{"id":1049,"date":"2026-05-14T17:06:04","date_gmt":"2026-05-14T09:06:04","guid":{"rendered":"https:\/\/www.eutaboo.com\/index.php\/2026\/05\/14\/%e3%80%8annu-net-0%e5%9f%ba%e7%a1%80%e5%85%a5%e9%97%a8%ef%bc%881%ef%bc%89%ef%bc%9annu-net-v2-%e6%98%af%e4%bb%80%e4%b9%88%ef%bc%8c%e4%b8%ba%e4%bb%80%e4%b9%88%e5%ae%83%e6%98%af%e5%8c%bb%e5%ad%a6\/"},"modified":"2026-05-14T17:06:04","modified_gmt":"2026-05-14T09:06:04","slug":"%e3%80%8annu-net-0%e5%9f%ba%e7%a1%80%e5%85%a5%e9%97%a8%ef%bc%881%ef%bc%89%ef%bc%9annu-net-v2-%e6%98%af%e4%bb%80%e4%b9%88%ef%bc%8c%e4%b8%ba%e4%bb%80%e4%b9%88%e5%ae%83%e6%98%af%e5%8c%bb%e5%ad%a6","status":"publish","type":"post","link":"https:\/\/www.eutaboo.com\/index.php\/2026\/05\/14\/%e3%80%8annu-net-0%e5%9f%ba%e7%a1%80%e5%85%a5%e9%97%a8%ef%bc%881%ef%bc%89%ef%bc%9annu-net-v2-%e6%98%af%e4%bb%80%e4%b9%88%ef%bc%8c%e4%b8%ba%e4%bb%80%e4%b9%88%e5%ae%83%e6%98%af%e5%8c%bb%e5%ad%a6\/","title":{"rendered":"\u300annU-Net 0\u57fa\u7840\u5165\u95e8\uff081\uff09\uff1annU-Net v2 \u662f\u4ec0\u4e48\uff0c\u4e3a\u4ec0\u4e48\u5b83\u662f\u533b\u5b66\u56fe\u50cf\u5206\u5272\u7684\u5f3a\u57fa\u7ebf\u300b"},"content":{"rendered":"<h2>\u672c\u7bc7\u5b66\u4e60\u76ee\u6807<\/h2>\n<p>\u8fd9\u662f\u300annU-Net 0\u57fa\u7840\u5165\u95e8\u300b\u7cfb\u5217\u7684\u7b2c 1 \u7bc7\u3002\u8bfb\u5b8c\u672c\u6587\uff0c\u4f60\u5e94\u8be5\u80fd\u591f\u56de\u7b54\u56db\u4e2a\u95ee\u9898\uff1a<\/p>\n<ol>\n<li>nnU-Net v2 \u5230\u5e95\u662f\u4ec0\u4e48\uff0c\u5b83\u89e3\u51b3\u7684\u4e0d\u662f\u201c\u9020\u4e00\u4e2a\u65b0\u7f51\u7edc\u201d\uff0c\u800c\u662f\u4ec0\u4e48\u95ee\u9898\uff1f<\/li>\n<li>\u4e3a\u4ec0\u4e48\u533b\u5b66\u56fe\u50cf\u5206\u5272\u4e0d\u80fd\u7b80\u5355\u5957\u4e00\u4e2a\u666e\u901a U-Net \u5c31\u7ed3\u675f\uff1f<\/li>\n<li>nnU-Net v2 \u4ece\u6570\u636e\u5230\u9884\u6d4b\u7684\u5927\u81f4\u6d41\u7a0b\u662f\u4ec0\u4e48\uff1f<\/li>\n<li>\u8fd9\u4e2a\u7cfb\u5217\u540e\u9762\u4f1a\u600e\u6837\u5e26\u4f60\u4ece\u4f1a\u7528 nnU-Net \u8d70\u5230\u80fd\u6539 Trainer\u3001loss\u3001augmentation \u548c network architecture\uff1f<\/li>\n<\/ol>\n<h2>1. \u5148\u8bf4\u7ed3\u8bba\uff1annU-Net v2 \u662f\u4e00\u4e2a\u201c\u81ea\u52a8\u914d\u7f6e\u533b\u5b66\u56fe\u50cf\u5206\u5272\u6d41\u7a0b\u201d\u7684\u6846\u67b6<\/h2>\n<p><strong>nnU-Net<\/strong> \u53ef\u4ee5\u62c6\u5f00\u7406\u89e3\uff1a<strong>U-Net<\/strong> \u662f\u533b\u5b66\u56fe\u50cf\u5206\u5272\u91cc\u975e\u5e38\u7ecf\u5178\u7684\u7f51\u7edc\u7ed3\u6784\uff0c\u800c\u524d\u9762\u7684 <strong>nn<\/strong> \u6765\u81ea \u201cno-new-Net\u201d\u3002\u8fd9\u4e2a\u540d\u5b57\u7684\u91cd\u70b9\u4e0d\u662f\u8bf4\u5b83\u6c38\u8fdc\u4e0d\u7528\u65b0\u7f51\u7edc\uff0c\u800c\u662f\u5f3a\u8c03\uff1a\u5f88\u591a\u533b\u5b66\u5206\u5272\u4efb\u52a1\u771f\u6b63\u96be\u7684\u5730\u65b9\uff0c\u4e0d\u662f\u4e34\u65f6\u8bbe\u8ba1\u4e00\u4e2a\u770b\u8d77\u6765\u66f4\u590d\u6742\u7684\u7f51\u7edc\uff0c\u800c\u662f\u628a\u6574\u4e2a\u8bad\u7ec3\u6d41\u7a0b\u914d\u7f6e\u5bf9\u3002<\/p>\n<p>\u5b98\u65b9\u6587\u6863\u628a nnU-Net v2 \u5b9a\u4f4d\u4e3a\u4e00\u4e2a <strong>semantic segmentation framework<\/strong>\uff0c\u4e5f\u5c31\u662f\u8bed\u4e49\u5206\u5272\u6846\u67b6\u3002\u8bed\u4e49\u5206\u5272\u6307\u7684\u662f\uff1a\u6a21\u578b\u9700\u8981\u7ed9\u56fe\u50cf\u4e2d\u7684\u6bcf\u4e2a\u50cf\u7d20\u6216\u4f53\u7d20\u5206\u914d\u4e00\u4e2a\u7c7b\u522b\u3002\u4f8b\u5982\u5728 CT \u4e2d\u5206\u5272\u809d\u810f\uff0c\u5728 MRI \u4e2d\u5206\u5272\u8111\u80bf\u7624\uff0c\u5728\u663e\u5fae\u56fe\u50cf\u4e2d\u5206\u5272\u7ec6\u80de\u533a\u57df\uff0c\u90fd\u5c5e\u4e8e\u8fd9\u4e2a\u8303\u7574\u3002<\/p>\n<p>\u66f4\u51c6\u786e\u5730\u8bf4\uff0cnnU-Net v2 \u7684\u6838\u5fc3\u4ef7\u503c\u662f\uff1a<\/p>\n<blockquote>\n<p>\u7ed9\u5b9a\u4e00\u4e2a\u7b26\u5408\u683c\u5f0f\u7684\u8bad\u7ec3\u6570\u636e\u96c6\uff0cnnU-Net \u4f1a\u5206\u6790\u6570\u636e\u7279\u5f81\uff0c\u7136\u540e\u81ea\u52a8\u914d\u7f6e\u9884\u5904\u7406\u3001\u7f51\u7edc\u7ed3\u6784\u3001patch size\u3001\u8bad\u7ec3\u7b56\u7565\u3001\u6a21\u578b\u9009\u62e9\u548c\u63a8\u7406\u6d41\u7a0b\uff0c\u5f97\u5230\u4e00\u4e2a\u5f3a\u57fa\u7ebf\u3002<\/p>\n<\/blockquote>\n<p>\u8fd9\u91cc\u7684<strong>\u5f3a\u57fa\u7ebf<\/strong>\u5f88\u91cd\u8981\u3002\u57fa\u7ebf\u4e0d\u662f\u201c\u6700\u4f4e\u6c34\u5e73\u201d\uff0c\u800c\u662f\u4e00\u4e2a\u8db3\u591f\u53ef\u9760\u3001\u53ef\u590d\u73b0\u3001\u5f88\u96be\u968f\u4fbf\u8d85\u8fc7\u7684\u53c2\u8003\u7cfb\u7edf\u3002\u5bf9\u521a\u5f00\u59cb\u505a\u533b\u5b66\u56fe\u50cf\u5206\u5272\u7684\u4eba\u6765\u8bf4\uff0cnnU-Net v2 \u901a\u5e38\u6bd4\u624b\u5199\u4e00\u4e2a U-Net \u66f4\u9002\u5408\u4f5c\u4e3a\u8d77\u70b9\u3002<\/p>\n<h2>2. \u4e3a\u4ec0\u4e48\u533b\u5b66\u56fe\u50cf\u5206\u5272\u4e0d\u80fd\u53ea\u9760\u201c\u4e00\u4e2a U-Net \u6a21\u578b\u201d<\/h2>\n<p>\u5982\u679c\u4f60\u5df2\u7ecf\u6709\u6df1\u5ea6\u5b66\u4e60\u57fa\u7840\uff0c\u53ef\u80fd\u4f1a\u81ea\u7136\u5730\u60f3\uff1a\u65e2\u7136 U-Net \u662f\u5206\u5272\u7f51\u7edc\uff0c\u90a3\u6211\u5199\u4e00\u4e2a U-Net\uff0c\u6362\u6210\u533b\u5b66\u6570\u636e\u8bad\u7ec3\u4e0d\u5c31\u884c\u4e86\u5417\uff1f\u95ee\u9898\u5728\u4e8e\u533b\u5b66\u56fe\u50cf\u7684\u5dee\u5f02\u8fdc\u6bd4\u81ea\u7136\u56fe\u50cf\u5927\u3002<\/p>\n<p>\u5b98\u65b9\u89e3\u91ca\u4e2d\u5217\u51fa\u7684\u5173\u952e\u5dee\u5f02\u5305\u62ec\uff1a2D \u4e0e 3D \u6570\u636e\u3001\u4e0d\u540c\u6a21\u6001\u548c\u901a\u9053\u6570\u3001\u56fe\u50cf\u5c3a\u5bf8\u3001\u4f53\u7d20\u95f4\u8ddd\u3001\u5404\u5411\u5f02\u6027\u3001\u7c7b\u522b\u4e0d\u5e73\u8861\u3001\u76ee\u6807\u7ed3\u6784\u5c5e\u6027\u7b49\u3002\u628a\u8fd9\u4e9b\u7ffb\u8bd1\u6210\u5b9e\u9645\u8bad\u7ec3\u95ee\u9898\uff0c\u5c31\u662f\uff1a<\/p>\n<ul>\n<li><strong>2D\/3D \u5dee\u5f02<\/strong>\uff1a\u4e00\u5f20\u75c5\u7406\u5207\u7247\u53ef\u80fd\u662f 2D \u56fe\u50cf\uff0c\u800c CT\/MRI \u5f80\u5f80\u662f 3D \u4f53\u6570\u636e\u30022D \u5377\u79ef\u548c 3D \u5377\u79ef\u7684\u663e\u5b58\u3001\u4e0a\u4e0b\u6587\u548c\u8bad\u7ec3\u65b9\u5f0f\u5b8c\u5168\u4e0d\u540c\u3002<\/li>\n<li><strong>spacing<\/strong>\uff1aspacing \u6307\u533b\u5b66\u56fe\u50cf\u4e2d\u76f8\u90bb\u50cf\u7d20\u6216\u4f53\u7d20\u5728\u771f\u5b9e\u4e16\u754c\u91cc\u7684\u7269\u7406\u8ddd\u79bb\u3002\u540c\u6837\u662f 512\u00d7512\u00d7100 \u7684\u6570\u636e\uff0c\u4e0d\u540c spacing \u4ee3\u8868\u7684\u771f\u5b9e\u5668\u5b98\u5c3a\u5ea6\u53ef\u80fd\u4e0d\u540c\u3002<\/li>\n<li><strong>anisotropy<\/strong>\uff1a\u5404\u5411\u5f02\u6027\u6307\u4e0d\u540c\u65b9\u5411\u7684\u5206\u8fa8\u7387\u5dee\u5f02\u5f88\u5927\u3002\u4f8b\u5982\u5e73\u9762\u5185\u5f88\u6e05\u695a\uff0c\u4f46\u5c42\u539a\u5f88\u5927\uff0c\u8fd9\u4f1a\u5f71\u54cd\u662f\u5426\u9002\u5408\u76f4\u63a5\u7528 3D \u5377\u79ef\u3002<\/li>\n<li><strong>class imbalance<\/strong>\uff1a\u7c7b\u522b\u4e0d\u5e73\u8861\u6307\u80cc\u666f\u533a\u57df\u5de8\u5927\uff0c\u800c\u76ee\u6807\u533a\u57df\u5f88\u5c0f\u3002\u6bd4\u5982\u5c0f\u80bf\u7624\u53ea\u5360\u6574\u5e45\u56fe\u50cf\u6781\u5c11\u90e8\u5206\uff0c\u666e\u901a\u8bad\u7ec3\u5f88\u5bb9\u6613\u5b66\u6210\u201c\u5168\u9884\u6d4b\u80cc\u666f\u201d\u3002<\/li>\n<li><strong>modality\/channel<\/strong>\uff1a\u6a21\u6001\u6216\u901a\u9053\u53ef\u80fd\u662f CT\u3001T1\u3001T2\u3001FLAIR\u3001\u663e\u5fae\u56fe\u50cf\u4e0d\u540c\u67d3\u8272\u65b9\u5f0f\u7b49\u3002\u4e0d\u540c\u901a\u9053\u7684\u5f3a\u5ea6\u5206\u5e03\u548c\u9884\u5904\u7406\u65b9\u5f0f\u5dee\u5f02\u5f88\u5927\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e9b\u56e0\u7d20\u4f1a\u76f8\u4e92\u5f71\u54cd\u3002\u4f60\u6539 patch size\uff0c\u4f1a\u5f71\u54cd\u663e\u5b58\uff1b\u4f60\u6539 spacing\uff0c\u4f1a\u5f71\u54cd\u76ee\u6807\u5927\u5c0f\uff1b\u4f60\u6539 2D\/3D \u914d\u7f6e\uff0c\u4f1a\u5f71\u54cd\u4e0a\u4e0b\u6587\u8303\u56f4\uff1b\u4f60\u6539 loss\uff0c\u53c8\u4f1a\u548c\u7c7b\u522b\u4e0d\u5e73\u8861\u76f8\u5173\u3002nnU-Net v2 \u7684\u601d\u8def\u662f\uff1a\u4e0d\u8981\u8ba9\u7528\u6237\u624b\u5de5\u62cd\u8111\u888b\u914d\u7f6e\u6240\u6709\u73af\u8282\uff0c\u800c\u662f\u8ba9\u6846\u67b6\u5148\u4ece\u6570\u636e\u4e2d\u63d0\u53d6\u89c4\u5f8b\uff0c\u518d\u7528\u4e00\u5957\u53ef\u590d\u73b0\u7684\u89c4\u5219\u548c\u7ecf\u9a8c\u7b56\u7565\u751f\u6210\u5408\u7406\u914d\u7f6e\u3002<\/p>\n<h2>3. nnU-Net v2 \u7684\u5de5\u4f5c\u6d41\u957f\u4ec0\u4e48\u6837<\/h2>\n<p>\u4ece\u7528\u6237\u89c6\u89d2\u770b\uff0cnnU-Net v2 \u7684\u5b8c\u6574\u6d41\u7a0b\u53ef\u4ee5\u6982\u62ec\u6210\u4e94\u6b65\uff1a<\/p>\n<pre><code class=\"language-mermaid\">flowchart TD\n    A[\u51c6\u5907\u7b26\u5408 nnU-Net \u683c\u5f0f\u7684\u6570\u636e\u96c6] --> B[\u63d0\u53d6 dataset fingerprint]\n    B --> C[\u751f\u6210 plans \u5e76\u6267\u884c preprocessing]\n    C --> D[\u8bad\u7ec3 2d \/ 3d_fullres \/ 3d_lowres \/ cascade \u7b49\u914d\u7f6e]\n    D --> E[\u6bd4\u8f83\u914d\u7f6e\u5e76\u786e\u5b9a postprocessing]\n    E --> F[\u7528\u6700\u4f73\u914d\u7f6e\u8fdb\u884c inference]\n<\/code><\/pre>\n<p>\u8fd9\u91cc\u6709\u51e0\u4e2a\u672f\u8bed\u9700\u8981\u5148\u8ba4\u8bc6\uff1a<\/p>\n<ul>\n<li><strong>dataset fingerprint<\/strong>\uff1a\u53ef\u4ee5\u7406\u89e3\u4e3a\u6570\u636e\u96c6\u201c\u6307\u7eb9\u201d\u3002\u5b83\u603b\u7ed3\u56fe\u50cf\u5c3a\u5bf8\u3001spacing\u3001\u5f3a\u5ea6\u5206\u5e03\u7b49\u6570\u636e\u7279\u5f81\u3002nnU-Net \u540e\u7eed\u5f88\u591a\u914d\u7f6e\u90fd\u4f9d\u8d56\u5b83\u3002<\/li>\n<li><strong>plans<\/strong>\uff1a\u53ef\u4ee5\u7406\u89e3\u4e3a\u5b9e\u9a8c\u8ba1\u5212\u6587\u4ef6\u3002\u5b83\u8bb0\u5f55 nnU-Net \u6839\u636e fingerprint \u751f\u6210\u7684\u9884\u5904\u7406\u3001\u7f51\u7edc\u62d3\u6251\u3001patch size\u3001batch size \u7b49\u914d\u7f6e\u3002<\/li>\n<li><strong>preprocessing<\/strong>\uff1a\u9884\u5904\u7406\u3002\u5305\u62ec\u91cd\u91c7\u6837\u3001\u5f52\u4e00\u5316\u3001\u88c1\u526a\u3001\u4fdd\u5b58\u8bad\u7ec3\u6240\u9700\u7684\u4e2d\u95f4\u6570\u636e\u7b49\u3002<\/li>\n<li><strong>configuration<\/strong>\uff1a\u8bad\u7ec3\u914d\u7f6e\u3002nnU-Net v2 \u5e38\u89c1\u914d\u7f6e\u5305\u62ec <code>2d<\/code>\u3001<code>3d_fullres<\/code>\u3001<code>3d_lowres<\/code> \u548c <code>3d_cascade_fullres<\/code>\u3002<\/li>\n<li><strong>inference<\/strong>\uff1a\u63a8\u7406\u3002\u4e5f\u5c31\u662f\u7528\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u5bf9\u65b0\u75c5\u4f8b\u751f\u6210\u5206\u5272\u7ed3\u679c\u3002<\/li>\n<\/ul>\n<p>\u6ce8\u610f\uff0cnnU-Net v2 \u5e76\u4e0d\u4fdd\u8bc1\u6bcf\u4e2a\u6570\u636e\u96c6\u90fd\u4f1a\u751f\u6210\u6240\u6709\u914d\u7f6e\u3002\u5b98\u65b9\u6587\u6863\u660e\u786e\u8bf4\u660e\uff0ccascade \u53ea\u4f1a\u5728\u6570\u636e\u96c6\u7279\u5f81\u786e\u5b9e\u9700\u8981\u65f6\u624d\u751f\u6210\u3002\u6bd4\u5982\u56fe\u50cf\u672c\u8eab\u8f83\u5c0f\u3001\u5168\u5206\u8fa8\u7387 U-Net \u7684 patch \u5df2\u7ecf\u80fd\u8986\u76d6\u8db3\u591f\u5927\u7684\u533a\u57df\u65f6\uff0c\u5c31\u672a\u5fc5\u9700\u8981\u4f4e\u5206\u8fa8\u7387\u548c cascade\u3002<\/p>\n<h2>4. nnU-Net v2 \u81ea\u52a8\u914d\u7f6e\u7684\u4e0d\u662f\u4e00\u4e2a\u70b9\uff0c\u800c\u662f\u4e00\u6761 pipeline<\/h2>\n<p>\u5f88\u591a\u521d\u5b66\u8005\u7b2c\u4e00\u6b21\u63a5\u89e6 nnU-Net\uff0c\u4f1a\u628a\u5b83\u7406\u89e3\u6210\u201c\u4e00\u4e2a\u7f51\u7edc\u201d\u3002\u8fd9\u4e0d\u591f\u51c6\u786e\u3002nnU-Net v2 \u66f4\u50cf\u662f\u4e00\u6761\u5b8c\u6574 pipeline\u3002pipeline \u6307\u7684\u662f\u4ece\u8f93\u5165\u6570\u636e\u5230\u6700\u7ec8\u8f93\u51fa\u7ed3\u679c\u7684\u4e00\u6574\u5957\u6d41\u6c34\u7ebf\u3002<\/p>\n<table>\n<thead>\n<tr>\n<th>\u73af\u8282<\/th>\n<th>nnU-Net v2 \u4f1a\u505a\u4ec0\u4e48<\/th>\n<th>\u521d\u5b66\u8005\u5e38\u89c1\u8bef\u89e3<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u6570\u636e\u5206\u6790<\/td>\n<td>\u8bfb\u53d6\u8bad\u7ec3\u6570\u636e\uff0c\u63d0\u53d6\u5c3a\u5bf8\u3001spacing\u3001\u5f3a\u5ea6\u7b49 fingerprint<\/td>\n<td>\u4ee5\u4e3a\u53ea\u8981\u628a\u56fe\u7247\u653e\u8fdb\u53bb\u5c31\u80fd\u76f4\u63a5\u8bad\u7ec3<\/td>\n<\/tr>\n<tr>\n<td>\u9884\u5904\u7406<\/td>\n<td>\u6839\u636e\u6570\u636e\u7279\u5f81\u51b3\u5b9a\u91cd\u91c7\u6837\u3001\u5f52\u4e00\u5316\u3001\u88c1\u526a\u7b49\u6b65\u9aa4<\/td>\n<td>\u4ee5\u4e3a\u6240\u6709\u533b\u5b66\u56fe\u50cf\u90fd\u7528\u540c\u4e00\u79cd\u5f52\u4e00\u5316<\/td>\n<\/tr>\n<tr>\n<td>\u7f51\u7edc\u914d\u7f6e<\/td>\n<td>\u81ea\u52a8\u786e\u5b9a 2D\/3D \u914d\u7f6e\u3001patch size\u3001\u7f51\u7edc\u62d3\u6251\u7b49<\/td>\n<td>\u4ee5\u4e3a\u7f51\u7edc\u7ed3\u6784\u5fc5\u987b\u5b8c\u5168\u624b\u52a8\u8bbe\u8ba1<\/td>\n<\/tr>\n<tr>\n<td>\u8bad\u7ec3<\/td>\n<td>\u4f7f\u7528\u6807\u51c6\u5316 Trainer\u3001loss\u3001augmentation \u548c cross-validation<\/td>\n<td>\u53ea\u5173\u6ce8\u6a21\u578b\u4ee3\u7801\uff0c\u5ffd\u7565\u8bad\u7ec3\u7b56\u7565<\/td>\n<\/tr>\n<tr>\n<td>\u6a21\u578b\u9009\u62e9<\/td>\n<td>\u6bd4\u8f83\u4e0d\u540c\u914d\u7f6e\uff0c\u5fc5\u8981\u65f6\u505a ensemble \u548c postprocessing<\/td>\n<td>\u8bad\u7ec3\u5b8c\u4e00\u4e2a fold \u5c31\u76f4\u63a5\u5f53\u6700\u7ec8\u7ed3\u679c<\/td>\n<\/tr>\n<tr>\n<td>\u63a8\u7406<\/td>\n<td>\u6309\u8bad\u7ec3\u65f6\u7684\u914d\u7f6e\u548c\u9884\u5904\u7406\u903b\u8f91\u5bf9\u65b0\u6570\u636e\u9884\u6d4b<\/td>\n<td>\u8bad\u7ec3\u548c\u63a8\u7406\u524d\u5904\u7406\u4e0d\u4e00\u81f4<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48 nnU-Net \u5728\u5f88\u591a\u533b\u5b66\u5206\u5272\u4efb\u52a1\u4e2d\u5f88\u5f3a\uff1a\u5b83\u628a\u5f88\u591a\u201c\u5bb9\u6613\u88ab\u5ffd\u89c6\u4f46\u51b3\u5b9a\u7ed3\u679c\u201d\u7684\u5de5\u7a0b\u7ec6\u8282\u7cfb\u7edf\u5316\u4e86\u3002<\/p>\n<h2>5. nnU-Net v2 \u9002\u5408\u4ec0\u4e48\uff0c\u4e0d\u9002\u5408\u4ec0\u4e48<\/h2>\n<p>nnU-Net v2 \u7279\u522b\u9002\u5408\u4ee5\u4e0b\u573a\u666f\uff1a<\/p>\n<ul>\n<li>\u4f60\u6709\u5e26\u6807\u7b7e\u7684\u533b\u5b66\u56fe\u50cf\u5206\u5272\u6570\u636e\uff0c\u60f3\u5feb\u901f\u5f97\u5230\u53ef\u9760 baseline\u3002<\/li>\n<li>\u4f60\u6b63\u5728\u53c2\u52a0\u533b\u5b66\u56fe\u50cf\u5206\u5272\u6311\u6218\u8d5b\uff0c\u9700\u8981\u5148\u5efa\u7acb\u5f3a\u57fa\u7ebf\u3002<\/li>\n<li>\u4f60\u7684\u6570\u636e\u548c\u81ea\u7136\u56fe\u50cf\u5dee\u5f02\u5f88\u5927\uff0c\u4f7f\u7528 ImageNet \u9884\u8bad\u7ec3\u6a21\u578b\u4e0d\u4e00\u5b9a\u5408\u9002\u3002<\/li>\n<li>\u4f60\u60f3\u7814\u7a76\u65b0\u7684 loss\u3001augmentation\u3001Trainer \u6216\u7f51\u7edc\u7ed3\u6784\uff0c\u9700\u8981\u4e00\u4e2a\u6210\u719f\u6846\u67b6\u4f5c\u4e3a\u5b9e\u9a8c\u5e73\u53f0\u3002<\/li>\n<\/ul>\n<p>\u4f46\u5b83\u4e0d\u9002\u5408\u88ab\u8bef\u7528\u5728\u8fd9\u4e9b\u573a\u666f\uff1a<\/p>\n<ul>\n<li>\u6ca1\u6709\u50cf\u7d20\u7ea7\u6216\u4f53\u7d20\u7ea7\u6807\u6ce8\uff0c\u5374\u5e0c\u671b\u5b83\u81ea\u52a8\u5b66\u4f1a\u5206\u5272\u3002<\/li>\n<li>\u6570\u636e\u683c\u5f0f\u6df7\u4e71\u3001\u6807\u7b7e\u4e0d\u4e00\u81f4\uff0c\u4f46\u4e0d\u613f\u610f\u5148\u6e05\u6d17\u6570\u636e\u3002<\/li>\n<li>\u5e0c\u671b\u4e0d\u7406\u89e3\u4efb\u4f55\u6d41\u7a0b\uff0c\u53ea\u9760\u4e00\u6761\u547d\u4ee4\u5f97\u5230\u53ef\u53d1\u8868\u7ed3\u679c\u3002<\/li>\n<li>\u628a nnU-Net \u5f53\u6210\u4e07\u80fd\u6a21\u578b\uff0c\u5ffd\u7565\u4efb\u52a1\u5b9a\u4e49\u3001\u6570\u636e\u8d28\u91cf\u548c\u5b9e\u9a8c\u9a8c\u8bc1\u3002<\/li>\n<\/ul>\n<p>\u4e00\u53e5\u8bdd\uff1annU-Net v2 \u80fd\u5e2e\u4f60\u5c11\u8d70\u5f88\u591a\u5de5\u7a0b\u5f2f\u8def\uff0c\u4f46\u4e0d\u80fd\u66ff\u4ee3\u6570\u636e\u8d28\u91cf\u3001\u533b\u5b66\u95ee\u9898\u5b9a\u4e49\u548c\u4e25\u8c28\u5b9e\u9a8c\u8bbe\u8ba1\u3002<\/p>\n<h2>6. \u521d\u5b66\u8005\u5e94\u8be5\u600e\u6837\u5b66\u4e60 nnU-Net v2<\/h2>\n<p>\u6211\u5efa\u8bae\u628a\u5b66\u4e60\u8def\u5f84\u5206\u6210\u56db\u5c42\uff1a<\/p>\n<ol>\n<li><strong>\u4f1a\u5b89\u88c5<\/strong>\uff1a\u7406\u89e3\u73af\u5883\u3001PyTorch\u3001CUDA\u3001\u4e09\u4e2a\u6838\u5fc3\u76ee\u5f55\u53d8\u91cf\u3002<\/li>\n<li><strong>\u4f1a\u8dd1\u901a<\/strong>\uff1a\u6574\u7406\u6570\u636e\u96c6\uff0c\u5b8c\u6210 plan\u3001preprocess\u3001train\u3001predict\u3002<\/li>\n<li><strong>\u4f1a\u89e3\u91ca<\/strong>\uff1a\u77e5\u9053 fingerprint\u3001plans\u3001configuration\u3001fold\u3001postprocessing \u5728\u505a\u4ec0\u4e48\u3002<\/li>\n<li><strong>\u4f1a\u4fee\u6539<\/strong>\uff1a\u80fd\u901a\u8fc7\u7ee7\u627f Trainer \u4fee\u6539 loss\u3001augmentation\u3001optimizer\u3001network architecture \u7b49\u6a21\u5757\u3002<\/li>\n<\/ol>\n<p>\u8fd9\u4e2a\u7cfb\u5217\u540e\u7eed\u4e5f\u4f1a\u6309\u8fd9\u4e2a\u987a\u5e8f\u5c55\u5f00\u3002\u524d\u51e0\u7bc7\u5148\u89e3\u51b3\u201c\u600e\u4e48\u8dd1\u8d77\u6765\u201d\uff0c\u4e2d\u95f4\u51e0\u7bc7\u89e3\u51b3\u201c\u600e\u4e48\u5224\u65ad\u8dd1\u5f97\u5bf9\u4e0d\u5bf9\u201d\uff0c\u6700\u540e\u51e0\u7bc7\u8fdb\u5165\u6e90\u7801\u548c\u81ea\u5b9a\u4e49\u6a21\u5757\u3002<\/p>\n<h2>7. \u672c\u7bc7\u53ef\u4ee5\u5148\u8bb0\u4f4f\u7684\u6700\u5c0f\u547d\u4ee4\u5730\u56fe<\/h2>\n<p>\u7b2c 1 \u7bc7\u4e0d\u8981\u6c42\u4f60\u9a6c\u4e0a\u8fd0\u884c nnU-Net\uff0c\u4f46\u4f60\u53ef\u4ee5\u5148\u5bf9\u540e\u9762\u4f1a\u9891\u7e41\u51fa\u73b0\u7684\u547d\u4ee4\u6709\u4e00\u4e2a\u5370\u8c61\u3002nnU-Net v2 \u7684\u547d\u4ee4\u4e00\u822c\u4ee5 <code>nnUNetv2<\/code> \u5f00\u5934\uff1a<\/p>\n<pre><code class=\"language-bash\"># \u67e5\u770b\u547d\u4ee4\u5e2e\u52a9\uff0c\u786e\u8ba4\u5f53\u524d\u5b89\u88c5\u7248\u672c\u652f\u6301\u54ea\u4e9b\u53c2\u6570\nnnUNetv2_plan_and_preprocess -h\nnnUNetv2_train -h\nnnUNetv2_predict -h\n\n# \u540e\u7eed\u5b8c\u6574\u6d41\u7a0b\u4e2d\u4f1a\u51fa\u73b0\u7684\u5178\u578b\u547d\u4ee4\u5f62\u6001\nnnUNetv2_plan_and_preprocess -d DATASET_ID --verify_dataset_integrity\nnnUNetv2_train DATASET_ID 3d_fullres 0\nnnUNetv2_predict -i INPUT_FOLDER -o OUTPUT_FOLDER -d DATASET_ID -c 3d_fullres\n<\/code><\/pre>\n<p>\u8fd9\u91cc\u7684 <code>DATASET_ID<\/code> \u662f\u6570\u636e\u96c6\u7f16\u53f7\uff0c\u4f8b\u5982\u5b98\u65b9\u683c\u5f0f\u4e2d\u5e38\u89c1\u7684 <code>Dataset001_XXX<\/code> \u5bf9\u5e94\u7f16\u53f7 1\u3002\u5177\u4f53\u6570\u636e\u96c6\u683c\u5f0f\u3001\u76ee\u5f55\u7ed3\u6784\u548c\u547d\u4ee4\u53c2\u6570\uff0c\u6211\u4eec\u4f1a\u5728\u540e\u7eed\u6587\u7ae0\u9010\u6b65\u5c55\u5f00\u3002<\/p>\n<h2>8. \u5b98\u65b9\u8d44\u6599\u5165\u53e3<\/h2>\n<p>\u672c\u6587\u4e3b\u8981\u57fa\u4e8e\u4ee5\u4e0b\u5b98\u65b9\u8d44\u6599\u6574\u7406\uff1a<\/p>\n<ul>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\" target=\"_blank\"  rel=\"nofollow\" >MIC-DKFZ\/nnUNet \u5b98\u65b9\u4ed3\u5e93<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/README.md\" target=\"_blank\"  rel=\"nofollow\" >nnU-Net \u5b98\u65b9 documentation \u9996\u9875<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/explanation\/how-nnunet-works.md\" target=\"_blank\"  rel=\"nofollow\" >How nnU-Net Works<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/tldr_migration_guide_from_v1.md\" target=\"_blank\"  rel=\"nofollow\" >TLDR Migration Guide from nnU-Net V1<\/a><\/li>\n<\/ul>\n<h2>\u672c\u7bc7\u603b\u7ed3<\/h2>\n<p>nnU-Net v2 \u4e0d\u662f\u5355\u7eaf\u7684 U-Net \u7f51\u7edc\u4ee3\u7801\uff0c\u800c\u662f\u4e00\u5957\u9762\u5411\u76d1\u7763\u5f0f\u533b\u5b66\u56fe\u50cf\u8bed\u4e49\u5206\u5272\u7684\u81ea\u52a8\u914d\u7f6e\u6846\u67b6\u3002\u5b83\u4f1a\u4ece\u6570\u636e\u4e2d\u63d0\u53d6 fingerprint\uff0c\u751f\u6210 plans\uff0c\u81ea\u52a8\u51b3\u5b9a\u9884\u5904\u7406\u3001\u7f51\u7edc\u914d\u7f6e\u3001\u8bad\u7ec3\u548c\u63a8\u7406\u6d41\u7a0b\u3002\u5bf9\u521d\u5b66\u8005\u6765\u8bf4\uff0c\u5148\u628a\u5b83\u5f53\u4f5c\u5f3a\u57fa\u7ebf\u8dd1\u901a\uff0c\u518d\u9010\u6b65\u7406\u89e3\u5185\u90e8\u673a\u5236\uff0c\u662f\u6700\u7a33\u59a5\u7684\u5b66\u4e60\u8def\u5f84\u3002<\/p>\n<h2>\u4e0b\u4e00\u7bc7\u9884\u544a<\/h2>\n<p>\u4e0b\u4e00\u7bc7\u6211\u4eec\u8fdb\u5165\u5b9e\u64cd\uff1a\u5b89\u88c5 nnU-Net v2\u3002\u91cd\u70b9\u4f1a\u8bb2\u6e05\u695a PyTorch \u5e94\u8be5\u5148\u600e\u4e48\u88c5\uff0c\u4e3a\u4ec0\u4e48\u9700\u8981 <code>nnUNet_raw<\/code>\u3001<code>nnUNet_preprocessed<\/code>\u3001<code>nnUNet_results<\/code> \u8fd9\u4e09\u4e2a\u8def\u5f84\u53d8\u91cf\uff0c\u4ee5\u53ca\u5982\u4f55\u68c0\u67e5\u73af\u5883\u662f\u5426\u771f\u7684\u51c6\u5907\u597d\u4e86\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u7bc7\u4ece\u96f6\u89e3\u91ca nnU-Net v2 \u7684\u5b9a\u4f4d\u3001\u533b\u5b66\u56fe\u50cf\u5206\u5272\u4e3a\u4ec0\u4e48\u9700\u8981\u81ea\u52a8\u914d\u7f6e\uff0c\u4ee5\u53ca nnU-Net \u4ece\u6570\u636e\u5230\u63a8\u7406\u7684\u5b8c\u6574 pipeline\u3002<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"emotion":"","emotion_color":"","title_style":"","license":"","footnotes":""},"categories":[83],"tags":[],"class_list":["post-1049","post","type-post","status-publish","format-standard","hentry","category-83"],"views":2,"_links":{"self":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1049","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/comments?post=1049"}],"version-history":[{"count":0,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1049\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/media?parent=1049"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/categories?post=1049"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/tags?post=1049"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}