\u675f\u7684\u4f4d\u7f6e<br \/>                    # \u5207\u7247\u4f4d\u7f6e\u5b9a\u4f4d\u597d\u4e86\u6211\u4eec\u5c31\u628a\u5b83\u53d6\u51fa\u6765,\u9700\u8981\u6ce8\u610f\u7684\u662f\u6211\u4eec\u662f\u201c\u7a7f\u900f\u201d\u53d6\u51fa\u6765\u7684\uff0c<br \/>                    # \u81ea\u884c\u8111\u8865\u4e00\u4e0b\u5438\u7ba1\u63d2\u5165\u4e00\u5c42\u5c42\u7684\u6a61\u76ae\u6ce5\u5c31\u660e\u767d\u4e86<br \/>                    a_slice_prev = a_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :]<br \/>                    # \u6267\u884c\u5355\u6b65\u5377\u79ef<br \/>                    Z[i, h, w, c] = conv_single_step(a_slice_prev, W[:, :, :, c], b[0, 0, 0, c])<br \/><br \/>    # \u6570\u636e\u5904\u7406\u5b8c\u6bd5\uff0c\u9a8c\u8bc1\u6570\u636e\u683c\u5f0f\u662f\u5426\u6b63\u786e<br \/>    assert (Z.shape == (m, n_H, n_W, n_C))<br \/><br \/>    # \u5b58\u50a8\u4e00\u4e9b\u7f13\u5b58\u503c\uff0c\u4ee5\u4fbf\u4e8e\u53cd\u5411\u4f20\u64ad\u4f7f\u7528<br \/>    cache = (A_prev, W, b, hparameters)<br \/><br \/>    return (Z, cache)<\/pre>\n\n\n<p>\u7136\u540e\u5c31\u662f\u6c60\u5316\u5c42\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def pool_forward(A_prev, hparameters, mode=\"max\"):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u6c60\u5316\u5c42<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> A_prev:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> hparameters:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> mode:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em><br \/><\/em><em>    <\/em># \u83b7\u53d6\u8f93\u5165\u6570\u636e\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape<br \/><br \/>    # \u83b7\u53d6\u8d85\u53c2\u6570\u7684\u4fe1\u606f<br \/>    f = hparameters[\"f\"]<br \/>    stride = hparameters[\"stride\"]<br \/><br \/>    # \u8ba1\u7b97\u8f93\u51fa\u7ef4\u5ea6<br \/>    n_H = int((n_H_prev - f) \/ stride) + 1<br \/>    n_W = int((n_W_prev - f) \/ stride) + 1<br \/>    n_C = n_C_prev<br \/><br \/>    # \u521d\u59cb\u5316\u8f93\u51fa\u77e9\u9635<br \/>    A = np.zeros((m, n_H, n_W, n_C))<br \/><br \/>    for i in range(m):  # \u904d\u5386\u6837\u672c<br \/>        for h in range(n_H):  # \u5728\u8f93\u51fa\u7684\u5782\u76f4\u8f74\u4e0a\u5faa\u73af<br \/>            for w in range(n_W):  # \u5728\u8f93\u51fa\u7684\u6c34\u5e73\u8f74\u4e0a\u5faa\u73af<br \/>                for c in range(n_C):  # \u5faa\u73af\u904d\u5386\u8f93\u51fa\u7684\u901a\u9053<br \/>                    # \u5b9a\u4f4d\u5f53\u524d\u7684\u5207\u7247\u4f4d\u7f6e<br \/>                    vert_start = h * stride  # \u7ad6\u5411\uff0c\u5f00\u59cb\u7684\u4f4d\u7f6e<br \/>                    vert_end = vert_start + f  # \u7ad6\u5411\uff0c\u7ed3\u675f\u7684\u4f4d\u7f6e<br \/>                    horiz_start = w * stride  # \u6a2a\u5411\uff0c\u5f00\u59cb\u7684\u4f4d\u7f6e<br \/>                    horiz_end = horiz_start + f  # \u6a2a\u5411\uff0c\u7ed3\u675f\u7684\u4f4d\u7f6e<br \/>                    # \u5b9a\u4f4d\u5b8c\u6bd5\uff0c\u5f00\u59cb\u5207\u5272<br \/>                    a_slice_prev = A_prev[i, vert_start:vert_end, horiz_start:horiz_end, c]<br \/><br \/>                    # \u5bf9\u5207\u7247\u8fdb\u884c\u6c60\u5316\u64cd\u4f5c<br \/>                    if mode == \"max\":<br \/>                        A[i, h, w, c] = np.max(a_slice_prev)<br \/>                    elif mode == \"average\":<br \/>                        A[i, h, w, c] = np.mean(a_slice_prev)<br \/><br \/>    # \u6c60\u5316\u5b8c\u6bd5\uff0c\u6821\u9a8c\u6570\u636e\u683c\u5f0f<br \/>    assert (A.shape == (m, n_H, n_W, n_C))<br \/><br \/>    # \u6821\u9a8c\u5b8c\u6bd5\uff0c\u5f00\u59cb\u5b58\u50a8\u7528\u4e8e\u53cd\u5411\u4f20\u64ad\u7684\u503c<br \/>    cache = (A_prev, hparameters)<br \/><br \/>    return A, cache<\/pre>\n\n\n<p>\u7136\u540e\u5c31\u662f\u5377\u79ef\u7684\u53cd\u5411\u4f20\u64ad\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def conv_backward(dZ, cache):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5377\u79ef\u5c42\u53cd\u5411\u4f20\u64ad<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> dZ:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> cache:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em># \u83b7\u53d6cache\u7684\u503c<br \/>    (A_prev, W, b, hparameters) = cache<br \/><br \/>    # \u83b7\u53d6A_prev\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape<br \/><br \/>    # \u83b7\u53d6dZ\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (m, n_H, n_W, n_C) = dZ.shape<br \/><br \/>    # \u83b7\u53d6\u6743\u503c\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (f, f, n_C_prev, n_C) = W.shape<br \/><br \/>    # \u83b7\u53d6hparaeters\u7684\u503c<br \/>    pad = hparameters[\"pad\"]<br \/>    stride = hparameters[\"stride\"]<br \/><br \/>    # \u521d\u59cb\u5316\u5404\u4e2a\u68af\u5ea6\u7684\u7ed3\u6784<br \/>    dA_prev = np.zeros((m, n_H_prev, n_W_prev, n_C_prev))<br \/>    dW = np.zeros((f, f, n_C_prev, n_C))<br \/>    db = np.zeros((1, 1, 1, n_C))<br \/><br \/>    # \u524d\u5411\u4f20\u64ad\u4e2d\u6211\u4eec\u4f7f\u7528\u4e86pad\uff0c\u53cd\u5411\u4f20\u64ad\u4e5f\u9700\u8981\u4f7f\u7528\uff0c\u8fd9\u662f\u4e3a\u4e86\u4fdd\u8bc1\u6570\u636e\u7ed3\u6784\u4e00\u81f4<br \/>    A_prev_pad = zero_pad(A_prev, pad)<br \/>    dA_prev_pad = zero_pad(dA_prev, pad)<br \/><br \/>    # \u73b0\u5728\u5904\u7406\u6570\u636e<br \/>    for i in range(m):<br \/>        # \u9009\u62e9\u7b2ci\u4e2a\u6269\u5145\u4e86\u7684\u6570\u636e\u7684\u6837\u672c,\u964d\u4e86\u4e00\u7ef4\u3002<br \/>        a_prev_pad = A_prev_pad[i]<br \/>        da_prev_pad = dA_prev_pad[i]<br \/><br \/>        for h in range(n_H):<br \/>            for w in range(n_W):<br \/>                for c in range(n_C):<br \/>                    # \u5b9a\u4f4d\u5207\u7247\u4f4d\u7f6e<br \/>                    vert_start = h<br \/>                    vert_end = vert_start + f<br \/>                    horiz_start = w<br \/>                    horiz_end = horiz_start + f<br \/><br \/>                    # \u5b9a\u4f4d\u5b8c\u6bd5\uff0c\u5f00\u59cb\u5207\u7247<br \/>                    a_slice = a_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :]<br \/><br \/>                    # \u5207\u7247\u5b8c\u6bd5\uff0c\u4f7f\u7528\u4e0a\u9762\u7684\u516c\u5f0f\u8ba1\u7b97\u68af\u5ea6<br \/>                    da_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :] += W[:, :, :, c] * dZ[i, h, w, c]<br \/>                    dW[:, :, :, c] += a_slice * dZ[i, h, w, c]<br \/>                    db[:, :, :, c] += dZ[i, h, w, c]<br \/>        # \u8bbe\u7f6e\u7b2ci\u4e2a\u6837\u672c\u6700\u7ec8\u7684dA_prev,\u5373\u628a\u975e\u586b\u5145\u7684\u6570\u636e\u53d6\u51fa\u6765\u3002<br \/>        dA_prev[i, :, :, :] = da_prev_pad[pad:-pad, pad:-pad, :]<br \/><br \/>    # \u6570\u636e\u5904\u7406\u5b8c\u6bd5\uff0c\u9a8c\u8bc1\u6570\u636e\u683c\u5f0f\u662f\u5426\u6b63\u786e<br \/>    assert (dA_prev.shape == (m, n_H_prev, n_W_prev, n_C_prev))<br \/><br \/>    return (dA_prev, dW, db)<\/pre>\n\n\n<p>\u7136\u540e\u662f\u6c60\u5316\u5c42\u7684\u53cd\u5411\u4f20\u64ad\uff0c\u56e0\u4e3a\u5bf9\u4e8e\u6700\u5927\u6c60\u5316\u5c42\u6211\u4eec\u8981\u8bb0\u5f55\u6700\u5927\u7684\u4f4d\u7f6e\u4ee5\u4fbf\u6211\u4eec\u53cd\u5411\u4f20\u64ad\uff0c\u56e0\u6b64\u9700\u8981\u591a\u4e00\u4e2a\u65b9\u6cd5\u6765\u8bb0\u5f55\u8fd9\u4e9b\u6700\u5927\u503c\u7684\u4f4d\u7f6e\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def create_mask_from_window(x):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u4ece\u8f93\u5165\u77e9\u9635\u4e2d\u521b\u5efa\u63a9\u7801\uff0c\u4ee5\u4fdd\u5b58\u6700\u5927\u503c\u7684\u77e9\u9635\u7684\u4f4d\u7f6e\u3002<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> x:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>mask = x == np.max(x)<br \/><br \/>    return mask<br \/><br \/><br \/>def distribute_value(dz, shape):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5e73\u5747\u6c60\u5316\u7684\u53cd\u5411\u4f20\u64ad<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> dz:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> shape:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em># \u83b7\u53d6\u77e9\u9635\u7684\u5927\u5c0f<br \/>    (n_H, n_W) = shape<br \/><br \/>    # \u8ba1\u7b97\u5e73\u5747\u503c<br \/>    average = dz \/ (n_H * n_W)<br \/><br \/>    # \u586b\u5145\u5165\u77e9\u9635<br \/>    a = np.ones(shape) * average<br \/><br \/>    return a<br \/><br \/><br \/>def pool_backward(dA, cache, mode=\"max\"):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5b9e\u73b0\u6c60\u5316\u5c42\u7684\u53cd\u5411\u4f20\u64ad<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> dA:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> cache:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> mode:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em># \u83b7\u53d6cache\u4e2d\u7684\u503c<br \/>    (A_prev, hparaeters) = cache<br \/><br \/>    # \u83b7\u53d6hparaeters\u7684\u503c<br \/>    f = hparaeters[\"f\"]<br \/>    stride = hparaeters[\"stride\"]<br \/><br \/>    # \u83b7\u53d6A_prev\u548cdA\u7684\u57fa\u672c\u4fe1\u606f<br \/>    (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape<br \/>    (m, n_H, n_W, n_C) = dA.shape<br \/><br \/>    # \u521d\u59cb\u5316\u8f93\u51fa\u7684\u7ed3\u6784<br \/>    dA_prev = np.zeros_like(A_prev)<br \/><br \/>    # \u5f00\u59cb\u5904\u7406\u6570\u636e<br \/>    for i in range(m):<br \/>        a_prev = A_prev[i]<br \/>        for h in range(n_H):<br \/>            for w in range(n_W):<br \/>                for c in range(n_C):<br \/>                    # \u5b9a\u4f4d\u5207\u7247\u4f4d\u7f6e<br \/>                    vert_start = h<br \/>                    vert_end = vert_start + f<br \/>                    horiz_start = w<br \/>                    horiz_end = horiz_start + f<br \/><br \/>                    # \u9009\u62e9\u53cd\u5411\u4f20\u64ad\u7684\u8ba1\u7b97\u65b9\u5f0f<br \/>                    if mode == \"max\":<br \/>                        # \u5f00\u59cb\u5207\u7247<br \/>                        a_prev_slice = a_prev[vert_start:vert_end, horiz_start:horiz_end, c]<br \/>                        # \u521b\u5efa\u63a9\u7801<br \/>                        mask = create_mask_from_window(a_prev_slice)<br \/>                        # \u8ba1\u7b97dA_prev,\u5c31\u662f\u53ea\u5728\u6700\u5927\u503c\u7684\u4f4d\u7f6e\u8fdb\u884c\u52a0\u51cf<br \/>                        dA_prev[i, vert_start:vert_end, horiz_start:horiz_end, c] += np.multiply(mask, dA[i, h, w, c])<br \/><br \/>                    elif mode == \"average\":<br \/>                        # \u83b7\u53d6dA\u7684\u503c<br \/>                        da = dA[i, h, w, c]<br \/>                        # \u5b9a\u4e49\u8fc7\u6ee4\u5668\u5927\u5c0f<br \/>                        shape = (f, f)<br \/>                        # \u5e73\u5747\u5206\u914d<br \/>                        dA_prev[i, vert_start:vert_end, horiz_start:horiz_end, c] += distribute_value(da, shape)<br \/>    # \u6570\u636e\u5904\u7406\u5b8c\u6bd5\uff0c\u5f00\u59cb\u9a8c\u8bc1\u683c\u5f0f<br \/>    assert (dA_prev.shape == A_prev.shape)<br \/><br \/>    return dA_prev<br \/><\/pre>\n\n\n<p>\u5230\u8fd9\u57fa\u672c\u4e0a\u6211\u4eec\u5c31\u624b\u52a8\u7684\u5199\u4e86\u4e00\u8fb9\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u3002<\/p>\n\n\n<p>\u7136\u540e\u6211\u4eec\u4f7f\u7528TensorFlow\u6846\u67b6\u8fdb\u884c\u624b\u52bf\u8bc6\u522b\uff1a<\/p>\n\n\n<p>\u9996\u5148\u8fd8\u662f\u5bfc\u5305\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">import math<br \/>import numpy as np<br \/>import h5py<br \/>import matplotlib.pyplot as plt<br \/>import matplotlib.image as mpimg<br \/>import tensorflow as tf<br \/>from tensorflow.python.framework import ops<br \/><br \/>from course_4_week_1 import cnn_utils<\/pre>\n\n\n<p>\u8bfb\u53d6\u6570\u636e\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">np.random.seed(1)<br \/>X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = cnn_utils.load_dataset()<\/pre>\n\n\n<p>\u7136\u540e\u8fdb\u884c\u6570\u636e\u6807\u51c6\u5316\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\"># \u6570\u636e\u6807\u51c6\u5316<br \/>X_train = X_train_orig \/ 255<br \/>X_test = X_test_orig \/ 255<br \/><br \/>Y_train = cnn_utils.convert_to_one_hot(Y_train_orig, 6).T<br \/>Y_test = cnn_utils.convert_to_one_hot(Y_test_orig, 6).T<\/pre>\n\n\n<p>\u521d\u59cb\u5316\u6211\u4eec\u7684palceholder\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def create_placeholders(n_H0, n_W0, n_C0, n_y):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521b\u5efa\u5360\u4f4d\u7b26<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_H0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_W0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_C0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_y:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>X = tf.placeholder(tf.float32, [None, n_H0, n_W0, n_C0])<br \/>    Y = tf.placeholder(tf.float32, [None, n_y])<br \/><br \/>    return X, Y<\/pre>\n\n\n<p>\u521d\u59cb\u5316\u53c2\u6570\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def initialize_parameters():<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521d\u59cb\u5316\u6743\u503c\u77e9\u9635\uff0c\u8fd9\u91cc\u6211\u4eec\u628a\u6743\u503c\u77e9\u9635\u786c\u7f16\u7801\uff1a<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em><br \/><\/em><em>    <\/em>W1 = tf.get_variable(\"W1\", [4, 4, 3, 8], initializer=tf.contrib.layers.xavier_initializer(seed=1))<br \/>    W2 = tf.get_variable(\"W2\", [2, 2, 8, 16], initializer=tf.contrib.layers.xavier_initializer(seed=1))<br \/><br \/>    parameters = {\"W1\": W1,<br \/>                  \"W2\": W2}<br \/><br \/>    return parameters<\/pre>\n\n\n<p>\u7136\u540e\u5bf9\u7f51\u7edc\u7684\u524d\u5411\u4f20\u64ad\u8fdb\u884c\u5b9a\u4e49\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def forward_propagation(X, parameters):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5b9e\u73b0\u524d\u5411\u4f20\u64ad<\/em><em><br \/><\/em><em>    CONV2D -&gt; RELU -&gt; MAXPOOL -&gt; CONV2D -&gt; RELU -&gt; MAXPOOL -&gt; FLATTEN -&gt; FULLYCONNECTED<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> parameters:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>W1 = parameters['W1']<br \/>    W2 = parameters['W2']<br \/><br \/>    # Conv2d : \u6b65\u4f10\uff1a1\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    Z1 = tf.nn.conv2d(X, W1, strides=[1, 1, 1, 1], padding=\"SAME\")<br \/>    # ReLU \uff1a<br \/>    A1 = tf.nn.relu(Z1)<br \/>    # Max pool : \u7a97\u53e3\u5927\u5c0f\uff1a8x8\uff0c\u6b65\u4f10\uff1a8x8\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    P1 = tf.nn.max_pool(A1, ksize=[1, 8, 8, 1], strides=[1, 8, 8, 1], padding=\"SAME\")<br \/><br \/>    # Conv2d : \u6b65\u4f10\uff1a1\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    Z2 = tf.nn.conv2d(P1, W2, strides=[1, 1, 1, 1], padding=\"SAME\")<br \/>    # ReLU \uff1a<br \/>    A2 = tf.nn.relu(Z2)<br \/>    # Max pool : \u8fc7\u6ee4\u5668\u5927\u5c0f\uff1a4x4\uff0c\u6b65\u4f10\uff1a4x4\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    P2 = tf.nn.max_pool(A2, ksize=[1, 4, 4, 1], strides=[1, 4, 4, 1], padding=\"SAME\")<br \/><br \/>    # \u4e00\u7ef4\u5316\u4e0a\u4e00\u5c42\u7684\u8f93\u51fa<br \/>    P = tf.contrib.layers.flatten(P2)<br \/><br \/>    # \u5168\u8fde\u63a5\u5c42\uff08FC\uff09\uff1a\u4f7f\u7528\u6ca1\u6709\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u7684\u5168\u8fde\u63a5\u5c42<br \/>    Z3 = tf.contrib.layers.fully_connected(P, 6, activation_fn=None)<br \/><br \/>    return Z3<\/pre>\n\n\n<p>\u8ba1\u7b97\u4ee3\u4ef7\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def compute_cost(Z3, Y):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u8ba1\u7b97\u4ee3\u4ef7<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Z3:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=Z3, labels=Y))<br \/><br \/>    return cost<\/pre>\n\n\n<p>\u7136\u540e\u6211\u4eec\u5b9a\u4e49\u6a21\u578b\u5e76\u8fd0\u884c\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">def model(X_train, Y_train, X_test, Y_test, learning_rate=0.009,<br \/>          num_epochs=100, minibatch_size=64, print_cost=True, isPlot=True):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521b\u5efa\u6a21\u578b<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X_train:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y_train:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X_test:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y_test:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> learning_rate:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> num_epochs:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> minibatch_size:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> print_cost:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> isPlot:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>tf.get_default_graph()<br \/>    seed = 3<br \/>    tf.set_random_seed(1)<br \/>    # m\u4e2a\u56fe\u7247  h*w\u5927\u5c0f  c\u4e2a\u5377\u79ef\u5c42<br \/>    m, n_H0, n_W0, n_C0 = X_train.shape<br \/>    # \u6709\u591a\u5c11\u79cd\u7ed3\u679c<br \/>    n_y = Y_train.shape[1]<br \/>    costs = []<br \/>    X, Y = create_placeholders(n_H0, n_W0, n_C0, n_y)<br \/>    # \u521d\u59cb\u5316\u53c2\u6570<br \/>    parameters = initialize_parameters()<br \/>    # \u524d\u5411\u4f20\u64ad<br \/>    Z3 = forward_propagation(X, parameters)<br \/>    # \u8ba1\u7b97\u6210\u672c<br \/>    cost = compute_cost(Z3, Y)<br \/>    # \u53cd\u5411\u4f20\u64ad<br \/>    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)<br \/>    # \u4fdd\u5b58\u6570\u636e<br \/>    saver = tf.train.Saver()<br \/>    # \u5168\u5c40\u521d\u59cb\u5316\u53d8\u91cf<br \/>    init = tf.global_variables_initializer()<br \/>    # \u5f00\u59cb\u8fd0\u884c<br \/>    with tf.Session() as session:<br \/>        # \u521d\u59cb\u5316\u53c2\u6570<br \/>        session.run(init)<br \/>        # \u904d\u5386\u6570\u636e<br \/>        for epoch in range(num_epochs):<br \/>            minibatch_cost = 0<br \/>            num_minibatches = int(m \/ minibatch_size)<br \/>            seed = seed + 1<br \/>            minibatches = cnn_utils.random_mini_batches(X_train, Y_train, minibatch_size, seed)<br \/><br \/>            for minibatch in minibatches:<br \/>                minibatch_X, minibatch_Y = minibatch<br \/>                _, temp_cost = session.run([optimizer, cost], feed_dict={X: minibatch_X, Y: minibatch_Y})<br \/><br \/>                minibatch_cost += temp_cost \/ num_minibatches<br \/><br \/>            # \u4fdd\u5b58\u53c2\u6570<br \/>            saver.save(session, '.\/model\/my_model', global_step=5)<br \/><br \/>            # \u6253\u5370\u6210\u672c<br \/>            if print_cost:<br \/>                if epoch % 5 == 0:<br \/>                    print(\"\u5f53\u524d\u662f\u7b2c \" + str(epoch) + \" \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a\" + str(minibatch_cost))<br \/><br \/>            # \u8bb0\u5f55\u6210\u672c<br \/>            costs.append(minibatch_cost)<br \/><br \/>        # \u7ed8\u5236\u6210\u672c\u66f2\u7ebf<br \/>        if isPlot:<br \/>            plt.plot(np.squeeze(costs))<br \/>            plt.ylabel('cost')<br \/>            plt.xlabel('iterations (per tens)')<br \/>            plt.title(\"Learning rate =\" + str(learning_rate))<br \/>            plt.show()<br \/><br \/>        # \u5f00\u59cb\u9884\u6d4b<br \/>        predict_op = tf.argmax(Z3, 1)<br \/>        corrent_prediction = tf.equal(predict_op, tf.argmax(Y, 1))<br \/>        accuracy = tf.reduce_mean(tf.cast(corrent_prediction, 'float',name='accuracy'))<br \/>        print(\"corrent_prediction accuracy= \" + str(accuracy))<br \/>        train_accuracy = accuracy.eval({X: X_train, Y: Y_train})<br \/>        test_accuracy = accuracy.eval({X: X_test, Y: Y_test})<br \/>        print(\"\u8bad\u7ec3\u96c6\u51c6\u786e\u5ea6\uff1a\" + str(train_accuracy))<br \/>        print(\"\u6d4b\u8bd5\u96c6\u51c6\u786e\u5ea6\uff1a\" + str(test_accuracy))<br \/><br \/>        return train_accuracy, test_accuracy, parameters<br \/><br \/><br \/>_, _, parameters = model(X_train, Y_train, X_test, Y_test, num_epochs=150)<\/pre>\n\n\n<p>\u7a0b\u5e8f\u8f93\u51fa\u5982\u4e0b\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\">\u5f53\u524d\u662f\u7b2c 0 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.9079020842909813<br \/>\u5f53\u524d\u662f\u7b2c 5 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.8338764905929565<br \/>\u5f53\u524d\u662f\u7b2c 10 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.3530711755156517<br \/>\u5f53\u524d\u662f\u7b2c 15 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.0513618439435959<br \/>\u5f53\u524d\u662f\u7b2c 20 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.7954236306250095<br \/>\u5f53\u524d\u662f\u7b2c 25 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.6776404492557049<br \/>\u5f53\u524d\u662f\u7b2c 30 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.6062654200941324<br \/>\u5f53\u524d\u662f\u7b2c 35 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.5560709461569786<br \/>\u5f53\u524d\u662f\u7b2c 40 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.50390131957829<br \/>\u5f53\u524d\u662f\u7b2c 45 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.45013799145817757<br \/>\u5f53\u524d\u662f\u7b2c 50 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.43954286724328995<br \/>\u5f53\u524d\u662f\u7b2c 55 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3888242533430457<br \/>\u5f53\u524d\u662f\u7b2c 60 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.37428596522659063<br \/>\u5f53\u524d\u662f\u7b2c 65 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.35892391856759787<br \/>\u5f53\u524d\u662f\u7b2c 70 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.33011145144701004<br \/>\u5f53\u524d\u662f\u7b2c 75 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3231000145897269<br \/>\u5f53\u524d\u662f\u7b2c 80 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.31615445110946894<br \/>\u5f53\u524d\u662f\u7b2c 85 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3158330311998725<br \/>\u5f53\u524d\u662f\u7b2c 90 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.26991996727883816<br \/>\u5f53\u524d\u662f\u7b2c 95 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2658209102228284<br \/>\u5f53\u524d\u662f\u7b2c 100 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2894991096109152<br \/>\u5f53\u524d\u662f\u7b2c 105 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.24726102221757174<br \/>\u5f53\u524d\u662f\u7b2c 110 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2515424760058522<br \/>\u5f53\u524d\u662f\u7b2c 115 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3262146320194006<br \/>\u5f53\u524d\u662f\u7b2c 120 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.23388438019901514<br \/>\u5f53\u524d\u662f\u7b2c 125 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.23200181871652603<br \/>\u5f53\u524d\u662f\u7b2c 130 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2730541517958045<br \/>\u5f53\u524d\u662f\u7b2c 135 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.22053561871871352<br \/>\u5f53\u524d\u662f\u7b2c 140 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.25617739744484425<br \/>\u5f53\u524d\u662f\u7b2c 145 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.19133262312971056<br \/>corrent_prediction accuracy= Tensor(\"Mean_1:0\", shape=(), dtype=float32)<br \/>\u8bad\u7ec3\u96c6\u51c6\u786e\u5ea6\uff1a0.93796295<br \/>\u6d4b\u8bd5\u96c6\u51c6\u786e\u5ea6\uff1a0.81666666<\/pre>\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img loading=\"lazy\" decoding=\"async\" width=\"342\" height=\"258\" src=\"\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-87.png\" alt=\"\" class=\"wp-image-3233\" srcset=\"http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-87.png 342w, http:\/\/www.sniper97.cn\/wp-content\/uploads\/2020\/03\/\u56fe\u7247-87-300x226.png 300w\" sizes=\"(max-width: 342px) 100vw, 342px\" \/><\/figure><\/div>\n\n\n<p>\u5b8c\u6574\u4ee3\u7801\uff1a<\/p>\n\n\n<pre class=\"wp-block-preformatted\"># -*- coding:utf-8 -*-<br \/><br \/><em>\"\"\"<br \/><\/em><em>      \u250f\u251b \u253b\u2501\u2501\u2501\u2501\u2501\u251b \u253b\u2513<br \/><\/em><em>      \u2503<\/em><em>\u3000\u3000\u3000\u3000\u3000\u3000<\/em><em> \u2503<br \/><\/em><em>      \u2503<\/em><em>\u3000\u3000\u3000<\/em><em>\u2501<\/em><em>\u3000\u3000\u3000<\/em><em>\u2503<br \/><\/em><em>      \u2503<\/em><em>\u3000<\/em><em>\u2533\u251b<\/em><em>\u3000<\/em><em>  \u2517\u2533<\/em><em>\u3000<\/em><em>\u2503<br \/><\/em><em>      \u2503<\/em><em>\u3000\u3000\u3000\u3000\u3000\u3000<\/em><em> \u2503<br \/><\/em><em>      \u2503<\/em><em>\u3000\u3000\u3000<\/em><em>\u253b<\/em><em>\u3000\u3000\u3000<\/em><em>\u2503<br \/><\/em><em>      \u2503<\/em><em>\u3000\u3000\u3000\u3000\u3000\u3000<\/em><em> \u2503<br \/><\/em><em>      \u2517\u2501\u2513<\/em><em>\u3000\u3000\u3000<\/em><em>\u250f\u2501\u2501\u2501\u251b<br \/><\/em><em>        \u2503<\/em><em>\u3000\u3000\u3000<\/em><em>\u2503   <\/em><em>\u795e\u517d\u4fdd\u4f51<\/em><em><br \/><\/em><em>        \u2503<\/em><em>\u3000\u3000\u3000<\/em><em>\u2503   <\/em><em>\u4ee3\u7801\u65e0<\/em><em>BUG<\/em><em>\uff01<\/em><em><br \/><\/em><em>        \u2503<\/em><em>\u3000\u3000\u3000<\/em><em>\u2517\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2513<br \/><\/em><em>        \u2503<\/em><em>\u3000\u3000\u3000\u3000\u3000\u3000\u3000<\/em><em>    \u2523\u2513<br \/><\/em><em>        \u2503<\/em><em>\u3000\u3000\u3000\u3000<\/em><em>         \u250f\u251b<br \/><\/em><em>        \u2517\u2501\u2513 \u2513 \u250f\u2501\u2501\u2501\u2533 \u2513 \u250f\u2501\u251b<br \/><\/em><em>          \u2503 \u252b \u252b   \u2503 \u252b \u252b<br \/><\/em><em>          \u2517\u2501\u253b\u2501\u251b   \u2517\u2501\u253b\u2501\u251b<br \/><\/em><em>\"\"\"<br \/><\/em><em><br \/><\/em>import math<br \/>import numpy as np<br \/>import h5py<br \/>import matplotlib.pyplot as plt<br \/>import matplotlib.image as mpimg<br \/>import tensorflow as tf<br \/>from tensorflow.python.framework import ops<br \/><br \/>from course_4_week_1 import cnn_utils<br \/><br \/>np.random.seed(1)<br \/>X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = cnn_utils.load_dataset()<br \/># index = 6<br \/># plt.imshow(X_train_orig[index])<br \/># print(\"y = \" + str(np.squeeze(Y_train_orig[:, index])))<br \/># plt.show()<br \/><br \/># \u6570\u636e\u6807\u51c6\u5316<br \/>X_train = X_train_orig \/ 255<br \/>X_test = X_test_orig \/ 255<br \/><br \/>Y_train = cnn_utils.convert_to_one_hot(Y_train_orig, 6).T<br \/>Y_test = cnn_utils.convert_to_one_hot(Y_test_orig, 6).T<br \/><br \/>conv_layers = {}<br \/><br \/><br \/>def create_placeholders(n_H0, n_W0, n_C0, n_y):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521b\u5efa\u5360\u4f4d\u7b26<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_H0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_W0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_C0:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> n_y:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>X = tf.placeholder(tf.float32, [None, n_H0, n_W0, n_C0])<br \/>    Y = tf.placeholder(tf.float32, [None, n_y])<br \/><br \/>    return X, Y<br \/><br \/><br \/>def initialize_parameters():<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521d\u59cb\u5316\u6743\u503c\u77e9\u9635\uff0c\u8fd9\u91cc\u6211\u4eec\u628a\u6743\u503c\u77e9\u9635\u786c\u7f16\u7801\uff1a<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em><br \/><\/em><em>    <\/em>W1 = tf.get_variable(\"W1\", [4, 4, 3, 8], initializer=tf.contrib.layers.xavier_initializer(seed=1))<br \/>    W2 = tf.get_variable(\"W2\", [2, 2, 8, 16], initializer=tf.contrib.layers.xavier_initializer(seed=1))<br \/><br \/>    parameters = {\"W1\": W1,<br \/>                  \"W2\": W2}<br \/><br \/>    return parameters<br \/><br \/><br \/>def forward_propagation(X, parameters):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u5b9e\u73b0\u524d\u5411\u4f20\u64ad<\/em><em><br \/><\/em><em>    CONV2D -&gt; RELU -&gt; MAXPOOL -&gt; CONV2D -&gt; RELU -&gt; MAXPOOL -&gt; FLATTEN -&gt; FULLYCONNECTED<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> parameters:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>W1 = parameters['W1']<br \/>    W2 = parameters['W2']<br \/><br \/>    # Conv2d : \u6b65\u4f10\uff1a1\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    Z1 = tf.nn.conv2d(X, W1, strides=[1, 1, 1, 1], padding=\"SAME\")<br \/>    # ReLU \uff1a<br \/>    A1 = tf.nn.relu(Z1)<br \/>    # Max pool : \u7a97\u53e3\u5927\u5c0f\uff1a8x8\uff0c\u6b65\u4f10\uff1a8x8\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    P1 = tf.nn.max_pool(A1, ksize=[1, 8, 8, 1], strides=[1, 8, 8, 1], padding=\"SAME\")<br \/><br \/>    # Conv2d : \u6b65\u4f10\uff1a1\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    Z2 = tf.nn.conv2d(P1, W2, strides=[1, 1, 1, 1], padding=\"SAME\")<br \/>    # ReLU \uff1a<br \/>    A2 = tf.nn.relu(Z2)<br \/>    # Max pool : \u8fc7\u6ee4\u5668\u5927\u5c0f\uff1a4x4\uff0c\u6b65\u4f10\uff1a4x4\uff0c\u586b\u5145\u65b9\u5f0f\uff1a\u201cSAME\u201d<br \/>    P2 = tf.nn.max_pool(A2, ksize=[1, 4, 4, 1], strides=[1, 4, 4, 1], padding=\"SAME\")<br \/><br \/>    # \u4e00\u7ef4\u5316\u4e0a\u4e00\u5c42\u7684\u8f93\u51fa<br \/>    P = tf.contrib.layers.flatten(P2)<br \/><br \/>    # \u5168\u8fde\u63a5\u5c42\uff08FC\uff09\uff1a\u4f7f\u7528\u6ca1\u6709\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u7684\u5168\u8fde\u63a5\u5c42<br \/>    Z3 = tf.contrib.layers.fully_connected(P, 6, activation_fn=None)<br \/><br \/>    return Z3<br \/><br \/><br \/>def compute_cost(Z3, Y):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u8ba1\u7b97\u4ee3\u4ef7<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Z3:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=Z3, labels=Y))<br \/><br \/>    return cost<br \/><br \/><br \/>def model(X_train, Y_train, X_test, Y_test, learning_rate=0.009,<br \/>          num_epochs=100, minibatch_size=64, print_cost=True, isPlot=True):<br \/>    <em>\"\"\"<br \/><\/em><em>    <\/em><em>\u521b\u5efa\u6a21\u578b<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X_train:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y_train:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> X_test:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> Y_test:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> learning_rate:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> num_epochs:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> minibatch_size:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> print_cost:<br \/><\/em><em>    <\/em><strong><em>:param<\/em><\/strong><em> isPlot:<br \/><\/em><em>    <\/em><strong><em>:return<\/em><\/strong><em>:<br \/><\/em><em>    \"\"\"<br \/><\/em><em>    <\/em>tf.get_default_graph()<br \/>    seed = 3<br \/>    tf.set_random_seed(1)<br \/>    # m\u4e2a\u56fe\u7247  h*w\u5927\u5c0f  c\u4e2a\u5377\u79ef\u5c42<br \/>    m, n_H0, n_W0, n_C0 = X_train.shape<br \/>    # \u6709\u591a\u5c11\u79cd\u7ed3\u679c<br \/>    n_y = Y_train.shape[1]<br \/>    costs = []<br \/>    X, Y = create_placeholders(n_H0, n_W0, n_C0, n_y)<br \/>    # \u521d\u59cb\u5316\u53c2\u6570<br \/>    parameters = initialize_parameters()<br \/>    # \u524d\u5411\u4f20\u64ad<br \/>    Z3 = forward_propagation(X, parameters)<br \/>    # \u8ba1\u7b97\u6210\u672c<br \/>    cost = compute_cost(Z3, Y)<br \/>    # \u53cd\u5411\u4f20\u64ad<br \/>    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)<br \/>    # \u4fdd\u5b58\u6570\u636e<br \/>    saver = tf.train.Saver()<br \/>    # \u5168\u5c40\u521d\u59cb\u5316\u53d8\u91cf<br \/>    init = tf.global_variables_initializer()<br \/>    # \u5f00\u59cb\u8fd0\u884c<br \/>    with tf.Session() as session:<br \/>        # \u521d\u59cb\u5316\u53c2\u6570<br \/>        session.run(init)<br \/>        # \u904d\u5386\u6570\u636e<br \/>        for epoch in range(num_epochs):<br \/>            minibatch_cost = 0<br \/>            num_minibatches = int(m \/ minibatch_size)<br \/>            seed = seed + 1<br \/>            minibatches = cnn_utils.random_mini_batches(X_train, Y_train, minibatch_size, seed)<br \/><br \/>            for minibatch in minibatches:<br \/>                minibatch_X, minibatch_Y = minibatch<br \/>                _, temp_cost = session.run([optimizer, cost], feed_dict={X: minibatch_X, Y: minibatch_Y})<br \/><br \/>                minibatch_cost += temp_cost \/ num_minibatches<br \/><br \/>            # \u4fdd\u5b58\u53c2\u6570<br \/>            saver.save(session, '.\/model\/my_model', global_step=5)<br \/><br \/>            # \u6253\u5370\u6210\u672c<br \/>            if print_cost:<br \/>                if epoch % 5 == 0:<br \/>                    print(\"\u5f53\u524d\u662f\u7b2c \" + str(epoch) + \" \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a\" + str(minibatch_cost))<br \/><br \/>            # \u8bb0\u5f55\u6210\u672c<br \/>            costs.append(minibatch_cost)<br \/><br \/>        # \u7ed8\u5236\u6210\u672c\u66f2\u7ebf<br \/>        if isPlot:<br \/>            plt.plot(np.squeeze(costs))<br \/>            plt.ylabel('cost')<br \/>            plt.xlabel('iterations (per tens)')<br \/>            plt.title(\"Learning rate =\" + str(learning_rate))<br \/>            plt.show()<br \/><br \/>        # \u5f00\u59cb\u9884\u6d4b<br \/>        predict_op = tf.argmax(Z3, 1)<br \/>        corrent_prediction = tf.equal(predict_op\n, tf.argmax(Y, 1))<br \/>        accuracy = tf.reduce_mean(tf.cast(corrent_prediction, 'float',name='accuracy'))<br \/>        print(\"corrent_prediction accuracy= \" + str(accuracy))<br \/>        train_accuracy = accuracy.eval({X: X_train, Y: Y_train})<br \/>        test_accuracy = accuracy.eval({X: X_test, Y: Y_test})<br \/>        print(\"\u8bad\u7ec3\u96c6\u51c6\u786e\u5ea6\uff1a\" + str(train_accuracy))<br \/>        print(\"\u6d4b\u8bd5\u96c6\u51c6\u786e\u5ea6\uff1a\" + str(test_accuracy))<br \/><br \/>        return train_accuracy, test_accuracy, parameters<br \/><br \/><br \/>_, _, parameters = model(X_train, Y_train, X_test, Y_test, num_epochs=150)<br \/>\"\"\"<br \/>\u5f53\u524d\u662f\u7b2c 0 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.9079020842909813<br \/>\u5f53\u524d\u662f\u7b2c 5 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.8338764905929565<br \/>\u5f53\u524d\u662f\u7b2c 10 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.3530711755156517<br \/>\u5f53\u524d\u662f\u7b2c 15 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a1.0513618439435959<br \/>\u5f53\u524d\u662f\u7b2c 20 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.7954236306250095<br \/>\u5f53\u524d\u662f\u7b2c 25 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.6776404492557049<br \/>\u5f53\u524d\u662f\u7b2c 30 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.6062654200941324<br \/>\u5f53\u524d\u662f\u7b2c 35 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.5560709461569786<br \/>\u5f53\u524d\u662f\u7b2c 40 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.50390131957829<br \/>\u5f53\u524d\u662f\u7b2c 45 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.45013799145817757<br \/>\u5f53\u524d\u662f\u7b2c 50 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.43954286724328995<br \/>\u5f53\u524d\u662f\u7b2c 55 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3888242533430457<br \/>\u5f53\u524d\u662f\u7b2c 60 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.37428596522659063<br \/>\u5f53\u524d\u662f\u7b2c 65 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.35892391856759787<br \/>\u5f53\u524d\u662f\u7b2c 70 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.33011145144701004<br \/>\u5f53\u524d\u662f\u7b2c 75 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3231000145897269<br \/>\u5f53\u524d\u662f\u7b2c 80 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.31615445110946894<br \/>\u5f53\u524d\u662f\u7b2c 85 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3158330311998725<br \/>\u5f53\u524d\u662f\u7b2c 90 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.26991996727883816<br \/>\u5f53\u524d\u662f\u7b2c 95 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2658209102228284<br \/>\u5f53\u524d\u662f\u7b2c 100 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2894991096109152<br \/>\u5f53\u524d\u662f\u7b2c 105 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.24726102221757174<br \/>\u5f53\u524d\u662f\u7b2c 110 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2515424760058522<br \/>\u5f53\u524d\u662f\u7b2c 115 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.3262146320194006<br \/>\u5f53\u524d\u662f\u7b2c 120 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.23388438019901514<br \/>\u5f53\u524d\u662f\u7b2c 125 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.23200181871652603<br \/>\u5f53\u524d\u662f\u7b2c 130 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.2730541517958045<br \/>\u5f53\u524d\u662f\u7b2c 135 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.22053561871871352<br \/>\u5f53\u524d\u662f\u7b2c 140 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.25617739744484425<br \/>\u5f53\u524d\u662f\u7b2c 145 \u4ee3\uff0c\u6210\u672c\u503c\u4e3a\uff1a0.19133262312971056<br \/>corrent_prediction accuracy= Tensor(\"Mean_1:0\", shape=(), dtype=float32)<br \/>\u8bad\u7ec3\u96c6\u51c6\u786e\u5ea6\uff1a0.93796295<br \/>\u6d4b\u8bd5\u96c6\u51c6\u786e\u5ea6\uff1a0.81666666<br \/>\"\"\"<\/pre>\n","protected":false},"excerpt":{"rendered":"<p>\u5434\u6069\u8fbe\u6df1\u5ea6\u5b66\u4e60\u7b2c\u56db\u8bfe \u7b2c\u4e00\u5468 \u5377\u79ef\u795e\u7ecf\u7f51\u7edc 1.\u8ba1\u7b97\u673a\u89c6\u89c9 \u6df1\u5ea6\u5b66\u4e60\u4e0e\u8ba1\u7b97\u673a\u89c6\u89c9\u53ef\u4ee5\u5e2e\u52a9\u6c7d\u8f66\uff0c\u67e5\u660e\u5468 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_mi_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[7],"tags":[],"views":13650,"_links":{"self":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts\/3161"}],"collection":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/comments?post=3161"}],"version-history":[{"count":0,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/posts\/3161\/revisions"}],"wp:attachment":[{"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/media?parent=3161"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/categories?post=3161"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.sniper97.cn\/index.php\/wp-json\/wp\/v2\/tags?post=3161"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}