{"id":1059,"date":"2026-05-14T19:29:50","date_gmt":"2026-05-14T11:29:50","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%8811%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-data-augmentation-%e4%b8%8e%e9%87%87%e6%a0%b7%e7%ad%96%e7%95%a5%ef%bc%8c%e8%ae%a9-nnu-net\/"},"modified":"2026-05-14T19:29:50","modified_gmt":"2026-05-14T11:29:50","slug":"%e3%80%8annu-net-0%e5%9f%ba%e7%a1%80%e5%85%a5%e9%97%a8%ef%bc%8811%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-data-augmentation-%e4%b8%8e%e9%87%87%e6%a0%b7%e7%ad%96%e7%95%a5%ef%bc%8c%e8%ae%a9-nnu-net","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%8811%ef%bc%89%ef%bc%9a%e4%bf%ae%e6%94%b9-data-augmentation-%e4%b8%8e%e9%87%87%e6%a0%b7%e7%ad%96%e7%95%a5%ef%bc%8c%e8%ae%a9-nnu-net\/","title":{"rendered":"\u300annU-Net 0\u57fa\u7840\u5165\u95e8\uff0811\uff09\uff1a\u4fee\u6539 data augmentation \u4e0e\u91c7\u6837\u7b56\u7565\uff0c\u8ba9 nnU-Net \u9002\u5e94\u4f60\u7684\u6570\u636e\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 11 \u7bc7\u3002\u4e0a\u4e00\u7bc7\u6211\u4eec\u4fee\u6539\u4e86 loss\u3002\u672c\u6587\u7ee7\u7eed\u4fee\u6539\u8bad\u7ec3\u6d41\u7a0b\u4e2d\u7684\u53e6\u4e00\u4e2a\u5173\u952e\u73af\u8282\uff1adata augmentation \u548c\u91c7\u6837\u7b56\u7565\u3002<\/p>\n<p>\u8bfb\u5b8c\u672c\u6587\uff0c\u4f60\u5e94\u8be5\u80fd\u591f\uff1a<\/p>\n<ol>\n<li>\u7406\u89e3 data augmentation \u5728 nnU-Net v2 \u8bad\u7ec3\u6d41\u7a0b\u4e2d\u7684\u4f4d\u7f6e\u3002<\/li>\n<li>\u77e5\u9053\u9ed8\u8ba4\u589e\u5f3a\u5927\u81f4\u5305\u62ec\u54ea\u4e9b\u7c7b\u578b\u3002<\/li>\n<li>\u7406\u89e3\u533b\u5b66\u56fe\u50cf\u589e\u5f3a\u4e3a\u4ec0\u4e48\u4e0d\u80fd\u65e0\u8111\u52a0\u3002<\/li>\n<li>\u901a\u8fc7\u81ea\u5b9a\u4e49 Trainer \u8c03\u6574 mirror \u548c foreground oversampling\u3002<\/li>\n<\/ol>\n<h2>1. data augmentation \u662f\u4ec0\u4e48<\/h2>\n<p><strong>data augmentation<\/strong>\uff0c\u4e2d\u6587\u5e38\u8bd1\u4e3a\u6570\u636e\u589e\u5f3a\uff0c\u6307\u8bad\u7ec3\u65f6\u5bf9\u56fe\u50cf\u548c\u6807\u7b7e\u505a\u968f\u673a\u53d8\u6362\uff0c\u8ba9\u6a21\u578b\u770b\u5230\u66f4\u591a\u53d8\u5316\u5f62\u5f0f\u3002\u5e38\u89c1\u589e\u5f3a\u5305\u62ec\u65cb\u8f6c\u3001\u7f29\u653e\u3001\u7ffb\u8f6c\u3001\u52a0\u566a\u58f0\u3001\u6a21\u7cca\u3001\u4eae\u5ea6\u53d8\u5316\u3001\u5bf9\u6bd4\u5ea6\u53d8\u5316\u7b49\u3002<\/p>\n<p>\u5728\u533b\u5b66\u56fe\u50cf\u5206\u5272\u4e2d\uff0c\u589e\u5f3a\u7684\u76ee\u6807\u4e0d\u662f\u628a\u6570\u636e\u201c\u53d8\u82b1\u54e8\u201d\uff0c\u800c\u662f\u8ba9\u6a21\u578b\u5bf9\u5408\u7406\u53d8\u5316\u66f4\u9c81\u68d2\u3002\u4f8b\u5982\u4e0d\u540c\u626b\u63cf\u4eea\u3001\u4e0d\u540c\u75c5\u4eba\u59ff\u6001\u3001\u4e0d\u540c\u56fe\u50cf\u566a\u58f0\uff0c\u90fd\u53ef\u80fd\u9020\u6210\u8f93\u5165\u53d8\u5316\u3002<\/p>\n<pre><code class=\"language-mermaid\">flowchart TD\n    A[\u9884\u5904\u7406\u540e\u7684 case] --> B[nnUNetDataLoader \u91c7\u6837 patch]\n    B --> C[get_training_transforms]\n    C --> D[\u7a7a\u95f4\u589e\u5f3a \/ \u5f3a\u5ea6\u589e\u5f3a \/ mirror \/ cascade transforms]\n    D --> E[\u8bad\u7ec3 batch]\n    E --> F[network forward + loss]\n<\/code><\/pre>\n<p>\u6ce8\u610f\uff0caugmentation \u53d1\u751f\u5728\u8bad\u7ec3\u65f6\uff0c\u4e0d\u662f preprocessing \u9636\u6bb5\u6c38\u4e45\u5199\u5165\u78c1\u76d8\u3002\u6bcf\u4e2a epoch\u3001\u6bcf\u4e2a batch \u90fd\u53ef\u80fd\u770b\u5230\u4e0d\u540c\u968f\u673a\u53d8\u6362\u3002<\/p>\n<h2>2. nnU-Net v2 \u9ed8\u8ba4\u589e\u5f3a\u5305\u62ec\u4ec0\u4e48<\/h2>\n<p>\u6839\u636e\u5f53\u524d\u5b98\u65b9 <code>nnUNetTrainer.py<\/code> \u4e2d\u7684 <code>get_training_transforms<\/code>\uff0c\u9ed8\u8ba4\u8bad\u7ec3\u589e\u5f3a\u5927\u81f4\u5305\u62ec\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u589e\u5f3a\u7c7b\u578b<\/th>\n<th>\u6e90\u7801\u4e2d\u7684 transform<\/th>\n<th>\u76f4\u89c2\u4f5c\u7528<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u7a7a\u95f4\u53d8\u6362<\/td>\n<td><code>SpatialTransform<\/code><\/td>\n<td>\u65cb\u8f6c\u3001\u7f29\u653e\u3001\u88c1\u526a\u5230 patch<\/td>\n<\/tr>\n<tr>\n<td>\u9ad8\u65af\u566a\u58f0<\/td>\n<td><code>GaussianNoiseTransform<\/code><\/td>\n<td>\u6a21\u62df\u56fe\u50cf\u566a\u58f0<\/td>\n<\/tr>\n<tr>\n<td>\u9ad8\u65af\u6a21\u7cca<\/td>\n<td><code>GaussianBlurTransform<\/code><\/td>\n<td>\u6a21\u62df\u5206\u8fa8\u7387\u6216\u6e05\u6670\u5ea6\u53d8\u5316<\/td>\n<\/tr>\n<tr>\n<td>\u4eae\u5ea6\u53d8\u5316<\/td>\n<td><code>MultiplicativeBrightnessTransform<\/code><\/td>\n<td>\u6a21\u62df\u6574\u4f53\u5f3a\u5ea6\u53d8\u5316<\/td>\n<\/tr>\n<tr>\n<td>\u5bf9\u6bd4\u5ea6\u53d8\u5316<\/td>\n<td><code>ContrastTransform<\/code><\/td>\n<td>\u6a21\u62df\u7ec4\u7ec7\u5bf9\u6bd4\u5dee\u5f02<\/td>\n<\/tr>\n<tr>\n<td>\u4f4e\u5206\u8fa8\u7387\u6a21\u62df<\/td>\n<td><code>SimulateLowResolutionTransform<\/code><\/td>\n<td>\u6a21\u62df\u8f83\u4f4e\u91c7\u6837\u8d28\u91cf<\/td>\n<\/tr>\n<tr>\n<td>gamma \u53d8\u6362<\/td>\n<td><code>GammaTransform<\/code><\/td>\n<td>\u6539\u53d8\u5f3a\u5ea6\u5206\u5e03\u66f2\u7ebf<\/td>\n<\/tr>\n<tr>\n<td>\u955c\u50cf\u7ffb\u8f6c<\/td>\n<td><code>MirrorTransform<\/code><\/td>\n<td>\u6cbf\u5141\u8bb8\u8f74\u968f\u673a\u7ffb\u8f6c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u5982\u679c\u662f cascade \u8bad\u7ec3\uff0c\u8fd8\u4f1a\u6709\u548c\u4e0a\u4e00\u9636\u6bb5\u5206\u5272\u7ed3\u679c\u76f8\u5173\u7684 transform\u3002\u5982\u679c\u542f\u7528 deep supervision\uff0c\u8fd8\u4f1a\u52a0\u5165 <code>DownsampleSegForDSTransform<\/code>\uff0c\u4e3a\u591a\u5c3a\u5ea6\u8f93\u51fa\u51c6\u5907\u6807\u7b7e\u3002<\/p>\n<h2>3. \u533b\u5b66\u56fe\u50cf\u589e\u5f3a\u7684\u98ce\u9669<\/h2>\n<p>\u533b\u5b66\u56fe\u50cf\u589e\u5f3a\u4e0d\u80fd\u7b80\u5355\u7167\u642c\u81ea\u7136\u56fe\u50cf\u3002\u539f\u56e0\u662f\u533b\u5b66\u56fe\u50cf\u91cc\u7684\u65b9\u5411\u3001\u5f3a\u5ea6\u548c\u7a7a\u95f4\u5173\u7cfb\u53ef\u80fd\u6709\u4e34\u5e8a\u542b\u4e49\u3002<\/p>\n<table>\n<thead>\n<tr>\n<th>\u589e\u5f3a<\/th>\n<th>\u53ef\u80fd\u98ce\u9669<\/th>\n<th>\u4ec0\u4e48\u65f6\u5019\u8981\u8c28\u614e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u5de6\u53f3\u7ffb\u8f6c<\/td>\n<td>\u53ef\u80fd\u6539\u53d8\u5de6\u53f3\u89e3\u5256\u8bed\u4e49<\/td>\n<td>\u5de6\u53f3\u5668\u5b98\u3001\u4fa7\u522b\u8bca\u65ad\u3001\u8111\u534a\u7403\u4efb\u52a1<\/td>\n<\/tr>\n<tr>\n<td>\u5927\u89d2\u5ea6\u65cb\u8f6c<\/td>\n<td>\u4ea7\u751f\u4e0d\u771f\u5b9e\u4f53\u4f4d<\/td>\n<td>\u65b9\u5411\u4e25\u683c\u6807\u51c6\u5316\u7684\u6570\u636e<\/td>\n<\/tr>\n<tr>\n<td>\u5f3a\u5ea6\u53d8\u5316<\/td>\n<td>\u7834\u574f\u5b9a\u91cf\u610f\u4e49<\/td>\n<td>CT HU \u503c\u3001\u5b9a\u91cf MRI\u3001\u663e\u5fae\u5b9a\u91cf\u67d3\u8272<\/td>\n<\/tr>\n<tr>\n<td>\u4f4e\u5206\u8fa8\u7387\u6a21\u62df<\/td>\n<td>\u6a21\u7cca\u5c0f\u75c5\u7076<\/td>\n<td>\u5c0f\u76ee\u6807\u5206\u5272\u6216\u8fb9\u754c\u654f\u611f\u4efb\u52a1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u6240\u4ee5\u4fee\u6539 augmentation \u524d\uff0c\u5148\u95ee\u4e00\u4e2a\u95ee\u9898\uff1a\u8fd9\u4e2a\u53d8\u6362\u5728\u4f60\u7684\u533b\u5b66\u4efb\u52a1\u4e2d\u662f\u5426\u4ecd\u7136\u7b26\u5408\u771f\u5b9e\u4e16\u754c\uff1f\u5982\u679c\u4e0d\u7b26\u5408\uff0c\u5b83\u53ef\u80fd\u4e0d\u662f\u589e\u5f3a\uff0c\u800c\u662f\u5728\u5236\u9020\u9519\u8bef\u6837\u672c\u3002<\/p>\n<h2>4. \u6700\u5c0f\u4fee\u6539\u4e00\uff1a\u5173\u95ed mirror<\/h2>\n<p>\u5982\u679c\u4f60\u7684\u4efb\u52a1\u5bf9\u5de6\u53f3\u65b9\u5411\u654f\u611f\uff0c\u53ef\u4ee5\u5148\u5c1d\u8bd5\u5173\u95ed mirror\u3002\u5f53\u524d\u5b98\u65b9 Trainer \u4e2d\uff0cmirror \u8f74\u7531 <code>configure_rotation_dummyDA_mirroring_and_inital_patch_size<\/code> \u8fd4\u56de\uff0c\u5e76\u4fdd\u5b58\u4e3a <code>inference_allowed_mirroring_axes<\/code>\u3002<\/p>\n<p>\u53ef\u4ee5\u5199\u4e00\u4e2a\u81ea\u5b9a\u4e49 Trainer\uff1a<\/p>\n<pre><code class=\"language-python\">from nnunetv2.training.nnUNetTrainer.nnUNetTrainer import nnUNetTrainer\n\n\nclass nnUNetTrainerNoMirror(nnUNetTrainer):\n    def configure_rotation_dummyDA_mirroring_and_inital_patch_size(self):\n        (\n            rotation_for_DA,\n            do_dummy_2d_data_aug,\n            initial_patch_size,\n            mirror_axes,\n        ) = super().configure_rotation_dummyDA_mirroring_and_inital_patch_size()\n\n        mirror_axes = None\n        self.inference_allowed_mirroring_axes = mirror_axes\n        return rotation_for_DA, do_dummy_2d_data_aug, initial_patch_size, mirror_axes<\/code><\/pre>\n<p>\u8bad\u7ec3\u547d\u4ee4\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_train 1 3d_fullres 0 -tr nnUNetTrainerNoMirror --npz<\/code><\/pre>\n<p>\u8fd9\u4e2a\u4fee\u6539\u6bd4\u91cd\u5199\u6574\u4e2a <code>get_training_transforms<\/code> \u66f4\u5c0f\uff0c\u4e5f\u66f4\u9002\u5408\u521d\u5b66\u8005\u9a8c\u8bc1\u4e00\u4e2a\u5177\u4f53\u5047\u8bbe\uff1amirror \u662f\u5426\u5bf9\u5f53\u524d\u4efb\u52a1\u6709\u5bb3\u3002<\/p>\n<h2>5. \u6700\u5c0f\u4fee\u6539\u4e8c\uff1a\u8c03\u6574 foreground oversampling<\/h2>\n<p><strong>foreground oversampling<\/strong> \u6307\u8bad\u7ec3 patch \u91c7\u6837\u65f6\uff0c\u63d0\u9ad8\u5305\u542b\u524d\u666f\u7c7b\u522b\u7684 patch \u6bd4\u4f8b\u3002\u533b\u5b66\u56fe\u50cf\u5206\u5272\u4e2d\uff0c\u76ee\u6807\u7ed3\u6784\u53ef\u80fd\u5f88\u5c0f\u3002\u5982\u679c\u968f\u673a\u91c7\u6837\uff0c\u5927\u91cf patch \u53ef\u80fd\u53ea\u6709\u80cc\u666f\uff0c\u6a21\u578b\u5f88\u96be\u5b66\u5230\u5c0f\u76ee\u6807\u3002<\/p>\n<p>\u5f53\u524d\u5b98\u65b9 Trainer \u4e2d\u6709\u4e00\u4e2a\u8d85\u53c2\u6570\uff1a<\/p>\n<pre><code class=\"language-python\">self.oversample_foreground_percent = 0.33<\/code><\/pre>\n<p>\u53ef\u4ee5\u521b\u5efa\u4e00\u4e2a\u66f4\u504f\u5411\u524d\u666f\u91c7\u6837\u7684 Trainer\uff1a<\/p>\n<pre><code class=\"language-python\">from nnunetv2.training.nnUNetTrainer.nnUNetTrainer import nnUNetTrainer\n\n\nclass nnUNetTrainerMoreForeground(nnUNetTrainer):\n    def __init__(self, *args, **kwargs):\n        super().__init__(*args, **kwargs)\n        self.oversample_foreground_percent = 0.5<\/code><\/pre>\n<p>\u8bad\u7ec3\u547d\u4ee4\uff1a<\/p>\n<pre><code class=\"language-bash\">nnUNetv2_train 1 3d_fullres 0 -tr nnUNetTrainerMoreForeground --npz<\/code><\/pre>\n<p>\u8fd9\u5e76\u4e0d\u7b49\u4e8e\u201c\u4e00\u5b9a\u66f4\u597d\u201d\u3002\u524d\u666f\u91c7\u6837\u592a\u591a\uff0c\u53ef\u80fd\u8ba9\u6a21\u578b\u4f4e\u4f30\u80cc\u666f\u5206\u5e03\uff0c\u6216\u8005\u5bf9\u771f\u5b9e\u63a8\u7406\u573a\u666f\u6cdb\u5316\u53d8\u5dee\u3002\u5b83\u9002\u5408\u5728\u5c0f\u76ee\u6807\u4efb\u52a1\u4e2d\u4f5c\u4e3a\u4e00\u4e2a\u53d7\u63a7\u5b9e\u9a8c\u3002<\/p>\n<h2>6. \u4e0d\u5efa\u8bae\u4e00\u5f00\u59cb\u91cd\u5199\u6574\u4e2a get_training_transforms<\/h2>\n<p><code>get_training_transforms<\/code> \u5f88\u5f3a\u5927\uff0c\u4f46\u4e5f\u5f88\u5bb9\u6613\u6539\u574f\u3002\u5b83\u4e0d\u4ec5\u5305\u542b\u56fe\u50cf\u589e\u5f3a\uff0c\u8fd8\u5305\u542b\uff1a<\/p>\n<ul>\n<li>dummy 2D augmentation\u3002<\/li>\n<li>cascade \u76f8\u5173 transform\u3002<\/li>\n<li>region-based training \u8f6c\u6362\u3002<\/li>\n<li>ignore label \u5904\u7406\u3002<\/li>\n<li>deep supervision \u6807\u7b7e\u4e0b\u91c7\u6837\u3002<\/li>\n<\/ul>\n<p>\u5982\u679c\u4f60\u76f4\u63a5\u590d\u5236\u6574\u6bb5\u5b98\u65b9\u51fd\u6570\u518d\u6539\u51e0\u884c\uff0c\u672a\u6765 nnU-Net \u66f4\u65b0\u540e\u5f88\u5bb9\u6613\u548c\u65b0\u7248\u672c\u8131\u8282\u3002\u66f4\u597d\u7684\u505a\u6cd5\u662f\u5148\u505a\u5c0f\u8303\u56f4\u8986\u76d6\uff0c\u4f8b\u5982\u5173\u95ed mirror\u3001\u8c03\u6574\u91c7\u6837\u6bd4\u4f8b\u6216\u6539\u4e00\u4e2a\u6982\u7387\u53c2\u6570\u3002\u53ea\u6709\u5728\u4f60\u660e\u786e\u77e5\u9053\u6bcf\u4e2a transform \u7684\u8f93\u5165\u8f93\u51fa\u65f6\uff0c\u518d\u91cd\u5199\u5b8c\u6574 pipeline\u3002<\/p>\n<h2>7. \u5b9e\u9a8c\u5bf9\u7167\u8bbe\u8ba1<\/h2>\n<p>augmentation \u548c\u91c7\u6837\u7b56\u7565\u7684\u5b9e\u9a8c\u5fc5\u987b\u505a\u5bf9\u7167\u3002\u63a8\u8350\u81f3\u5c11\u6bd4\u8f83\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u5b9e\u9a8c<\/th>\n<th>Trainer<\/th>\n<th>\u76ee\u7684<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u9ed8\u8ba4\u57fa\u7ebf<\/td>\n<td><code>nnUNetTrainer<\/code><\/td>\n<td>\u786e\u8ba4\u5b98\u65b9\u9ed8\u8ba4\u8868\u73b0<\/td>\n<\/tr>\n<tr>\n<td>\u5173\u95ed mirror<\/td>\n<td><code>nnUNetTrainerNoMirror<\/code><\/td>\n<td>\u9a8c\u8bc1\u7ffb\u8f6c\u662f\u5426\u4f24\u5bb3\u5de6\u53f3\u8bed\u4e49<\/td>\n<\/tr>\n<tr>\n<td>\u66f4\u591a\u524d\u666f\u91c7\u6837<\/td>\n<td><code>nnUNetTrainerMoreForeground<\/code><\/td>\n<td>\u9a8c\u8bc1\u5c0f\u76ee\u6807\u53ec\u56de\u662f\u5426\u6539\u5584<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u6bd4\u8f83\u65f6\u4e0d\u8981\u53ea\u770b\u5e73\u5747 Dice\u3002\u8fd8\u5e94\u68c0\u67e5\u6bcf\u4e2a\u7c7b\u522b\u7684 Dice\u3001\u5c0f\u76ee\u6807\u53ec\u56de\u3001\u8bef\u62a5\u533a\u57df\u548c\u53ef\u89c6\u5316\u7ed3\u679c\u3002\u589e\u5f3a\u7b56\u7565\u6709\u65f6\u4f1a\u6539\u5584\u6574\u4f53\u5e73\u5747\u503c\uff0c\u5374\u8ba9\u67d0\u4e2a\u4e34\u5e8a\u5173\u952e\u7c7b\u522b\u53d8\u5dee\u3002<\/p>\n<h2>8. \u8c03\u8bd5\u5efa\u8bae<\/h2>\n<ul>\n<li>\u5148\u7528\u77ed\u8bad\u7ec3\u53d8\u4f53\u68c0\u67e5\u81ea\u5b9a\u4e49 Trainer \u80fd\u5426\u8dd1\u8d77\u6765\u3002<\/li>\n<li>\u6bcf\u6b21\u53ea\u6539\u4e00\u4e2a\u589e\u5f3a\u6216\u91c7\u6837\u53c2\u6570\u3002<\/li>\n<li>\u6539 mirror \u65f6\uff0c\u540c\u6b65\u5173\u6ce8 inference mirroring\uff0c\u56e0\u4e3a\u8bad\u7ec3\u548c\u63a8\u7406\u65f6\u7684 mirroring \u8bed\u4e49\u5e94\u4e00\u81f4\u3002<\/li>\n<li>\u6539\u91c7\u6837\u6bd4\u4f8b\u540e\uff0c\u89c2\u5bdf\u8bad\u7ec3 loss\u3001\u9a8c\u8bc1 Dice \u548c\u5c0f\u76ee\u6807\u7c7b\u522b\u6307\u6807\u3002<\/li>\n<li>\u53ef\u89c6\u5316\u589e\u5f3a\u540e\u7684 patch\uff0c\u786e\u8ba4\u6ca1\u6709\u4ea7\u751f\u660e\u663e\u4e0d\u5408\u7406\u56fe\u50cf\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\/nnunetv2\/training\/nnUNetTrainer\/nnUNetTrainer.py\" target=\"_blank\"  rel=\"nofollow\" >nnUNetTrainer.py \u5b98\u65b9\u6e90\u7801<\/a><\/li>\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\/how-to\/train-models.md\" target=\"_blank\"  rel=\"nofollow\" >Train Models<\/a><\/li>\n<\/ul>\n<h2>\u672c\u7bc7\u603b\u7ed3<\/h2>\n<p>nnU-Net v2 \u9ed8\u8ba4 augmentation \u5df2\u7ecf\u8986\u76d6\u7a7a\u95f4\u53d8\u6362\u3001\u5f3a\u5ea6\u53d8\u5316\u3001\u566a\u58f0\u3001\u6a21\u7cca\u3001\u4f4e\u5206\u8fa8\u7387\u6a21\u62df\u3001gamma \u548c mirror \u7b49\u591a\u79cd\u60c5\u51b5\u3002\u4fee\u6539\u589e\u5f3a\u7b56\u7565\u65f6\uff0c\u5e94\u4ece\u533b\u5b66\u4efb\u52a1\u5408\u7406\u6027\u51fa\u53d1\uff0c\u800c\u4e0d\u662f\u76f2\u76ee\u589e\u52a0\u53d8\u6362\u3002\u521d\u5b66\u8005\u6700\u9002\u5408\u4ece\u5c0f\u6539\u52a8\u5f00\u59cb\uff0c\u4f8b\u5982\u5173\u95ed mirror \u6216\u8c03\u6574 foreground oversampling\uff0c\u5e76\u7528\u4e25\u683c\u5bf9\u7167\u5b9e\u9a8c\u5224\u65ad\u662f\u5426\u771f\u7684\u6709\u6548\u3002<\/p>\n<h2>\u4e0b\u4e00\u7bc7\u9884\u544a<\/h2>\n<p>\u4e0b\u4e00\u7bc7\u662f\u672c\u7cfb\u5217\u6700\u540e\u4e00\u7bc7\uff1a\u4fee\u6539 network architecture \u4e0e plans\u3002\u6211\u4eec\u4f1a\u8bb2 quick-and-dirty \u7684 Trainer \u8986\u76d6\u8def\u7ebf\u3001proper planner \u8def\u7ebf\u3001ResEnc preset\uff0c\u4ee5\u53ca\u4e3a\u4ec0\u4e48\u7f51\u7edc\u66ff\u6362\u5fc5\u987b\u5904\u7406 deep supervision\u3001patch size \u548c\u663e\u5b58\u7ea6\u675f\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u7bc7\u8bb2\u89e3 nnU-Net v2 \u9ed8\u8ba4 data augmentation\u3001\u533b\u5b66\u56fe\u50cf\u589e\u5f3a\u98ce\u9669\uff0c\u4ee5\u53ca\u5982\u4f55\u901a\u8fc7 Trainer \u8c03\u6574 mirror \u548c foreground oversampling\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-1059","post","type-post","status-publish","format-standard","hentry","category-83"],"views":6,"_links":{"self":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1059","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=1059"}],"version-history":[{"count":0,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/posts\/1059\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/media?parent=1059"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/categories?post=1059"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.eutaboo.com\/index.php\/wp-json\/wp\/v2\/tags?post=1059"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}