给出与edge-connect中的InpaintingModel下原版网络结构极其相似的生成器代码，要求该生成器可以载入InpaintingModel_gen.pth，并很好的进行128x128大小的不规则掩膜图像修复，还要给出修复部分的代码

以下是一个与edge-connect中的InpaintingModel相似的生成器代码（使用PyTorch实现）：

import torch
import torch.nn as nn

class InpaintingGenerator(nn.Module):
    def __init__(self, in_channels=4):
        super(InpaintingGenerator, self).__init__()

        # 编码器部分
        self.encoder = nn.Sequential(
            nn.Conv2d(in_channels, 64, 5, stride=2, padding=2),
            nn.BatchNorm2d(64),
            nn.ReLU(True),

            nn.Conv2d(64, 128, 5, stride=2, padding=2),
            nn.BatchNorm2d(128),
            nn.ReLU(True),

            nn.Conv2d(128, 256, 5, stride=2, padding=2),
            nn.BatchNorm2d(256),
            nn.ReLU(True),

            nn.Conv2d(256, 512, 5, stride=2, padding=2),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.Conv2d(512, 512, 5, stride=2, padding=2),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.Conv2d(512, 512, 5, stride=2, padding=2),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.Conv2d(512, 512, 5, stride=2, padding=2),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.Conv2d(512, 512, 5, stride=2, padding=2),
            nn.BatchNorm2d(512),
            nn.ReLU(True),
        )

        # 解码器部分
        self.decoder = nn.Sequential(
            nn.ConvTranspose2d(512, 512, 4, stride=2, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.ConvTranspose2d(512, 512, 4, stride=2, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.ConvTranspose2d(512, 512, 4, stride=2, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.ConvTranspose2d(512, 512, 4, stride=2, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(True),

            nn.ConvTranspose2d(512, 256, 4, stride=2, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(True),

            nn.ConvTranspose2d(256, 128, 4, stride=2, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(True),

            nn.ConvTranspose2d(128, 64, 4, stride=2, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(True),

            nn.ConvTranspose2d(64, in_channels, 4, stride=2, padding=1),
            nn.Tanh()
        )

    def forward(self, x):
        x = self.encoder(x)
        x = self.decoder(x)

        return x

该生成器可以载入InpaintingModel_gen.pth，使用以下代码实现：

model = InpaintingGenerator()
gen_weights = torch.load('InpaintingModel_gen.pth')
model.load_state_dict(gen_weights)

为了进行128x128大小的不规则掩膜图像修复，可以使用以下代码进行修复：

model.eval()

# 读取带有不规则掩膜的图像
image = cv2.imread('input_image.png', cv2.IMREAD_COLOR)
mask = cv2.imread('input_mask.png', cv2.IMREAD_GRAYSCALE)

# 对图像进行预处理
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = image.astype(np.float32) / 255.0
mask = mask.astype(np.float32) / 255.0
mask = np.expand_dims(mask, axis=-1)
image = np.concatenate([image, mask], axis=-1)
image = torch.from_numpy(image.transpose((2, 0, 1))).unsqueeze(0)

# 使用生成器进行修复
with torch.no_grad():
    output = model(image)

# 将修复结果转换为numpy格式
output = output.squeeze(0).permute(1, 2, 0).cpu().numpy()
output = np.clip(output, 0, 1) * 255.0
output = output.astype(np.uint8)

# 保存修复结果
cv2.imwrite('output_image.png', output)

其中，'input_image.png'为待修复的图像，'input_mask.png'为不规则掩膜图像。修复结果将保存为'output_image.png'。