【Hackathon 8th No.23】Improved Training of Wasserstein GANs 论文复现

[robinbg]

PR types

PR changes

Describe

[paddle-bot]

Thanks for your contribution!

[lijialin03]

感谢提交～
当前代码直接运行时会报错，请先检查参数/代码，谢谢

[lijialin03]

运行时找会不到文件，改为动态修正的路径

ROOT_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.append(ROOT_DIR)
from models.wgan_gp import WGAN_GP

[lijialin03]

这里运行时会报错，Paddle 的 nn.Sequential 只能包含继承自 nn.Layer 的对象，不能包含 lambda 表达式，会触发 assert isinstance(layer, Layer) 错误，可以改为类似如下代码：

def __init__(self, noise_dim=100, output_channels=3):
      super(CIFAR10Generator, self).__init__()

      self.layers1 = nn.Sequential(
          nn.Linear(noise_dim, 512 * 4 * 4),
          nn.BatchNorm1D(512 * 4 * 4),
          nn.ReLU(),
      )
      self.layers2 = nn.Sequential(
          nn.Conv2DTranspose(512, 256, 4, 2, 1),
          nn.BatchNorm2D(256),
          nn.ReLU(),
          nn.Conv2DTranspose(256, 128, 4, 2, 1),
          nn.BatchNorm2D(128),
          nn.ReLU(),
          nn.Conv2DTranspose(128, output_channels, 4, 2, 1),
          nn.Tanh(),
      )

    def forward(self, x):
        x = self.layers1(x)
        x = x.reshape([-1, 512, 4, 4])
        x = self.layers2(x)
        return x

[lijialin03]

运行过程中报错：RuntimeError: (Unavailable) The Op flatten_grad doesn't have any grad op. If you don't intend calculating higher order derivatives, please set create_graphto False. (at ../paddle/fluid/eager/api/generated/eager_generated/backwards/nodes.cc:15349)