{"id":17964,"date":"2022-12-08T10:00:00","date_gmt":"2022-12-08T02:00:00","guid":{"rendered":"https:\/\/www.circuspi.com\/?p=17964"},"modified":"2022-12-09T09:28:34","modified_gmt":"2022-12-09T01:28:34","slug":"jetson-nano-dli-image-classification","status":"publish","type":"post","link":"https:\/\/www.circuspi.com\/index.php\/2022\/12\/08\/jetson-nano-dli-image-classification\/","title":{"rendered":"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e"},"content":{"rendered":"\n<p><style>\n\/* ###\u5be6\u9a57\u4e0b\u4e00\u5217### *\/\n\na:link {color:#0695e4\n;}    \/* \u8a2d\u5b9a\u5c1a\u672a\u9ede\u95b1\u904e\u7684\u9023\u7d50\u6a23\u5f0f *\/\na:visited {color:#0695e4\n;} \/* \u8a2d\u5b9a\u904e\u53bb\u66fe\u7d93\u95b1\u904e\u7684\u9023\u7d50 *\/\na:hover {color:#0695e4\n;}   \/* \u8a2d\u5b9a\u6ed1\u9f20\u6e38\u6a19\u6307\u5728\u9023\u7d50\u4f4d\u7f6e\u4e0a\u7684\u6a23\u5f0f *\/\na:active {color:#0000BB;}  \/* \u8a2d\u5b9a\u4e0a\u9ede\u64ca\u904e\u7684\u9023\u7d50\u6a23\u5f0f *\/\n<\/style><\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1000\" height=\"527\" src=\"https:\/\/www.circuspi.com\/wp-content\/uploads\/2022\/12\/JetsonNano_DLICH2_L.jpg\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\" class=\"wp-image-18157\"\/><\/figure><\/div>\n\n\n\n<p>\u95b1\u8b80\u5b8c\u300cJetson Nano DLI \u6559\u5b78(\u4e00)\uff1a\u74b0\u5883\u6e96\u5099\u8207 hello Camera\u300d\u5f8c\uff0c\u76f8\u4fe1\u5927\u5bb6\u61c9\u8a72\u90fd\u5df2\u7d93\u5c0d <a href=\"http:\/\/nvidia.com\/zh-tw\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA<\/a> DLI (Deep Learning Institute)\u6709\u4e00\u5b9a\u7684\u8a8d\u8b58\u4e86\uff0c\u672c\u7bc7\u6587\u7ae0\u5c07\u5728 JupyterLab \u4f7f\u7528\u4e92\u52d5\u5f0f\u4ecb\u9762\u5be6\u4f5c\u5f71\u50cf\u5206\u985e\u6a21\u578b\u7684<strong>\u8cc7\u6599\u6536\u96c6(data Collect)<\/strong>\u3001<strong>\u8a13\u7df4(Train)\u8207\u9810\u6e2c(Predict)<\/strong>\u3002<\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/massive-danthus-1c2.notion.site\/image\/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F30bd1035-efce-4787-ba6b-a6e4dfa9631c%2FUntitled.png?table=block&amp;id=2d622915-2c4c-4d6d-86ea-577e303b5032&amp;spaceId=b4fd4f91-6283-40d5-bc09-2016e5fc107b&amp;width=2000&amp;userId=&amp;cache=v2\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\" width=\"673\" height=\"384\"\/><figcaption>Resnet-18 \u67b6\u69cb\u793a\u610f (source: NVIDIA)<\/figcaption><\/figure><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e00\u3001\u958b\u555f\u5f71\u50cf\u5206\u985e ipynb \u7b46\u8a18\u672c<\/h2>\n\n\n\n<p>\u548c Jetson Nano DLI \u6559\u5b78(\u4e00) \u76f8\u540c\u7684\u65b9\u5f0f\u555f\u52d5 DLI docker image \u4e26\u4e14\u5b8c\u6210\u767b\u9304\uff0c\u5f9e JupyerLab \u7db2\u9801\u4ecb\u9762\u5de6\u5074\u6a94\u6848\u700f\u89bd\u5668\u9032\u53bb <strong>Classification<\/strong> \u76ee\u9304\uff0c\u958b\u555f <strong>classification_interactive.ipynb<\/strong> \u6a94\u6848\u3002<\/p>\n\n\n\n<p>\u7b2c\u4e00\u500b\u6bb5\u843d\u76f8\u6a5f Camera \u6703\u958b\u555f\u4f60\u7684 <a href=\"https:\/\/www.icshop.com.tw\/product-page.php?27390\" target=\"_blank\" rel=\"noreferrer noopener\">USB Webcam<\/a> \u6216 <a href=\"https:\/\/www.icshop.com.tw\/product-page.php?24894\" target=\"_blank\" rel=\"noreferrer noopener\">CSI \u76f8\u6a5f<\/a>\uff0c\u9810\u8a2d\u662f\u4f7f\u7528 USB \u4ecb\u9762\uff0c\u5982\u679c\u958b\u767c\u8005\u7684\u74b0\u5883\u662f\u4f7f\u7528 CSI \u4ecb\u9762\u7684\u9808\u4fee\u6539\u7a0b\u5f0f\u78bc\u628a\u539f\u59cb USB \u76f8\u6a5f\u555f\u52d5\u7a0b\u5f0f\u78bc\u8a3b\u89e3\uff0c\u4e26\u53d6\u6d88\u8a3b\u89e3 CSI \u76f8\u6a5f\u7684\u555f\u52d5\uff0c\u5982\u4e0b\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># for USB Camera (Logitech C270 webcam), uncomment the following line\n# camera = USBCamera(width=224, height=224, capture_device=0) # confirm the capture_device number\n\n# for CSI Camera (Raspberry Pi Camera Module V2), uncomment the following line\ncamera = CSICamera(width=224, height=224, capture_device=0) # confirm the capture_device number<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e8c\u3001\u4ee5\u7d42\u70ba\u59cb\uff0c\u5148\u770b\u7d50\u679c\uff01<\/h2>\n\n\n\n<p>\u4ee5\u7b46\u8005\u7684\u7d93\u9a57\u5efa\u8b70\u6240\u6709\u5165\u9580\u8005\u5148\u770b\u5230\u7d50\u679c\uff0c\u6709\u904e\u9ad4\u9a57\u5927\u6982\u77e5\u9053\u4ec0\u9ebc\u662f\u6df1\u5ea6\u5b78\u7fd2\u4e4b\u5f8c\u518d\u56de\u982d\u53bb\u770b\u7a0b\u5f0f\u78bc\u7d30\u7bc0\uff0c\u5b78\u7fd2\u6548\u679c\u6703\u6bd4\u8f03\u597d\u3002\u56e0\u6b64\u9019\u908a\u6703\u8b93\u5404\u4f4d\u5148\u73a9\u4e92\u52d5\u5f0f\u6a21\u578b\u8a13\u7df4\u8207\u8a55\u4f30\uff0c\u7a0b\u5f0f\u78bc\u7d30\u7bc0\u7a0d\u5f8c\u518d\u505a\u8aaa\u660e\u3002\u5c31\u5148\u8b93\u6211\u5011\u4f9d\u5e8f\u57f7\u884c\u6240\u6709\u7a0b\u5f0f\u78bc cell \u6bb5\u843d\uff0c\u76f4\u5230\u300c<strong>\u986f\u793a\u4e92\u52d5\u5f0f\u5de5\u5177\uff01\u300d<\/strong> \u9019\u6bb5\u843d\u5b8c\u6210\u5c31\u6703\u770b\u5230\u4e00\u500b\u5982\u7b46\u8a18\u672c\u4e0a\u8aaa\u660e\u7684\u4e92\u52d5\u5f0f\u4ecb\u9762\uff1a<\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/massive-danthus-1c2.notion.site\/image\/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F8173945d-3039-4cb1-8fbe-229acf3ddf5a%2F130a.png?table=block&amp;id=be13030d-9111-4750-bb30-fe9a01d54007&amp;spaceId=b4fd4f91-6283-40d5-bc09-2016e5fc107b&amp;width=2000&amp;userId=&amp;cache=v2\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\" width=\"834\" height=\"449\"\/><\/figure><\/div>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u6536\u96c6\u8cc7\u6599<\/h3>\n\n\n\n<p>\u9019\u88e1\u770b\u5230<strong>\u8cc7\u6599\u96c6 dataset<\/strong> \u53ef\u85c9\u7531\u4e0b\u62c9\u5f0f\u9078\u55ae\u9078\u64c7 A \u6216 B\uff0c\u9019\u6703\u6709\u5169\u7d44\u4e0d\u540c\u7684\u8cc7\u6599\u96c6\uff0c\u5f7c\u6b64\u5206\u958b\u5132\u5b58\u4e0d\u5171\u7528\uff0c\u4e26\u4e14\u6703\u5be6\u969b\u5beb\u5165\u5230\u6a94\u6848\u7cfb\u7d71\u4e2d\u3002<strong>\u985e\u5225 category<\/strong> \u5247\u662f\u5f71\u50cf\u5206\u985e\u7684\u7d50\u679c\u7e3d\u5171\u6709\u54ea\u5e7e\u985e\uff0c\u5728\u9810\u8a2d\u7684\u7bc4\u4f8b\u4e2d\u662f\u4e00\u500b\u4e8c\u5143\u5206\u985e\uff0c\u5206\u5225\u70ba <strong>thumbs_up<\/strong> \u8207 <strong>thumbs_down<\/strong>\uff0c\u4e5f\u5c31\u662f\u8b93\u6211\u5011\u505a <strong>\u8b9a<\/strong> \u8207 <strong>\u5012\u8b9a<\/strong> \u9019\u5169\u8005\u7684\u5f71\u50cf\u5206\u985e\u6a21\u578b\u3002\u6700\u5f8c<strong>\u8a08\u6578 count<\/strong> \u5247\u6703\u8a18\u9304\u76ee\u524d\u6b64\u985e\u578b\u7684\u8a13\u7df4\u8cc7\u6599\u6709\u5e7e\u5f35\u3002<\/p>\n\n\n\n<p>\u9019\u88e1\u6211\u5011\u5c31\u5206\u5225\u5c0d\u8457\u93e1\u982d\u6bd4\u8b9a\uff0c\u6bcf\u9ede\u9078\u4e00\u6b21\u300cadd\u300d\u6309\u9215\uff0c\u5c31\u6703\u5c07\u7576\u524d\u5b58\u6a94\u6210 jpg \u6a94\u3002\u8a66\u8457\u7d66\u4e88\u4e0d\u540c\u7684\u89d2\u5ea6\u8207\u5927\u5c0f\uff0c\u4e26\u6536\u96c6\u5f71\u50cf\u6578\u91cf\u7d04\u9054 40~50 \u5f35\u7684\u7167\u7247\u5373\u53ef\u3002\u63a5\u4e0b\u4f86\u5728\u628a\u985e\u578b\u5207\u63db\u5230 t<strong>humbs_down<\/strong>\uff0c\u4e00\u6a23\u62cd\u651d 40~50 \u5f35\u7167\u7247\uff0c\u5c31\u53ef\u5b8c\u6210\u6536\u96c6\u8cc7\u6599\u7684\u90e8\u5206\u3002<\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/massive-danthus-1c2.notion.site\/image\/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F83135fda-87b3-4c58-ad16-3225fa0d9c45%2FUntitled.png?table=block&amp;id=3e6bad1c-7632-47f1-bdf3-db25bb4a3a93&amp;spaceId=b4fd4f91-6283-40d5-bc09-2016e5fc107b&amp;width=1970&amp;userId=&amp;cache=v2\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\" width=\"812\" height=\"682\"\/><\/figure><\/div>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u8a13\u7df4\u6a21\u578b<\/h3>\n\n\n\n<p>\u8a13\u7df4\u6a21\u578b\u5176\u5be6\u53ea\u6709\u4e00\u500b\u53c3\u6578\u8981\u8abf\u6574\uff0c\u5c31\u662f<strong>\u8a13\u7df4\u6b21\u6578 epochs<\/strong>\uff0c\u5efa\u8b70\u53ef\u4ee5\u8a2d\u5b9a 5~10 \u6b21\u5c31\u53ef\u4ee5\u6709\u4e0d\u932f\u7684\u6548\u679c\uff0c\u4e5f\u4e0d\u6703\u82b1\u8cbb\u592a\u591a\u7684\u6642\u9593\u3002\u9ede\u9078\u300c<strong>train<\/strong>\u300d\u6309\u9215\u5c31\u6703\u76f4\u63a5\u5728 Jetson Nano \u4e0a\u8dd1\u6a21\u578b\u8a13\u7df4\uff0c\u4f9d\u64da\u8cc7\u6599\u91cf\u4ee5\u53ca<strong>\u8a13\u7df4\u6b21\u6578 epochs<\/strong> \u7684\u4e0d\u540c\uff0c\u82b1\u8cbb\u7684\u6642\u9593\u4e5f\u6703\u6709\u6240\u5dee\u7570\u3002\u8a13\u7df4\u904e\u7a0b\u540c\u6642\u4e5f\u6703\u770b\u5230<strong>\u9032\u5ea6\u689d progress bat<\/strong>\u3001<strong>\u640d\u5931 loss<\/strong>\u3001<strong>\u7cbe\u78ba\u5ea6 accuracy<\/strong> \u7684\u52d5\u614b\u8b8a\u5316\u3002\u7406\u8ad6\u4e0a <strong>loss<\/strong> \u8d8a\u5c0f\u8d8a\u597d\uff0c <strong>accuracy<\/strong> \u8d8a\u8da8\u8fd1\u65bc\u4e00\u8d8a\u597d\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u6e2c\u8a66<\/h3>\n\n\n\n<p>\u9ede\u9078\u300clive\u300d\u6309\u9215\u5f8c\uff0c\u6703\u5373\u6642\u6293\u53d6\u5982\u540c\u9810\u89bd\u7684 camera \u5f71\u50cf\u8cc7\u6599\uff0c\u4e26\u4e14\u9032\u884c\u63a8\u8ad6\u3002\u63a8\u8ad6\u5224\u5b9a\u7684\u7d50\u679c\uff08\u6a21\u578b\u5224\u5b9a\u7684\u6a5f\u7387\uff09\u4ee5\u5169\u689d\u6ed1\u52d5\u687f Slider Bar \u5448\u73fe\uff0c\u6240\u6709\u6a5f\u7387\u52a0\u7e3d\u70ba 1 \u3002\u7b46\u8005\u6574\u9ad4\u6e2c\u8a66\u8d77\u4f86\u8fa8\u8b58\u5ea6\u9084\u76f8\u7576\u4e0d\u932f\u3002<\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter\"><img decoding=\"async\" src=\"https:\/\/lh3.googleusercontent.com\/ow0XQRBBwYnVfDt5SPlWYEz6S2mZtfqEjHAp_pFH3D-TuecUSytHhHuQDpWDsbP7gu2uMQ04VxeYjPtN9zIBgmz-7FEDuLOmMlIWh_6riqYulnIRQ0ajvwYU1gZEUhU5-3ijGecyz3o=w1200\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\"\/><\/figure><\/div>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u5132\u5b58\u6a21\u578b<\/h3>\n\n\n\n<p>\u82e5\u89ba\u5f97\u7576\u524d\u7684\u6a21\u578b\u6548\u679c\u4e0d\u932f\uff0c\u4e5f\u53ef\u4ee5\u5f9e\u9019\u88e1\u9ede\u9078\u300c<strong>save model<\/strong>\u300d\u6309\u9215\uff0c\u5132\u5b58\u5f8c\u62ff\u53bb\u5176\u4ed6\u88dd\u7f6e\u8f09\u5165\u63a8\u8ad6\u4f7f\u7528\uff08\u8a8d\u771f( \u0361\u00b0 \u035c\u0296 \u0361\u00b0)\uff09\u3002\u5132\u5b58\u7684\u683c\u5f0f\u70ba Pytorch \u4f7f\u7528\u7684 pth \u6a21\u578b\u6a94\u3002\u4e0d\u904e\u8a18\u5f97\u8981\u532f\u5165\u6a21\u578b\u7684\u8a71\uff0c\u8a18\u5f97\u628a\u985e\u5225\u7684\u540d\u7a31\u6578\u91cf\u6539\u70ba\u548c\u539f\u6a21\u578b\u4e00\u81f4\uff0c\u624d\u4e0d\u6703\u767c\u751f\u554f\u984c\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e09\u3001\u9077\u79fb\u5b78\u7fd2 Transfer Learning<\/h2>\n\n\n\n<p>\u5404\u4f4d\u53ef\u80fd\u6703\u89ba\u5f97\u597d\u5947\uff0c\u70ba\u4ec0\u9ebc\u8a13\u7df4\u4e00\u500b\u6df1\u5ea6\u5b78\u7fd2\u6a21\u578b\u53ef\u4ee5\u9019\u9ebc\u5feb\uff01\uff1f\u5176\u5be6\u9019\u7bc4\u4f8b\u7528\u4e86\u53eb\u505a<strong>\u9077\u79fb\u5b78\u7fd2 Transfer Learning<\/strong> \u7684\u6280\u8853\uff0c\u6982\u5ff5\u5c31\u8ddf\u4eba\u5728\u5b78\u7fd2\u65b0\u4e8b\u7269\u6642\u6703\u628a\u820a\u6709\u7684\u7d93\u9a57\u5e36\u5165\uff0c\u9054\u5230\u7528\u8f03\u77ed\u6642\u9593\u5b78\u7fd2\u65b0\u6280\u8853\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<p>\u5728\u5377\u7a4d\u795e\u7d93\u7db2\u8def (Convolutional Neural Networks, CNN) \u4e2d\u524d\u534a\u6bb5\u7684\u6372\u7a4d (Convolution) \u8207\u6c60\u5316 (Pooling) \u7684\u904b\u7b97\uff0c\u53ef\u4ee5\u7406\u89e3\u70ba\u662f\u5728\u8403\u53d6\u5716\u50cf\u4e2d\u7684\u7279\u5fb5\u503c\u3002\u5f8c\u6bb5\u7684\u5168\u9023\u63a5\u5c64 (fully connected layers) \u4f86\u505a\u5404\u985e\u7279\u5fb5\u503c\u5c0d\u65bc\u4e0d\u540c\u985e\u5225\u7684\u6b0a\u91cd\u8abf\u6574\u3002\u56e0\u6b64\u524d\u534a\u6bb5\u7684\u795e\u7d93\u7db2\u8def\u6b0a\u91cd\u53ef\u4ee5\u76f4\u63a5\u62ff\u5176\u4ed6\u4eba\u8a13\u7df4\u597d\u7684\u6a21\u578b\u4f86\u4f7f\u7528\uff0c\u800c\u6b64\u7bc4\u4f8b\u4e2d\u5be6\u969b\u8981\u8a13\u7df4\u7684\u6b0a\u91cd\u5c31\u53ea\u6709\u5f8c\u6bb5\u5c0f\u90e8\u5206\u7684\u5168\u9023\u63a5\u7684\u795e\u7d93\u5143\u4e86\uff01<\/p>\n\n\n\n<p>\u4ee5\u6b64\u7bc4\u4f8b\u70ba\u4f8b\uff0c\u4f7f\u7528\u7684\u662f ResNet-18 \u795e\u7d93\u7db2\u8def\u67b6\u69cb\uff0c\u9810\u8a13\u7df4\u6b0a\u91cd\u5247\u662f LSVRC 2012 \u7af6\u8cfd\u4f7f\u7528\u7684 <a href=\"https:\/\/image-net.org\/challenges\/LSVRC\/2012\/browse-synsets\">ImageNet<\/a> \u5f71\u50cf\u5206\u985e\u8cc7\u6599\u96c6\uff0c\u6db5\u84cb\u6578\u4ee5\u5343\u8a08\u7684\u5206\u985e\u8207\u4e0a\u767e\u842c\u5f35\u7684\u5716\u50cf\u8cc7\u6599\u3002\u95dc\u65bc\u9032\u4e00\u6b65 CNN \u7684\u7d30\u7bc0\u9019\u88e1\u5c31\u4e0d\u505a\u8d05\u8ff0\u4e86\uff0c\u6709\u8208\u8da3\u5404\u4f4d\u53ef\u4ee5\u53bb google \u6df1\u5ea6\u5b78\u7fd2\u76f8\u95dc\u6559\u5b78\u8cc7\u6e90\uff0c\u90fd\u6709\u8a31\u591a\u9ad8\u624b\u7684\u6587\u737b\u53ef\u4ee5\u5b78\u7fd2\u3002<\/p>\n\n\n\n<div class=\"wp-block-image\"><figure class=\"aligncenter is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/massive-danthus-1c2.notion.site\/image\/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F96a980d3-ac4e-4565-87fc-8deba8d5b3d7%2FUntitled.png?table=block&amp;id=b691f9d8-6c63-4110-9b2d-8cf9f1e18a55&amp;spaceId=b4fd4f91-6283-40d5-bc09-2016e5fc107b&amp;width=1120&amp;userId=&amp;cache=v2\" alt=\"Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\" width=\"725\" height=\"327\"\/><figcaption>CNN \u67b6\u69cb\u793a\u610f (source: NVIDIA)<\/figcaption><\/figure><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">\u56db\u3001\u8a2d\u5b9a\u4efb\u52d9 Task<\/h2>\n\n\n\n<p>\u56de\u5230\u6b64\u5c08\u6848\u7684\u524d\u6bb5\u7a0b\u5f0f\u78bc\uff0c\u5728 <strong>\u4efb\u52d9 Task<\/strong> \u7684\u7a0b\u5f0f\u78bc cell \u53ef\u4ee5\u770b\u5230 TASK, CATEGORIES, DATASETS \u9019\u4e09\u500b\u53c3\u6578\u7684\u8a2d\u5b9a\uff0c\u5206\u5225\u5c31\u662f\u5c0d\u61c9\u5230\u7a0d\u65e9\u4f7f\u7528\u7684\u4e92\u52d5\u5f0f\u5de5\u5177\u4e0a\u9762\u7684\u9078\u9805\u3002\u9810\u8a2d\u4f7f\u7528\u7684 thumbs \u53ea\u6709\u5169\u500b\u5206\u985e\uff0c\u6211\u5011\u53ef\u4ee5\u81ea\u884c\u8abf\u6574\u70ba\u5176\u4ed6\u7bc4\u4f8b\u7684<strong>\u60c5\u7dd2\u5206\u985e<\/strong>\u6216\u662f<strong>\u624b\u6307\u6578\u91cf\u5206\u985e<\/strong>\uff0c\u751a\u81f3\u8981\u81ea\u884c\u8a2d\u5b9a\u4efb\u52d9\u4ee5\u53ca\u5404\u500b\u985e\u5225\u7684\u540d\u7a31\u90fd\u662f\u53ef\u4ee5\u5f48\u6027\u8abf\u6574\u7684\uff01DATASETS \u6307\u7684\u662f\u8aaa\u6b64\u4efb\u52d9\u6240\u6536\u96c6\u7684\u8cc7\u6599\u8981\u5206\u70ba\u5e7e\u500b\u8cc7\u6599\u96c6\uff0c\u5be6\u969b\u4e0a\u53ea\u662f\u628a\u7167\u7247\u5b58\u5728\u6a94\u6848\u7cfb\u7d71\u6642\uff0c\u7528\u4e0d\u540c\u8cc7\u6599\u593e\u53bb\u5132\u5b58\u800c\u5df2\uff0c\u65b9\u4fbf\u5c07\u8a13\u7df4\u8cc7\u6599\u6b78\u6a94\u6574\u7406\u3002<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>TASK = 'thumbs'\n# TASK = 'emotions'\n# TASK = 'fingers'\n# TASK = 'diy'\n\nCATEGORIES = &#91;'thumbs_up', 'thumbs_down']\n# CATEGORIES = &#91;'none', 'happy', 'sad', 'angry']\n# CATEGORIES = &#91;'1', '2', '3', '4', '5']\n# CATEGORIES = &#91; 'diy_1', 'diy_2', 'diy_3']\n\nDATASETS = &#91;'A', 'B']\n# DATASETS = &#91;'A', 'B', 'C']<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e94\u3001\u6a21\u578b\u8a2d\u5b9a Model<\/h2>\n\n\n\n<p>\u6a21\u578b\u8a2d\u5b9a\u7684\u7a0b\u5f0f\u78bc\u6bb5\u843d\u4e5f\u6709\u4e9b\u53ef\u4ee5\u8abf\u6574\u7684\u9078\u9805\uff0c\u7a0b\u5f0f\u78bc\u8a3b\u89e3\u90e8\u5206\u5ef6\u4f38\u51fa\u9664\u4e86\u9810\u8a2d\u7684 ResNet-18 \u4ee5\u5916\uff0c\u5176\u4ed6\u4e09\u500b\u6a21\u578b\u53ef\u4ee5\u505a\u66ff\u63db\uff0c\u5206\u5225\u70ba AlexNet, SqueezeNet, ResNet-34\uff0c\u90fd\u7b97\u662f\u8f15\u91cf\u7d1a\u53c3\u6578\u91cf\u8f03\u5c0f\u7684 CNN \u795e\u7d93\u7db2\u8def\u3002\u9019\u5e7e\u500b\u795e\u7d93\u7db2\u8def\u5728 pytorch \u90fd\u6709\u652f\u63f4\uff0c\u9078\u5b9a\u597d\u6a21\u578b\u4ee3\u5165\u53c3\u6578 <strong>pretrained=True<\/strong>\uff0c\u5c31\u53ef\u4ee5\u76f4\u63a5\u8f09\u5165\u9810\u8a13\u7df4\u597d\u7684\u6b0a\u91cd\u3002\u4e0b\u9762\u4e00\u884c\u57f7\u884c\u7684 <strong>torch.nn.Linear<\/strong> \u5247\u662f\u8981\u8b93\u539f\u59cb\u6a21\u578b\u8f38\u51fa\u7684\u5168\u9023\u63a5\u795e\u7d93\u5143 512 \u500b\u7dda\u6027\u5c0d\u61c9\u5230\u6211\u5011\u9700\u8981\u7684\u5206\u985e\u6578\u91cf\u3002\u6a21\u578b\u4e4b\u9593\u8981\u66ff\u63db\u53ea\u8981\u7528\u8a3b\u89e3\u53bb\u8abf\u6574\uff0c\u4e4b\u5f8c\u518d\u91cd\u65b0\u57f7\u884c\u6b64\u5340\u584a\u4ee5\u4e0b\u7684\u7a0b\u5f0f\u78bc\u5373\u53ef\uff01<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>device = torch.device('cuda')\n\n# ALEXNET\n# model = torchvision.models.alexnet(pretrained=True)\n# model.classifier&#91;-1] = torch.nn.Linear(4096, len(dataset.categories))\n\n# SQUEEZENET \n# model = torchvision.models.squeezenet1_1(pretrained=True)\n# model.classifier&#91;1] = torch.nn.Conv2d(512, len(dataset.categories), kernel_size=1)\n# model.num_classes = len(dataset.categories)\n\n# RESNET 18\nmodel = torchvision.models.resnet18(pretrained=True)\nmodel.fc = torch.nn.Linear(512, len(dataset.categories))\n\n# RESNET 34\n# model = torchvision.models.resnet34(pretrained=True)\n# model.fc = torch.nn.Linear(512, len(dataset.categories))\n    \nmodel = model.to(device)<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u516d\u3001\u5c0f\u7d50<\/h2>\n\n\n\n<p>\u5982\u679c\u662f\u521d\u6b21\u9ad4\u9a57\u9019\u6b21\u5f71\u50cf\u5206\u985e\u7bc4\u4f8b\u7684\u5925\u4f34\uff0c\u662f\u5426\u6703\u89ba\u5f97\u5728\u908a\u7de3\u88dd\u7f6e\u505a Edge AI \u904b\u7b97\u5c31\u5982\u540c\u73a9 <a href=\"https:\/\/teachablemachine.withgoogle.com\/\" target=\"_blank\" rel=\"noreferrer noopener\">Teachable Machine<\/a> \u4e00\u6a23\u7c21\u55ae\u5462\uff1f~~\uff08\u4e0d\u9019\u6a23\u641e\u600e\u9ebc\u9a19\u4eba\u5165\u5751\uff01(\u00ac\u25e1\u00ac)\u2727\uff09~~\u8aaa\u5be6\u8a71 Edge AI \u771f\u7684\u4e5f\u662f\u5bb9\u6613\u73a9\uff0c\u4e0d\u904e\u5f48\u6027\u76f8\u7576\u5927\uff0c\u53ef\u6dfa\u4e5f\u53ef\u6df1\uff0c\u800c\u4e14\u4e5f\u90fd\u6703\u6db5\u84cb\u5230\u4e00\u4e9b\u96fb\u5b50\u786c\u9ad4\u7684\u6574\u5408\u8207\u77e5\u8b58\u91cf\uff0c\u9019\u4e5f\u662f <a href=\"https:\/\/www.circuspi.com\/index.php\/category\/technical-articles\/ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">Edge AI<\/a> \u6709\u8da3\u7684\u5730\u65b9\u554a\uff01\u4e0b\u4e00\u7bc7\u6587\u7ae0\u5c31\u8b93\u6211\u5011\u7e7c\u7e8c\u4f86\u770b\u4e0b\u4e00\u500b\u7bc4\u4f8b\u56c9\uff01\uff3c(\u25cf\u00b4\u03d6`\u25cf)\uff0f<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Jetson Nano DLI \u6559\u5b78(\u4e8c)\uff1aImage Classification \u5f71\u50cf\u5206\u985e\uff0c\u5c07\u5728 JupyterLab \u4f7f\u7528\u4e92\u52d5\u5f0f\u4ecb\u9762\u5be6\u4f5c\u5f71\u50cf\u5206\u985e\u6a21\u578b\u7684\u8cc7\u6599\u6536\u96c6(data Collect)\u3001\u8a13\u7df4(Train)\u8207\u9810\u6e2c(Predict)\u3002<\/p>\n","protected":false},"author":1,"featured_media":18158,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[101,238],"tags":[],"table_tags":[],"_links":{"self":[{"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/posts\/17964"}],"collection":[{"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/comments?post=17964"}],"version-history":[{"count":7,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/posts\/17964\/revisions"}],"predecessor-version":[{"id":18164,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/posts\/17964\/revisions\/18164"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/media\/18158"}],"wp:attachment":[{"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/media?parent=17964"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/categories?post=17964"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/tags?post=17964"},{"taxonomy":"table_tags","embeddable":true,"href":"https:\/\/www.circuspi.com\/index.php\/wp-json\/wp\/v2\/table_tags?post=17964"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}