import torch
import torch.nn as nn


# Define GAN loss: [vanilla | lsgan]
class GANLoss(nn.Module):
    def __init__(self, gan_type, real_label_val=1.0, fake_label_val=0.0):
        super(GANLoss, self).__init__()
        self.gan_type = gan_type.lower()
        self.real_label_val = real_label_val
        self.fake_label_val = fake_label_val

        if self.gan_type == 'vanilla':
            self.loss = nn.BCEWithLogitsLoss()
        elif self.gan_type == 'lsgan':
            self.loss = nn.MSELoss()
        else:
            raise NotImplementedError('GAN type [{:s}] is not found'.format(self.gan_type))

    def get_target_label(self, input, target_is_real):
        if target_is_real:
            return torch.empty_like(input).fill_(self.real_label_val)
        else:
            return torch.empty_like(input).fill_(self.fake_label_val)

    def forward(self, input, target_is_real):
        target_label = self.get_target_label(input, target_is_real)
        loss = self.loss(input, target_label)
        return loss