{"id":1060,"date":"2026-05-14T20:04:41","date_gmt":"2026-05-14T12:04:41","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%8812%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-network-architecture-%e4%b8%8e-plans%ef%bc%8c%e4%bb%8e-resenc-preset-%e5%88%b0%e8%87%aa%e5%ae%9a\/"},"modified":"2026-05-14T20:04:41","modified_gmt":"2026-05-14T12:04:41","slug":"%e3%80%8annu-net-0%e5%9f%ba%e7%a1%80%e5%85%a5%e9%97%a8%ef%bc%8812%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-network-architecture-%e4%b8%8e-plans%ef%bc%8c%e4%bb%8e-resenc-preset-%e5%88%b0%e8%87%aa%e5%ae%9a","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%8812%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-network-architecture-%e4%b8%8e-plans%ef%bc%8c%e4%bb%8e-resenc-preset-%e5%88%b0%e8%87%aa%e5%ae%9a\/","title":{"rendered":"\u300annU-Net 0\u57fa\u7840\u5165\u95e8\uff0812\uff09\uff1a\u4fee\u6539 network architecture \u4e0e plans\uff0c\u4ece ResEnc preset \u5230\u81ea\u5b9a\u4e49\u7f51\u7edc\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 12 \u7bc7\uff0c\u4e5f\u662f\u672c\u7cfb\u5217\u6700\u540e\u4e00\u7bc7\u3002\u524d\u9762\u6211\u4eec\u5df2\u7ecf\u5b66\u4f1a\u5b89\u88c5\u3001\u6570\u636e\u51c6\u5907\u3001\u8bad\u7ec3\u3001\u63a8\u7406\u3001\u6a21\u578b\u9009\u62e9\u3001Trainer\u3001loss \u548c augmentation\u3002\u672c\u6587\u8fdb\u5165\u6700\u5bb9\u6613\u201c\u6539\u574f\u201d\u7684\u90e8\u5206\uff1anetwork architecture \u548c plans\u3002<\/p>\n<p>\u8bfb\u5b8c\u672c\u6587\uff0c\u4f60\u5e94\u8be5\u80fd\u591f\uff1a<\/p>\n<ol>\n<li>\u7406\u89e3\u4e3a\u4ec0\u4e48\u7f51\u7edc\u7ed3\u6784\u4e0d\u80fd\u8131\u79bb plans \u5355\u72ec\u4fee\u6539\u3002<\/li>\n<li>\u77e5\u9053 quick-and-dirty Trainer \u8986\u76d6\u8def\u7ebf\u548c proper planner \u8def\u7ebf\u7684\u533a\u522b\u3002<\/li>\n<li>\u4e86\u89e3\u5b98\u65b9 ResEnc presets \u7684\u7528\u6cd5\u548c\u663e\u5b58\u9884\u7b97\u3002<\/li>\n<li>\u77e5\u9053\u66ff\u6362\u7f51\u7edc\u65f6\u5fc5\u987b\u68c0\u67e5 deep supervision\u3001patch size\u3001\u8f93\u5165\u8f93\u51fa\u901a\u9053\u548c\u63a8\u7406\u517c\u5bb9\u3002<\/li>\n<\/ol>\n<h2>1. \u4e3a\u4ec0\u4e48 network architecture \u4e0d\u80fd\u968f\u4fbf\u6362<\/h2>\n<p><strong>network architecture<\/strong> \u6307\u7f51\u7edc\u7ed3\u6784\uff0c\u4f8b\u5982 U-Net\u3001ResEnc U-Net\u3001Transformer\/Mamba \u98ce\u683c\u7f51\u7edc\u7b49\u3002\u5f88\u591a\u4eba\u63a5\u89e6 nnU-Net \u540e\u7b2c\u4e00\u53cd\u5e94\u662f\uff1a\u201c\u6211\u80fd\u4e0d\u80fd\u628a\u91cc\u9762\u7684\u7f51\u7edc\u6362\u6210\u81ea\u5df1\u7684\u6a21\u578b\uff1f\u201d\u7b54\u6848\u662f\u53ef\u4ee5\uff0c\u4f46\u8981\u975e\u5e38\u8c28\u614e\u3002<\/p>\n<p>\u539f\u56e0\u662f nnU-Net v2 \u7684\u7f51\u7edc\u4e0d\u662f\u5b64\u7acb\u6a21\u5757\u3002\u5b83\u548c plans \u4e2d\u7684\u5f88\u591a\u914d\u7f6e\u5f3a\u76f8\u5173\uff1a<\/p>\n<ul>\n<li><code>patch_size<\/code> \u51b3\u5b9a\u7f51\u7edc\u8f93\u5165\u7a7a\u95f4\u5927\u5c0f\u3002<\/li>\n<li><code>num_input_channels<\/code> \u6765\u81ea\u6570\u636e\u96c6\u901a\u9053\u6570\u3002<\/li>\n<li><code>num_output_channels<\/code> \u6765\u81ea label manager\uff0c\u800c\u4e0d\u662f\u7b80\u5355\u7b49\u4e8e\u7c7b\u522b\u6570\u3002<\/li>\n<li><code>pool_op_kernel_sizes<\/code>\u3001<code>conv_kernel_sizes<\/code> \u5f71\u54cd U-Net \u4e0b\u91c7\u6837\u548c\u4e0a\u91c7\u6837\u7ed3\u6784\u3002<\/li>\n<li><code>enable_deep_supervision<\/code> \u51b3\u5b9a\u8bad\u7ec3\u65f6\u662f\u5426\u9700\u8981\u591a\u5c3a\u5ea6\u8f93\u51fa\u3002<\/li>\n<\/ul>\n<pre><code class=\"language-mermaid\">flowchart TD\n    A[nnUNetPlans.json] --> B[configuration_manager]\n    B --> C[patch_size \/ kernels \/ architecture kwargs]\n    C --> D[build_network_architecture]\n    E[label_manager] --> F[num_segmentation_heads]\n    F --> D\n    G[num_input_channels] --> D\n    D --> H[network]\n    H --> I[training and inference]\n<\/code><\/pre>\n<p>\u6240\u4ee5\uff0c\u66ff\u6362\u7f51\u7edc\u4e0d\u662f\u201c\u6539\u4e00\u884c\u6a21\u578b\u7c7b\u540d\u201d\u8fd9\u4e48\u7b80\u5355\u3002\u4f60\u5fc5\u987b\u4fdd\u8bc1\u8bad\u7ec3\u3001\u9a8c\u8bc1\u3001\u63a8\u7406\u90fd\u80fd\u7528\u540c\u4e00\u4e2a\u6784\u5efa\u903b\u8f91\u91cd\u5efa\u7f51\u7edc\u3002<\/p>\n<h2>2. \u5f53\u524d build_network_architecture \u7b7e\u540d<\/h2>\n<p>\u5f53\u524d\u5b98\u65b9 <code>nnUNetTrainer<\/code> \u4e2d\uff0c\u63a8\u8350\u7684 <code>build_network_architecture<\/code> \u7b7e\u540d\u662f\uff1a<\/p>\n<pre><code class=\"language-python\">@staticmethod\ndef build_network_architecture(\n    plans_manager,\n    configuration_manager,\n    num_input_channels: int,\n    num_output_channels: int,\n    enable_deep_supervision: bool = True,\n):\n    ...<\/code><\/pre>\n<p>\u5b98\u65b9\u6e90\u7801\u4ecd\u517c\u5bb9\u65e7\u7b7e\u540d\uff0c\u4f46\u4f1a\u7ed9\u51fa\u5f03\u7528\u8b66\u544a\u3002\u5199\u65b0 Trainer \u65f6\u5e94\u4f7f\u7528\u65b0\u7b7e\u540d\u3002\u8fd9\u4e2a\u51fd\u6570\u4e0d\u4ec5\u8bad\u7ec3\u65f6\u4f1a\u8c03\u7528\uff0c\u63a8\u7406\u52a0\u8f7d\u6a21\u578b\u65f6\u4e5f\u4f1a\u8c03\u7528\u3002\u56e0\u6b64\uff0c\u5982\u679c\u4f60\u901a\u8fc7\u81ea\u5b9a\u4e49 Trainer \u6539\u4e86\u7f51\u7edc\u7ed3\u6784\uff0c\u63a8\u7406\u73af\u5883\u4e5f\u5fc5\u987b\u80fd\u627e\u5230\u540c\u4e00\u4e2a Trainer \u7c7b\u3002<\/p>\n<h2>3. quick-and-dirty \u8def\u7ebf\uff1a\u5728 Trainer \u91cc\u8986\u76d6\u7f51\u7edc\u6784\u5efa<\/h2>\n<p>\u5b98\u65b9\u6269\u5c55\u6587\u6863\u63d0\u4f9b\u4e86\u4e00\u6761 quick-and-dirty \u8def\u7ebf\uff1a\u7ee7\u627f Trainer \u5e76\u8986\u76d6 <code>build_network_architecture<\/code>\u3002\u8fd9\u79cd\u65b9\u5f0f\u9002\u5408\u5feb\u901f\u9a8c\u8bc1\u4e00\u4e2a\u7f51\u7edc\u60f3\u6cd5\u3002<\/p>\n<p>\u4e0b\u9762\u662f\u4e00\u4e2a\u53ea\u7528\u4e8e\u6559\u5b66\u7684\u6700\u5c0f 3D \u7f51\u7edc\u793a\u4f8b\u3002\u4e3a\u4e86\u907f\u514d\u591a\u5c3a\u5ea6\u8f93\u51fa\u548c deep supervision \u7684\u590d\u6742\u6027\uff0c\u8fd9\u4e2a Trainer \u663e\u5f0f\u5173\u95ed deep supervision\u3002\u771f\u5b9e\u7814\u7a76\u4e0d\u8981\u4f7f\u7528\u8fd9\u4e2a\u73a9\u5177\u7f51\u7edc\u4f5c\u4e3a\u5f3a\u6a21\u578b\uff0c\u5b83\u53ea\u662f\u5e2e\u52a9\u4f60\u7406\u89e3\u63a5\u53e3\u3002<\/p>\n<pre><code class=\"language-python\">import torch\nfrom torch import nn\n\nfrom nnunetv2.training.nnUNetTrainer.nnUNetTrainer import nnUNetTrainer\n\n\nclass TinyExampleNet(nn.Module):\n    def __init__(self, in_channels: int, out_channels: int):\n        super().__init__()\n        self.net = nn.Sequential(\n            nn.Conv3d(in_channels, 32, kernel_size=3, padding=1),\n            nn.InstanceNorm3d(32),\n            nn.LeakyReLU(inplace=True),\n            nn.Conv3d(32, out_channels, kernel_size=1),\n        )\n\n    def forward(self, x):\n        return self.net(x)\n\n\nclass nnUNetTrainerTinyNet(nnUNetTrainer):\n    def __init__(self, *args, **kwargs):\n        super().__init__(*args, **kwargs)\n        self.enable_deep_supervision = False\n\n    @staticmethod\n    def build_network_architecture(\n        plans_manager,\n        configuration_manager,\n        num_input_channels: int,\n        num_output_channels: int,\n        enable_deep_supervision: bool = True,\n    ):\n        return TinyExampleNet(\n            num_input_channels,\n            num_output_channels,\n        )<\/code><\/pre>\n<p>\u8fd9\u4e2a\u793a\u4f8b\u53ea\u662f\u4e3a\u4e86\u8bf4\u660e\u63a5\u53e3\uff1a\u771f\u5b9e\u7f51\u7edc\u4e0d\u80fd\u8fd9\u4e48\u7b80\u5355\u3002\u5b83\u7f3a\u5c11 U-Net \u7684\u591a\u5c3a\u5ea6\u4e0a\u4e0b\u6587\u5efa\u6a21\uff0c\u4e5f\u6ca1\u6709 deep supervision \u591a\u5c3a\u5ea6\u8f93\u51fa\u3002\u56e0\u6b64\u5b83\u9002\u5408\u5e2e\u52a9\u4f60\u7406\u89e3\u63a5\u7ebf\u65b9\u5f0f\uff0c\u4e0d\u9002\u5408\u5f53\u6210\u5f3a\u6a21\u578b\u4f7f\u7528\u3002<\/p>\n<p>\u8bad\u7ec3\u547d\u4ee4\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_train 1 3d_fullres 0 -tr nnUNetTrainerTinyNet --npz<\/code><\/pre>\n<h2>4. proper \u8def\u7ebf\uff1a\u901a\u8fc7 planner \u548c plans \u96c6\u6210\u7f51\u7edc<\/h2>\n<p>quick-and-dirty \u8def\u7ebf\u9002\u5408\u5feb\u901f\u5b9e\u9a8c\uff0c\u4f46\u5982\u679c\u4f60\u5e0c\u671b\u7f51\u7edc\u7ed3\u6784\u957f\u671f\u53ef\u7ef4\u62a4\u3001\u53ef\u590d\u73b0\u3001\u53ef\u548c nnU-Net \u81ea\u52a8\u89c4\u5212\u7ed3\u5408\uff0c\u5e94\u8be5\u8003\u8651 proper \u8def\u7ebf\uff1a\u901a\u8fc7\u81ea\u5b9a\u4e49 planner \u6216 plans\uff0c\u8ba9 architecture \u4fe1\u606f\u8fdb\u5165 <code>nnUNetPlans.json<\/code>\u3002<\/p>\n<p>proper \u8def\u7ebf\u7684\u6838\u5fc3\u601d\u60f3\u662f\uff1a<\/p>\n<ul>\n<li>\u8ba9 plans \u660e\u786e\u8bb0\u5f55\u7f51\u7edc\u7c7b\u540d\u3001\u521d\u59cb\u5316\u53c2\u6570\u548c\u9700\u8981 import \u7684\u53c2\u6570\u3002<\/li>\n<li>\u8bad\u7ec3\u548c\u63a8\u7406\u90fd\u4ece plans \u4e2d\u8bfb\u53d6\u67b6\u6784\u914d\u7f6e\u3002<\/li>\n<li>\u4e0d\u540c\u67b6\u6784\u4f7f\u7528\u4e0d\u540c plans identifier\uff0c\u907f\u514d\u8986\u76d6\u9ed8\u8ba4 nnU-Net \u7ed3\u679c\u3002<\/li>\n<\/ul>\n<table>\n<thead>\n<tr>\n<th>\u8def\u7ebf<\/th>\n<th>\u9002\u5408\u573a\u666f<\/th>\n<th>\u4f18\u70b9<\/th>\n<th>\u98ce\u9669<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Trainer \u8986\u76d6<\/td>\n<td>\u5feb\u901f\u9a8c\u8bc1\u7f51\u7edc\u60f3\u6cd5<\/td>\n<td>\u4e0a\u624b\u5feb\uff0c\u6539\u52a8\u96c6\u4e2d<\/td>\n<td>\u5bb9\u6613\u548c plans \u8131\u8282\uff0c\u957f\u671f\u7ef4\u62a4\u5dee<\/td>\n<\/tr>\n<tr>\n<td>Planner \/ plans \u96c6\u6210<\/td>\n<td>\u6b63\u5f0f\u65b9\u6cd5\u3001\u8bba\u6587\u5b9e\u9a8c\u3001\u957f\u671f\u7ef4\u62a4<\/td>\n<td>\u914d\u7f6e\u53ef\u8ffd\u8e2a\uff0c\u8bad\u7ec3\u63a8\u7406\u4e00\u81f4<\/td>\n<td>\u9700\u8981\u7406\u89e3 plans \u548c planner<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>5. \u5b98\u65b9 ResEnc presets\uff1a\u5148\u7528\u5b98\u65b9\u5f3a\u57fa\u7ebf<\/h2>\n<p>\u5728\u81ea\u5df1\u9020\u7f51\u7edc\u4e4b\u524d\uff0c\u5efa\u8bae\u5148\u4e86\u89e3\u5b98\u65b9 Residual Encoder UNet presets\u3002\u5b98\u65b9\u6587\u6863\u7ed9\u51fa\u4e09\u79cd preset\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>Preset<\/th>\n<th>\u5b98\u65b9\u76ee\u6807\u663e\u5b58<\/th>\n<th>\u9002\u5408\u573a\u666f<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>ResEnc M<\/td>\n<td>\u7ea6 9-11GB VRAM<\/td>\n<td>\u663e\u5b58\u63a5\u8fd1\u6807\u51c6 U-Net \u914d\u7f6e<\/td>\n<\/tr>\n<tr>\n<td>ResEnc L<\/td>\n<td>\u7ea6 24GB VRAM<\/td>\n<td>\u5b98\u65b9\u63a8\u8350\u7684\u65b0\u9ed8\u8ba4\u914d\u7f6e<\/td>\n<\/tr>\n<tr>\n<td>ResEnc XL<\/td>\n<td>\u7ea6 40GB VRAM<\/td>\n<td>\u66f4\u5927\u663e\u5b58\u548c\u7b97\u529b\u9884\u7b97<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u5b98\u65b9\u4f7f\u7528\u65b9\u5f0f\u662f\u5148\u5728 plan\/preprocess \u9636\u6bb5\u6307\u5b9a planner\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_plan_and_preprocess -d 1 -pl nnUNetPlannerResEncM\nnnUNetv2_plan_and_preprocess -d 1 -pl nnUNetPlannerResEncL\nnnUNetv2_plan_and_preprocess -d 1 -pl nnUNetPlannerResEncXL<\/code><\/pre>\n<p>\u8bad\u7ec3\u548c\u63a8\u7406\u65f6\u6307\u5b9a\u5bf9\u5e94 plans\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_train 1 3d_fullres 0 -p nnUNetResEncUNetMPlans --npz\nnnUNetv2_predict -i .\/input_images -o .\/predictions -d 1 -c 3d_fullres -p nnUNetResEncUNetMPlans<\/code><\/pre>\n<p>\u5982\u679c\u4f60\u5df2\u7ecf\u6709\u6807\u51c6 2D \u6216 3D fullres \u9884\u5904\u7406\u6570\u636e\uff0c\u5b98\u65b9\u6587\u6863\u8bf4\u660e\u53ef\u7528 <code>nnUNetv2_plan_experiment<\/code> \u907f\u514d\u91cd\u590d preprocessing\uff1b\u4f46\u521d\u5b66\u8005\u7b2c\u4e00\u6b21\u4f7f\u7528\u65f6\uff0c\u76f4\u63a5\u6309\u5b98\u65b9 preset \u547d\u4ee4\u5b8c\u6574\u8dd1\u66f4\u4e0d\u5bb9\u6613\u6df7\u6dc6\u3002<\/p>\n<h2>6. scaling VRAM target \u65f6\u4e0d\u8981\u8986\u76d6\u9ed8\u8ba4 plans<\/h2>\n<p>\u5b98\u65b9 ResEnc preset \u6587\u6863\u8fd8\u8bf4\u660e\uff0c\u53ef\u4ee5\u901a\u8fc7 <code>-gpu_memory_target<\/code> \u8c03\u6574\u76ee\u6807\u663e\u5b58\u9884\u7b97\u3002\u4f46\u5fc5\u987b\u7528 <code>-overwrite_plans_name<\/code> \u6307\u5b9a\u65b0\u7684 plans \u540d\u79f0\uff0c\u907f\u514d\u8986\u76d6 preset plans\u3002<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_plan_experiment \\\n  -d 3 \\\n  -pl nnUNetPlannerResEncM \\\n  -gpu_memory_target 80 \\\n  -overwrite_plans_name nnUNetResEncUNetPlans_80G<\/code><\/pre>\n<p>\u540e\u7eed\u8bad\u7ec3\u7528\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_train 3 3d_fullres 0 -p nnUNetResEncUNetPlans_80G --npz<\/code><\/pre>\n<p>\u5982\u679c\u591a GPU \u8bad\u7ec3\uff0c\u4e0d\u8981\u7b80\u5355\u628a\u591a\u5f20 GPU \u7684\u663e\u5b58\u76f8\u52a0\u4f20\u7ed9 planner\u3002\u5b98\u65b9\u6587\u6863\u63d0\u9192\uff0cpatch size \u5fc5\u987b\u80fd\u88ab\u5355\u5f20 GPU \u5904\u7406\u3002\u591a GPU \u901a\u5e38\u5e94\u5148\u6309\u5355 GPU \u663e\u5b58\u89c4\u5212\uff0c\u518d\u901a\u8fc7 batch size \u6269\u5c55\u3002<\/p>\n<h2>7. \u66ff\u6362\u7f51\u7edc\u524d\u7684\u786c\u6027\u68c0\u67e5<\/h2>\n<table>\n<thead>\n<tr>\n<th>\u68c0\u67e5\u9879<\/th>\n<th>\u5fc5\u987b\u6ee1\u8db3\u4ec0\u4e48<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u8f93\u5165\u901a\u9053<\/td>\n<td>\u7f51\u7edc\u63a5\u53d7 <code>num_input_channels<\/code><\/td>\n<\/tr>\n<tr>\n<td>\u8f93\u51fa\u901a\u9053<\/td>\n<td>\u7f51\u7edc\u8f93\u51fa <code>num_output_channels<\/code>\uff0c\u4e0d\u8981\u81ea\u5df1\u731c\u7c7b\u522b\u6570<\/td>\n<\/tr>\n<tr>\n<td>\u7a7a\u95f4\u7ef4\u5ea6<\/td>\n<td>\u80fd\u5904\u7406\u5f53\u524d <code>patch_size<\/code><\/td>\n<\/tr>\n<tr>\n<td>deep supervision<\/td>\n<td>\u8bad\u7ec3\u542f\u7528\u65f6\u8f93\u51fa\u683c\u5f0f\u8981\u4e0e loss wrapper \u517c\u5bb9<\/td>\n<\/tr>\n<tr>\n<td>\u63a8\u7406\u517c\u5bb9<\/td>\n<td>\u63a8\u7406\u73af\u5883\u80fd import \u540c\u4e00\u4e2a Trainer \u6216 architecture \u7c7b<\/td>\n<\/tr>\n<tr>\n<td>\u663e\u5b58<\/td>\n<td>\u5355\u5f20 GPU \u80fd\u5904\u7406 patch \u548c batch<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>8. benchmarking\uff1a\u522b\u53ea\u548c\u5f31\u57fa\u7ebf\u6bd4<\/h2>\n<p>\u5b98\u65b9 ResEnc preset \u6587\u6863\u6700\u540e\u7279\u522b\u63d0\u9192\uff1a\u5982\u679c\u4f60\u63d0\u51fa\u4e00\u79cd\u65b0\u7684 segmentation method\uff0c\u5e94\u4f7f\u7528\u5408\u9002\u7684 nnU-Net baseline \u505a\u516c\u5e73\u6bd4\u8f83\uff0c\u5e76\u9f13\u52b1\u548c residual encoder variants \u6bd4\u8f83\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u4e0d\u8981\u53ea\u548c\u4e00\u4e2a\u5f88\u5f31\u6216\u672a\u8c03\u597d\u7684 U-Net \u6bd4\uff0c\u7136\u540e\u5ba3\u79f0\u65b0\u65b9\u6cd5\u6709\u6548\u3002<\/p>\n<p>\u5bf9\u7814\u7a76\u5b9e\u9a8c\u6765\u8bf4\uff0c\u81f3\u5c11\u8981\u505a\u5230\uff1a<\/p>\n<ul>\n<li>\u6bd4\u8f83\u9ed8\u8ba4 nnU-Net \u4e0e ResEnc preset\u3002<\/li>\n<li>\u663e\u5b58\u548c\u8bad\u7ec3\u65f6\u95f4\u9884\u7b97\u5c3d\u91cf\u516c\u5e73\u3002<\/li>\n<li>\u4f7f\u7528\u76f8\u540c\u6570\u636e\u5212\u5206\u548c\u76f8\u540c\u8bc4\u4ef7\u6307\u6807\u3002<\/li>\n<li>\u62a5\u544a\u591a\u4e2a fold\uff0c\u800c\u4e0d\u662f\u53ea\u6311\u4e00\u4e2a\u6700\u597d\u7ed3\u679c\u3002<\/li>\n<\/ul>\n<h2>9. \u5b98\u65b9\u8d44\u6599\u5165\u53e3<\/h2>\n<p>\u672c\u6587\u4e3b\u8981\u53c2\u8003\uff1a<\/p>\n<ul>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/extending_nnunet.md\" target=\"_blank\"  rel=\"nofollow\" >Extending nnU-Net<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/reference\/plans-and-configuration.md\" target=\"_blank\"  rel=\"nofollow\" >Plans and Configuration Reference<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/documentation\/resenc_presets.md\" target=\"_blank\"  rel=\"nofollow\" >Residual Encoder Presets in nnU-Net<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/MIC-DKFZ\/nnUNet\/blob\/master\/nnunetv2\/training\/nnUNetTrainer\/nnUNetTrainer.py\" target=\"_blank\"  rel=\"nofollow\" >nnUNetTrainer.py \u5b98\u65b9\u6e90\u7801<\/a><\/li>\n<\/ul>\n<h2>\u672c\u7bc7\u603b\u7ed3<\/h2>\n<p>\u4fee\u6539 network architecture \u662f nnU-Net v2 \u4e8c\u6b21\u5f00\u53d1\u4e2d\u98ce\u9669\u6700\u9ad8\u7684\u90e8\u5206\u3002\u5feb\u901f\u5b9e\u9a8c\u53ef\u4ee5\u901a\u8fc7\u81ea\u5b9a\u4e49 Trainer \u8986\u76d6 <code>build_network_architecture<\/code>\uff1b\u957f\u671f\u7ef4\u62a4\u548c\u6b63\u5f0f\u7814\u7a76\u66f4\u5e94\u901a\u8fc7 planner \u4e0e plans \u96c6\u6210\u3002\u6b63\u5f0f\u63d0\u51fa\u65b0\u7f51\u7edc\u524d\uff0c\u5efa\u8bae\u5148\u5c1d\u8bd5\u5b98\u65b9 ResEnc presets\uff0c\u5e76\u7528\u516c\u5e73\u7684\u663e\u5b58\u3001\u8bad\u7ec3\u65f6\u95f4\u548c 5-fold \u9a8c\u8bc1\u7ed3\u679c\u505a\u6bd4\u8f83\u3002<\/p>\n<h2>\u7cfb\u5217\u603b\u7ed3<\/h2>\n<p>\u5230\u8fd9\u91cc\uff0c\u300annU-Net 0\u57fa\u7840\u5165\u95e8\u300b\u7cfb\u5217\u5b8c\u6210\u4e86\u4ece 0 \u5230\u8fdb\u9636\u4fee\u6539\u7684\u5b8c\u6574\u8def\u5f84\uff1a\u4f60\u5df2\u7ecf\u5b66\u4e60\u4e86 nnU-Net v2 \u7684\u5b9a\u4f4d\u3001\u5b89\u88c5\u3001\u6570\u636e\u683c\u5f0f\u3001\u8bad\u7ec3\u3001\u63a8\u7406\u3001\u6a21\u578b\u9009\u62e9\u3001\u5185\u90e8\u6846\u67b6\uff0c\u4ee5\u53ca\u5982\u4f55\u4fee\u6539 Trainer\u3001loss\u3001augmentation \u548c network architecture\u3002\u4e0b\u4e00\u6b65\u4e0d\u518d\u662f\u7ee7\u7eed\u5806\u6559\u7a0b\uff0c\u800c\u662f\u9009\u62e9\u4e00\u4e2a\u771f\u5b9e\u6570\u636e\u96c6\uff0c\u5efa\u7acb\u9ed8\u8ba4 nnU-Net \u5f3a\u57fa\u7ebf\uff0c\u7136\u540e\u53ea\u6539\u4e00\u4e2a\u53d8\u91cf\uff0c\u505a\u53ef\u590d\u73b0\u7684\u5bf9\u7167\u5b9e\u9a8c\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u7bc7\u8bb2\u89e3 nnU-Net v2 \u4e2d network architecture \u4e0e plans \u7684\u5173\u7cfb\u3001Trainer \u8986\u76d6\u8def\u7ebf\u3001proper planner \u8def\u7ebf\u548c\u5b98\u65b9 ResEnc presets\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-1060","post","type-post","status-publish","format-standard","hentry","category-83"],"views":8,"_links":{"self":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1060","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=1060"}],"version-history":[{"count":0,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1060\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/media?parent=1060"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/categories?post=1060"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/tags?post=1060"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}