{"id":1164436,"date":"2025-01-15T15:07:22","date_gmt":"2025-01-15T07:07:22","guid":{"rendered":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/1164436.html"},"modified":"2025-01-15T15:07:26","modified_gmt":"2025-01-15T07:07:26","slug":"%e5%a6%82%e4%bd%95%e7%bc%96%e5%86%99python%e4%ba%ba%e7%89%a9%e5%8a%a8%e6%bc%ab%e5%8c%96","status":"publish","type":"post","link":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/1164436.html","title":{"rendered":"\u5982\u4f55\u7f16\u5199Python\u4eba\u7269\u52a8\u6f2b\u5316"},"content":{"rendered":"

\"\u5982\u4f55\u7f16\u5199Python\u4eba\u7269\u52a8\u6f2b\u5316\"<\/p>\n

\u7f16\u5199Python\u4eba\u7269\u52a8\u6f2b\u5316\u7684\u65b9\u6cd5\u5305\u62ec\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u3001\u56fe\u50cf\u5904\u7406\u6280\u672f\u4ee5\u53ca\u73b0\u6709\u7684\u5f00\u6e90\u9879\u76ee\u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u3002<\/strong> \u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u662f\u5176\u4e2d\u6700\u4e3a\u5173\u952e\u7684\u4e00\u70b9\uff0c\u901a\u8fc7\u5229\u7528\u9884\u8bad\u7ec3\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u73b0\u5b9e\u4eba\u7269\u7167\u7247\u8f6c\u6362\u4e3a\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u3002\u4e0b\u9762\u8be6\u7ec6\u63cf\u8ff0\u5982\u4f55\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u8fdb\u884c\u4eba\u7269\u52a8\u6f2b\u5316\u3002<\/p>\n<\/p>\n

\u4e00\u3001\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b<\/h3>\n<\/p>\n

\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u662f\u5b9e\u73b0\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u7684\u6838\u5fc3\u5de5\u5177\u3002\u8fd1\u5e74\u6765\uff0c\u8bb8\u591a\u7814\u7a76\u8005\u5f00\u53d1\u4e86\u5404\u79cd\u7528\u4e8e\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\uff0c\u5982CycleGAN\u3001Pix2Pix\u7b49\u3002\u8fd9\u4e9b\u6a21\u578b\u53ef\u4ee5\u6709\u6548\u5730\u5c06\u8f93\u5165\u7684\u7167\u7247\u8f6c\u6362\u4e3a\u76ee\u6807\u98ce\u683c\u7684\u56fe\u50cf\u3002<\/p>\n<\/p>\n

1.1 CycleGAN\u6a21\u578b<\/h4>\n<\/p>\n

CycleGAN\uff08Cycle-Consistent Generative Adversarial Network\uff09\u662f\u4e00\u79cd\u7528\u4e8e\u56fe\u50cf\u5230\u56fe\u50cf\u7ffb\u8bd1\u7684\u751f\u6210\u5bf9\u6297\u7f51\u7edc\uff08GAN\uff09\u6a21\u578b\u3002\u5b83\u901a\u8fc7\u4e24\u4e2a\u751f\u6210\u5668\u548c\u4e24\u4e2a\u5224\u522b\u5668\u5b9e\u73b0\u65e0\u76d1\u7763\u7684\u56fe\u50cf\u98ce\u683c\u8f6c\u6362\u3002\u4e00\u4e2a\u751f\u6210\u5668\u5c06\u73b0\u5b9e\u7167\u7247\u8f6c\u6362\u4e3a\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\uff0c\u53e6\u4e00\u4e2a\u751f\u6210\u5668\u5219\u5c06\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u8f6c\u6362\u56de\u73b0\u5b9e\u7167\u7247\uff0c\u4ece\u800c\u786e\u4fdd\u8f6c\u6362\u7684\u56fe\u50cf\u4fdd\u6301\u4e00\u81f4\u6027\u3002<\/p>\n<\/p>\n

1.1.1 CycleGAN\u7684\u57fa\u672c\u539f\u7406<\/h5>\n<\/p>\n

CycleGAN\u7684\u6838\u5fc3\u601d\u60f3\u662f\u901a\u8fc7\u4e24\u4e2a\u751f\u6210\u5668\u548c\u4e24\u4e2a\u5224\u522b\u5668\u8fdb\u884c\u8bad\u7ec3\u3002\u751f\u6210\u5668G\u5c06\u73b0\u5b9e\u7167\u7247\u8f6c\u6362\u4e3a\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\uff0c\u751f\u6210\u5668F\u5219\u5c06\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u8f6c\u6362\u56de\u73b0\u5b9e\u7167\u7247\u3002\u5224\u522b\u5668D_X\u7528\u4e8e\u533a\u5206\u73b0\u5b9e\u7167\u7247\u548c\u7531\u751f\u6210\u5668F\u751f\u6210\u7684\u56fe\u50cf\uff0c\u5224\u522b\u5668D_Y\u7528\u4e8e\u533a\u5206\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u548c\u7531\u751f\u6210\u5668G\u751f\u6210\u7684\u56fe\u50cf\u3002<\/p>\n<\/p>\n

1.1.2 CycleGAN\u7684\u5b9e\u73b0<\/h5>\n<\/p>\n

\u8981\u5b9e\u73b0CycleGAN\u6a21\u578b\uff0c\u53ef\u4ee5\u4f7f\u7528PyTorch\u6216TensorFlow\u7b49\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528PyTorch\u5b9e\u73b0CycleGAN\u6a21\u578b\u7684\u57fa\u672c\u6b65\u9aa4\uff1a<\/p>\n<\/p>\n

    \n
  1. \u5bfc\u5165\u5fc5\u8981\u7684\u5e93\uff1a<\/li>\n<\/ol>\n

    import torch<\/p>\n
    import torch.nn as nn<\/p>\n
    import torch.optim as optim<\/p>\n
    from torchvision import datasets, transforms<\/p>\n
    from torch.utils.data import DataLoader<\/p>\n
    <\/code><\/pre>\n<\/p>\n
      \n
    1. \u5b9a\u4e49\u751f\u6210\u5668\u548c\u5224\u522b\u5668\u7f51\u7edc\u7ed3\u6784\uff1a<\/li>\n<\/ol>\n
      class Generator(nn.Module):<\/p>\n
          def __init__(self):<\/p>\n
              super(Generator, self).__init__()<\/p>\n
              # \u5b9a\u4e49\u751f\u6210\u5668\u7684\u7f51\u7edc\u5c42<\/p>\n
          def forward(self, x):<\/p>\n
              # \u5b9e\u73b0\u751f\u6210\u5668\u7684\u524d\u5411\u4f20\u64ad<\/p>\n
              return x<\/p>\n
      class Discriminator(nn.Module):<\/p>\n
          def __init__(self):<\/p>\n
              super(Discriminator, self).__init__()<\/p>\n
              # \u5b9a\u4e49\u5224\u522b\u5668\u7684\u7f51\u7edc\u5c42<\/p>\n
          def forward(self, x):<\/p>\n
              # \u5b9e\u73b0\u5224\u522b\u5668\u7684\u524d\u5411\u4f20\u64ad<\/p>\n
              return x<\/p>\n
      <\/code><\/pre>\n<\/p>\n
        \n
      1. \u5b9a\u4e49\u635f\u5931\u51fd\u6570\u548c\u4f18\u5316\u5668\uff1a<\/li>\n<\/ol>\n
        # \u5b9a\u4e49\u635f\u5931\u51fd\u6570<\/p>\n
        criterion = nn.MSELoss()<\/p>\n
        \u5b9a\u4e49\u751f\u6210\u5668\u548c\u5224\u522b\u5668<\/strong><\/h2>\n
        G = Generator()<\/p>\n
        F = Generator()<\/p>\n
        D_X = Discriminator()<\/p>\n
        D_Y = Discriminator()<\/p>\n
        \u5b9a\u4e49\u4f18\u5316\u5668<\/strong><\/h2>\n
        optimizer_G = optim.Adam(G.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
        optimizer_F = optim.Adam(F.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
        optimizer_D_X = optim.Adam(D_X.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
        optimizer_D_Y = optim.Adam(D_Y.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
        <\/code><\/pre>\n<\/p>\n
          \n
        1. \u8bad\u7ec3CycleGAN\u6a21\u578b\uff1a<\/li>\n<\/ol>\n
          for epoch in range(num_epochs):<\/p>\n
              for i, data in enumerate(dataloader):<\/p>\n
                  real_X, real_Y = data<\/p>\n
                  # \u8bad\u7ec3\u751f\u6210\u5668G\u548cF<\/p>\n
                  optimizer_G.zero_grad()<\/p>\n
                  optimizer_F.zero_grad()<\/p>\n
                  fake_Y = G(real_X)<\/p>\n
                  rec_X = F(fake_Y)<\/p>\n
                  loss_G = criterion(D_Y(fake_Y), torch.ones_like(fake_Y)) + criterion(rec_X, real_X)<\/p>\n
                  loss_G.backward()<\/p>\n
                  optimizer_G.step()<\/p>\n
                  optimizer_F.step()<\/p>\n
                  # \u8bad\u7ec3\u5224\u522b\u5668D_X\u548cD_Y<\/p>\n
                  optimizer_D_X.zero_grad()<\/p>\n
                  optimizer_D_Y.zero_grad()<\/p>\n
                  loss_D_X = criterion(D_X(real_X), torch.ones_like(real_X)) + criterion(D_X(F(fake_Y)), torch.zeros_like(fake_Y))<\/p>\n
                  loss_D_Y = criterion(D_Y(real_Y), torch.ones_like(real_Y)) + criterion(D_Y(G(real_X)), torch.zeros_like(real_X))<\/p>\n
                  loss_D_X.backward()<\/p>\n
                  loss_D_Y.backward()<\/p>\n
                  optimizer_D_X.step()<\/p>\n
                  optimizer_D_Y.step()<\/p>\n
          <\/code><\/pre>\n<\/p>\n
          1.2 Pix2Pix\u6a21\u578b<\/h4>\n<\/p>\n
          Pix2Pix\u662f\u4e00\u79cd\u6761\u4ef6\u751f\u6210\u5bf9\u6297\u7f51\u7edc\uff08cGAN\uff09\u6a21\u578b\uff0c\u5b83\u901a\u8fc7\u6761\u4ef6\u8f93\u5165\u56fe\u50cf\u751f\u6210\u76ee\u6807\u98ce\u683c\u7684\u56fe\u50cf\u3002\u4e0eCycleGAN\u4e0d\u540c\uff0cPix2Pix\u9700\u8981\u6210\u5bf9\u7684\u8f93\u5165\u548c\u8f93\u51fa\u56fe\u50cf\u8fdb\u884c\u8bad\u7ec3\uff0c\u56e0\u6b64\u9002\u7528\u4e8e\u6709\u5bf9\u5e94\u5173\u7cfb\u7684\u6570\u636e\u96c6\u3002<\/p>\n<\/p>\n
          1.2.1 Pix2Pix\u7684\u57fa\u672c\u539f\u7406<\/h5>\n<\/p>\n
          Pix2Pix\u7684\u6838\u5fc3\u601d\u60f3\u662f\u5229\u7528\u6761\u4ef6\u751f\u6210\u5bf9\u6297\u7f51\u7edc\uff0c\u901a\u8fc7\u8f93\u5165\u56fe\u50cf\u548c\u76ee\u6807\u56fe\u50cf\u7684\u6210\u5bf9\u6570\u636e\u8fdb\u884c\u8bad\u7ec3\u3002\u751f\u6210\u5668G\u6839\u636e\u8f93\u5165\u56fe\u50cf\u751f\u6210\u76ee\u6807\u98ce\u683c\u7684\u56fe\u50cf\uff0c\u5224\u522b\u5668D\u5219\u5224\u65ad\u751f\u6210\u7684\u56fe\u50cf\u4e0e\u76ee\u6807\u56fe\u50cf\u662f\u5426\u5339\u914d\u3002<\/p>\n<\/p>\n
          1.2.2 Pix2Pix\u7684\u5b9e\u73b0<\/h5>\n<\/p>\n
          \u8981\u5b9e\u73b0Pix2Pix\u6a21\u578b\uff0c\u53ef\u4ee5\u4f7f\u7528\u4e0eCycleGAN\u7c7b\u4f3c\u7684\u6b65\u9aa4\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528PyTorch\u5b9e\u73b0Pix2Pix\u6a21\u578b\u7684\u57fa\u672c\u6b65\u9aa4\uff1a<\/p>\n<\/p>\n
            \n
          1. \u5bfc\u5165\u5fc5\u8981\u7684\u5e93\uff1a<\/li>\n<\/ol>\n
            import torch<\/p>\n
            import torch.nn as nn<\/p>\n
            import torch.optim as optim<\/p>\n
            from torchvision import datasets, transforms<\/p>\n
            from torch.utils.data import DataLoader<\/p>\n
            <\/code><\/pre>\n<\/p>\n
              \n
            1. \u5b9a\u4e49\u751f\u6210\u5668\u548c\u5224\u522b\u5668\u7f51\u7edc\u7ed3\u6784\uff1a<\/li>\n<\/ol>\n
              class Generator(nn.Module):<\/p>\n
                  def __init__(self):<\/p>\n
                      super(Generator, self).__init__()<\/p>\n
                      # \u5b9a\u4e49\u751f\u6210\u5668\u7684\u7f51\u7edc\u5c42<\/p>\n
                  def forward(self, x):<\/p>\n
                      # \u5b9e\u73b0\u751f\u6210\u5668\u7684\u524d\u5411\u4f20\u64ad<\/p>\n
                      return x<\/p>\n
              class Discriminator(nn.Module):<\/p>\n
                  def __init__(self):<\/p>\n
                      super(Discriminator, self).__init__()<\/p>\n
                      # \u5b9a\u4e49\u5224\u522b\u5668\u7684\u7f51\u7edc\u5c42<\/p>\n
                  def forward(self, x, y):<\/p>\n
                      # \u5b9e\u73b0\u5224\u522b\u5668\u7684\u524d\u5411\u4f20\u64ad<\/p>\n
                      return x<\/p>\n
              <\/code><\/pre>\n<\/p>\n
                \n
              1. \u5b9a\u4e49\u635f\u5931\u51fd\u6570\u548c\u4f18\u5316\u5668\uff1a<\/li>\n<\/ol>\n
                # \u5b9a\u4e49\u635f\u5931\u51fd\u6570<\/p>\n
                criterion = nn.BCELoss()<\/p>\n
                \u5b9a\u4e49\u751f\u6210\u5668\u548c\u5224\u522b\u5668<\/strong><\/h2>\n
                G = Generator()<\/p>\n
                D = Discriminator()<\/p>\n
                \u5b9a\u4e49\u4f18\u5316\u5668<\/strong><\/h2>\n
                optimizer_G = optim.Adam(G.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
                optimizer_D = optim.Adam(D.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/p>\n
                <\/code><\/pre>\n<\/p>\n
                  \n
                1. \u8bad\u7ec3Pix2Pix\u6a21\u578b\uff1a<\/li>\n<\/ol>\n
                  for epoch in range(num_epochs):<\/p>\n
                      for i, data in enumerate(dataloader):<\/p>\n
                          real_X, real_Y = data<\/p>\n
                          # \u8bad\u7ec3\u751f\u6210\u5668G<\/p>\n
                          optimizer_G.zero_grad()<\/p>\n
                          fake_Y = G(real_X)<\/p>\n
                          loss_G = criterion(D(real_X, fake_Y), torch.ones_like(fake_Y)) + criterion(fake_Y, real_Y)<\/p>\n
                          loss_G.backward()<\/p>\n
                          optimizer_G.step()<\/p>\n
                          # \u8bad\u7ec3\u5224\u522b\u5668D<\/p>\n
                          optimizer_D.zero_grad()<\/p>\n
                          loss_D_real = criterion(D(real_X, real_Y), torch.ones_like(real_Y))<\/p>\n
                          loss_D_fake = criterion(D(real_X, fake_Y.detach()), torch.zeros_like(fake_Y))<\/p>\n
                          loss_D = (loss_D_real + loss_D_fake) \/ 2<\/p>\n
                          loss_D.backward()<\/p>\n
                          optimizer_D.step()<\/p>\n
                  <\/code><\/pre>\n<\/p>\n
                  \u4e8c\u3001\u6570\u636e\u96c6\u51c6\u5907<\/h3>\n<\/p>\n
                  \u65e0\u8bba\u662fCycleGAN\u8fd8\u662fPix2Pix\u6a21\u578b\uff0c\u90fd\u9700\u8981\u5927\u91cf\u7684\u56fe\u50cf\u6570\u636e\u8fdb\u884c\u8bad\u7ec3\u3002\u6570\u636e\u96c6\u7684\u9009\u62e9\u548c\u51c6\u5907\u5bf9\u4e8e\u6a21\u578b\u7684\u6027\u80fd\u81f3\u5173\u91cd\u8981\u3002\u5e38\u7528\u7684\u6570\u636e\u96c6\u5305\u62ecCelebA\uff08\u4e00\u4e2a\u5305\u542b\u5927\u91cf\u4eba\u8138\u56fe\u50cf\u7684\u6570\u636e\u96c6\uff09\u548c\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u6570\u636e\u96c6\u3002<\/p>\n<\/p>\n
                  2.1 \u6570\u636e\u96c6\u4e0b\u8f7d\u548c\u9884\u5904\u7406<\/h4>\n<\/p>\n
                    \n
                  1. \n
                    \u4e0b\u8f7d\u6570\u636e\u96c6\uff1a<\/p>\n
                    \u53ef\u4ee5\u4ece\u516c\u5f00\u7684\u56fe\u50cf\u6570\u636e\u96c6\u7f51\u7ad9\u4e0b\u8f7d\u6240\u9700\u7684\u6570\u636e\u96c6\uff0c\u4f8b\u5982CelebA\u6570\u636e\u96c6\u53ef\u4ee5\u4ece\u5b98\u65b9\u7f51\u7ad9\u4e0b\u8f7d\u3002<\/p>\n<\/p>\n<\/li>\n
                  2. \n
                    \u6570\u636e\u9884\u5904\u7406\uff1a<\/p>\n
                    \u6570\u636e\u9884\u5904\u7406\u5305\u62ec\u56fe\u50cf\u7684\u88c1\u526a\u3001\u7f29\u653e\u548c\u6807\u51c6\u5316\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528PyTorch\u8fdb\u884c\u6570\u636e\u9884\u5904\u7406\u7684\u57fa\u672c\u6b65\u9aa4\uff1a<\/p>\n<\/p>\n<\/li>\n<\/ol>\n
                    transform = transforms.Compose([<\/p>\n
                        transforms.Resize((256, 256)),<\/p>\n
                        transforms.ToTensor(),<\/p>\n
                        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))<\/p>\n
                    ])<\/p>\n
                    dataset = datasets.ImageFolder(root='data_path', transform=transform)<\/p>\n
                    dataloader = DataLoader(dataset, batch_size=32, shuffle=True)<\/p>\n
                    <\/code><\/pre>\n<\/p>\n
                    \u4e09\u3001\u56fe\u50cf\u5904\u7406\u6280\u672f<\/h3>\n<\/p>\n
                    \u9664\u4e86\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u5916\uff0c\u8fd8\u53ef\u4ee5\u7ed3\u5408\u4e00\u4e9b\u56fe\u50cf\u5904\u7406\u6280\u672f\u6765\u8fdb\u4e00\u6b65\u63d0\u5347\u56fe\u50cf\u7684\u8d28\u91cf\u548c\u6548\u679c\u3002<\/p>\n<\/p>\n
                    3.1 \u56fe\u50cf\u589e\u5f3a<\/h4>\n<\/p>\n
                    \u56fe\u50cf\u589e\u5f3a\u6280\u672f\u53ef\u4ee5\u5e2e\u52a9\u63d0\u9ad8\u6a21\u578b\u7684\u9c81\u68d2\u6027\u548c\u6cdb\u5316\u80fd\u529b\u3002\u5e38\u7528\u7684\u56fe\u50cf\u589e\u5f3a\u6280\u672f\u5305\u62ec\u65cb\u8f6c\u3001\u7ffb\u8f6c\u3001\u88c1\u526a\u3001\u7f29\u653e\u3001\u989c\u8272\u8c03\u6574\u7b49\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528PyTorch\u8fdb\u884c\u56fe\u50cf\u589e\u5f3a\u7684\u793a\u4f8b\uff1a<\/p>\n<\/p>\n
                    transform = transforms.Compose([<\/p>\n
                        transforms.RandomHorizontalFlip(),<\/p>\n
                        transforms.RandomRotation(10),<\/p>\n
                        transforms.RandomResizedCrop(256),<\/p>\n
                        transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),<\/p>\n
                        transforms.ToTensor(),<\/p>\n
                        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))<\/p>\n
                    ])<\/p>\n
                    dataset = datasets.ImageFolder(root='data_path', transform=transform)<\/p>\n
                    dataloader = DataLoader(dataset, batch_size=32, shuffle=True)<\/p>\n
                    <\/code><\/pre>\n<\/p>\n
                    3.2 \u56fe\u50cf\u540e\u5904\u7406<\/h4>\n<\/p>\n
                    \u56fe\u50cf\u540e\u5904\u7406\u6280\u672f\u53ef\u4ee5\u5e2e\u52a9\u63d0\u5347\u751f\u6210\u56fe\u50cf\u7684\u8d28\u91cf\u548c\u89c6\u89c9\u6548\u679c\u3002\u5e38\u7528\u7684\u56fe\u50cf\u540e\u5904\u7406\u6280\u672f\u5305\u62ec\u53bb\u566a\u3001\u9510\u5316\u3001\u5bf9\u6bd4\u5ea6\u8c03\u6574\u7b49\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528OpenCV\u8fdb\u884c\u56fe\u50cf\u540e\u5904\u7406\u7684\u793a\u4f8b\uff1a<\/p>\n<\/p>\n
                    import cv2<\/p>\n
                    import numpy as np<\/p>\n
                    def post_process_image(image):<\/p>\n
                        # \u8f6c\u6362\u56fe\u50cf\u4e3aOpenCV\u683c\u5f0f<\/p>\n
                        image = image.permute(1, 2, 0).numpy() * 255<\/p>\n
                        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)<\/p>\n
                        # \u53bb\u566a<\/p>\n
                        image = cv2.fastNlMeansDenoisingColored(image, None, 10, 10, 7, 21)<\/p>\n
                        # \u9510\u5316<\/p>\n
                        kernel = np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]])<\/p>\n
                        image = cv2.filter2D(image, -1, kernel)<\/p>\n
                        # \u5bf9\u6bd4\u5ea6\u8c03\u6574<\/p>\n
                        image = cv2.convertScaleAbs(image, alpha=1.5, beta=0)<\/p>\n
                        # \u8f6c\u6362\u56dePyTorch\u683c\u5f0f<\/p>\n
                        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)<\/p>\n
                        image = torch.from_numpy(image).permute(2, 0, 1) \/ 255.0<\/p>\n
                        return image<\/p>\n
                    <\/code><\/pre>\n<\/p>\n
                    \u56db\u3001\u73b0\u6709\u5f00\u6e90\u9879\u76ee<\/h3>\n<\/p>\n
                    \u9664\u4e86\u81ea\u5df1\u5b9e\u73b0\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u6a21\u578b\u5916\uff0c\u8fd8\u53ef\u4ee5\u5229\u7528\u73b0\u6709\u7684\u5f00\u6e90\u9879\u76ee\u6765\u5b9e\u73b0\u4eba\u7269\u52a8\u6f2b\u5316\u3002\u4ee5\u4e0b\u662f\u51e0\u4e2a\u6d41\u884c\u7684\u5f00\u6e90\u9879\u76ee\uff1a<\/p>\n<\/p>\n
                    4.1 CartoonGAN<\/h4>\n<\/p>\n
                    CartoonGAN\u662f\u4e00\u4e2a\u4e13\u95e8\u7528\u4e8e\u5c06\u73b0\u5b9e\u7167\u7247\u8f6c\u6362\u4e3a\u5361\u901a\u98ce\u683c\u56fe\u50cf\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u3002\u5b83\u901a\u8fc7\u751f\u6210\u5bf9\u6297\u7f51\u7edc\u8fdb\u884c\u8bad\u7ec3\uff0c\u5e76\u5728\u751f\u6210\u5668\u4e2d\u4f7f\u7528\u4e86\u4e00\u4e9b\u7279\u5b9a\u7684\u635f\u5931\u51fd\u6570\u6765\u63d0\u5347\u5361\u901a\u98ce\u683c\u7684\u6548\u679c\u3002<\/p>\n<\/p>\n
                    4.1.1 CartoonGAN\u7684\u4f7f\u7528<\/h5>\n<\/p>\n
                    \u53ef\u4ee5\u4eceGitHub\u7b49\u5f00\u6e90\u5e73\u53f0\u4e0b\u8f7dCartoonGAN\u7684\u4ee3\u7801\u548c\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u5e76\u6309\u7167\u8bf4\u660e\u8fdb\u884c\u5b89\u88c5\u548c\u4f7f\u7528\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528CartoonGAN\u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u7684\u793a\u4f8b\uff1a<\/p>\n<\/p>\n
                    from cartoonize import CartoonGAN<\/p>\n
                    \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b<\/strong><\/h2>\n
                    model = CartoonGAN('pretrAI<\/a>ned_model_path')<\/p>\n
                    \u52a0\u8f7d\u8f93\u5165\u56fe\u50cf<\/strong><\/h2>\n
                    input_image = 'input_image_path'<\/p>\n
                    \u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb<\/strong><\/h2>\n
                    output_image = model.cartoonize(input_image)<\/p>\n
                    \u4fdd\u5b58\u8f93\u51fa\u56fe\u50cf<\/strong><\/h2>\n
                    output_image.save('output_image_path')<\/p>\n
                    <\/code><\/pre>\n<\/p>\n
                    4.2 AnimeGAN<\/h4>\n<\/p>\n
                    AnimeGAN\u662f\u53e6\u4e00\u4e2a\u6d41\u884c\u7684\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u6a21\u578b\uff0c\u4e13\u95e8\u7528\u4e8e\u5c06\u73b0\u5b9e\u7167\u7247\u8f6c\u6362\u4e3a\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u3002\u5b83\u901a\u8fc7\u751f\u6210\u5bf9\u6297\u7f51\u7edc\u8fdb\u884c\u8bad\u7ec3\uff0c\u5e76\u5728\u751f\u6210\u5668\u4e2d\u4f7f\u7528\u4e86\u7279\u5b9a\u7684\u635f\u5931\u51fd\u6570\u6765\u63d0\u5347\u52a8\u6f2b\u98ce\u683c\u7684\u6548\u679c\u3002<\/p>\n<\/p>\n
                    4.2.1 AnimeGAN\u7684\u4f7f\u7528<\/h5>\n<\/p>\n
                    \u53ef\u4ee5\u4eceGitHub\u7b49\u5f00\u6e90\u5e73\u53f0\u4e0b\u8f7dAnimeGAN\u7684\u4ee3\u7801\u548c\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u5e76\u6309\u7167\u8bf4\u660e\u8fdb\u884c\u5b89\u88c5\u548c\u4f7f\u7528\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528AnimeGAN\u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u7684\u793a\u4f8b\uff1a<\/p>\n<\/p>\n
                    from animegan import AnimeGAN<\/p>\n
                    \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b<\/strong><\/h2>\n
                    model = AnimeGAN('pretrained_model_path')<\/p>\n
                    \u52a0\u8f7d\u8f93\u5165\u56fe\u50cf<\/strong><\/h2>\n
                    input_image = 'input_image_path'<\/p>\n
                    \u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb<\/strong><\/h2>\n
                    output_image = model.animefy(input_image)<\/p>\n
                    \u4fdd\u5b58\u8f93\u51fa\u56fe\u50cf<\/strong><\/h2>\n
                    output_image.save('output_image_path')<\/p>\n
                    <\/code><\/pre>\n<\/p>\n
                    \u4e94\u3001\u603b\u7ed3<\/h3>\n<\/p>\n
                    \u901a\u8fc7\u672c\u6587\u7684\u4ecb\u7ecd\uff0c\u6211\u4eec\u4e86\u89e3\u4e86\u5982\u4f55\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u3001\u56fe\u50cf\u5904\u7406\u6280\u672f\u4ee5\u53ca\u73b0\u6709\u7684\u5f00\u6e90\u9879\u76ee\u6765\u5b9e\u73b0Python\u4eba\u7269\u52a8\u6f2b\u5316\u3002\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u662f\u5b9e\u73b0\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u7684\u6838\u5fc3\u5de5\u5177\uff0c\u901a\u8fc7\u4f7f\u7528CycleGAN\u3001Pix2Pix\u7b49\u751f\u6210\u5bf9\u6297\u7f51\u7edc\u6a21\u578b\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u73b0\u5b9e\u4eba\u7269\u7167\u7247\u8f6c\u6362\u4e3a\u52a8\u6f2b\u98ce\u683c\u7684\u56fe\u50cf\u3002<\/strong> \u6570\u636e\u96c6\u7684\u51c6\u5907\u548c\u56fe\u50cf\u5904\u7406\u6280\u672f\u4e5f\u662f\u63d0\u5347\u56fe\u50cf\u8d28\u91cf\u7684\u91cd\u8981\u73af\u8282\u3002\u6700\u540e\uff0c\u6211\u4eec\u8fd8\u53ef\u4ee5\u5229\u7528\u73b0\u6709\u7684\u5f00\u6e90\u9879\u76ee\uff0c\u5982CartoonGAN\u548cAnimeGAN\uff0c\u5feb\u901f\u5b9e\u73b0\u4eba\u7269\u52a8\u6f2b\u5316\u6548\u679c\u3002\u5e0c\u671b\u672c\u6587\u80fd\u4e3a\u4f60\u63d0\u4f9b\u6709\u4ef7\u503c\u7684\u53c2\u8003\u548c\u6307\u5bfc\u3002<\/p>\n<\/p>\n
                    \u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n
                     \u5982\u4f55\u9009\u62e9\u5408\u9002\u7684\u5de5\u5177\u6765\u5b9e\u73b0\u4eba\u7269\u52a8\u6f2b\u5316\uff1f<\/strong>
                    \u5728\u7f16\u5199Python\u7a0b\u5e8f\u4ee5\u5b9e\u73b0\u4eba\u7269\u52a8\u6f2b\u5316\u65f6\uff0c\u9009\u62e9\u5408\u9002\u7684\u5de5\u5177\u81f3\u5173\u91cd\u8981\u3002\u5e38\u7528\u7684\u5e93\u5305\u62ecOpenCV\u3001PIL\uff08Python Imaging Library\uff09\u548cTensorFlow\u7b49\u3002OpenCV\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u56fe\u50cf\u5904\u7406\u529f\u80fd\uff0c\u800cPIL\u5219\u9002\u5408\u8fdb\u884c\u57fa\u672c\u7684\u56fe\u50cf\u64cd\u4f5c\u3002\u5bf9\u4e8e\u66f4\u590d\u6742\u7684\u52a8\u6f2b\u5316\u6548\u679c\uff0c\u53ef\u4ee5\u8003\u8651\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u5982TensorFlow\uff0c\u7ed3\u5408\u9884\u8bad\u7ec3\u7684\u6a21\u578b\u6765\u751f\u6210\u66f4\u4e3a\u7ec6\u817b\u7684\u52a8\u6f2b\u98ce\u683c\u3002<\/p>\n
                    \u52a8\u6f2b\u5316\u7684\u6548\u679c\u53ef\u4ee5\u901a\u8fc7\u54ea\u4e9b\u65b9\u5f0f\u8fdb\u884c\u589e\u5f3a\uff1f<\/strong>
                    \u52a8\u6f2b\u5316\u6548\u679c\u7684\u589e\u5f3a\u53ef\u4ee5\u901a\u8fc7\u591a\u79cd\u6280\u672f\u5b9e\u73b0\u3002\u5e94\u7528\u98ce\u683c\u8fc1\u79fb\uff08Style Transfer\uff09\u7b97\u6cd5\u53ef\u4ee5\u5c06\u4eba\u7269\u56fe\u50cf\u4e0e\u7279\u5b9a\u7684\u52a8\u6f2b\u98ce\u683c\u56fe\u50cf\u7ed3\u5408\uff0c\u4ece\u800c\u83b7\u5f97\u66f4\u5177\u827a\u672f\u611f\u7684\u6548\u679c\u3002\u6b64\u5916\uff0c\u8c03\u8282\u8272\u5f69\u5bf9\u6bd4\u5ea6\u3001\u9971\u548c\u5ea6\uff0c\u4ee5\u53ca\u5e94\u7528\u4e0d\u540c\u7684\u6ee4\u955c\u6548\u679c\uff0c\u4e5f\u80fd\u663e\u8457\u63d0\u5347\u6700\u7ec8\u7684\u52a8\u6f2b\u5316\u6548\u679c\u3002\u7528\u6237\u8fd8\u53ef\u4ee5\u5c1d\u8bd5\u624b\u52a8\u8c03\u6574\u56fe\u50cf\u7684\u7ec6\u8282\uff0c\u4f7f\u7528\u56fe\u50cf\u7f16\u8f91\u5de5\u5177\u8fdb\u4e00\u6b65\u4f18\u5316\u3002<\/p>\n
                    \u5982\u4f55\u5904\u7406\u4eba\u7269\u56fe\u50cf\u4ee5\u83b7\u5f97\u6700\u4f73\u52a8\u6f2b\u5316\u6548\u679c\uff1f<\/strong>
                    \u5904\u7406\u4eba\u7269\u56fe\u50cf\u65f6\uff0c\u9996\u5148\u9700\u8981\u786e\u4fdd\u56fe\u50cf\u7684\u6e05\u6670\u5ea6\u548c\u5206\u8fa8\u7387\u9002\u4e2d\u3002\u56fe\u50cf\u9884\u5904\u7406\u6b65\u9aa4\u5305\u62ec\u53bb\u566a\u3001\u8fb9\u7f18\u68c0\u6d4b\u548c\u56fe\u50cf\u7f29\u653e\u7b49\u3002\u4f7f\u7528\u8fb9\u7f18\u68c0\u6d4b\u6280\u672f\u53ef\u4ee5\u5f3a\u5316\u4eba\u7269\u8f6e\u5ed3\uff0c\u800c\u53bb\u566a\u5219\u6709\u52a9\u4e8e\u6d88\u9664\u4e0d\u5fc5\u8981\u7684\u5e72\u6270\u3002\u6b64\u5916\uff0c\u786e\u4fdd\u4eba\u7269\u56fe\u50cf\u7684\u5149\u7ebf\u6761\u4ef6\u826f\u597d\uff0c\u8fd9\u6837\u5728\u52a8\u6f2b\u5316\u8fc7\u7a0b\u4e2d\uff0c\u7ec6\u8282\u548c\u989c\u8272\u4f1a\u66f4\u52a0\u9c9c\u660e\uff0c\u6700\u7ec8\u6548\u679c\u4e5f\u4f1a\u66f4\u52a0\u5438\u5f15\u4eba\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u7f16\u5199Python\u4eba\u7269\u52a8\u6f2b\u5316\u7684\u65b9\u6cd5\u5305\u62ec\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u3001\u56fe\u50cf\u5904\u7406\u6280\u672f\u4ee5\u53ca\u73b0\u6709\u7684\u5f00\u6e90\u9879\u76ee\u8fdb\u884c\u56fe\u50cf\u98ce\u683c\u8fc1\u79fb\u3002 \u6df1\u5ea6\u5b66 […]","protected":false},"author":3,"featured_media":1164438,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1164436"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=1164436"}],"version-history":[{"count":"1","href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1164436\/revisions"}],"predecessor-version":[{"id":1164439,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1164436\/revisions\/1164439"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/1164438"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=1164436"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=1164436"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=1164436"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}