e7a02b6293100e21ef15870fd0f9069beaae5290,train.py,,,#,21

Before Change

        g_update_first = True
        batch_size = data.size(0)
        running_batch_sizes += batch_size
        real_label = Variable(torch.ones(batch_size))
        fake_label = Variable(torch.zeros(batch_size))

        ////////////////////////////////////////////////////////
        // (1) Update D network: maximize log(D(x)) + log(1 - D(G(z)))
        //////////////////////////////////////////////////////
        real_img = Variable(target)
        if torch.cuda.is_available():
            real_img = real_img.cuda()
            real_label = real_label.cuda()

        // compute loss of real_img
        real_out = netD(real_img)
        d_loss_real = discriminator_criterion(real_out, real_label)
        real_scores = real_out.data.sum()
        running_real_scores += real_scores

        // compute loss of fake_img
        z = Variable(data)
        if torch.cuda.is_available():
            z = z.cuda()
            fake_label = fake_label.cuda()
        fake_img = netG(z)
        fake_out = netD(fake_img)
        d_loss_fake = discriminator_criterion(fake_out, fake_label)
        fake_scores = fake_out.data.sum()

        d_loss = d_loss_real + d_loss_fake

        // bp and optimize
        optimizerD.zero_grad()
        d_loss.backward(retain_graph=True)
        optimizerD.step()

        ////////////////////////////////////////////////////////
        // (2) Update G network: maximize log(D(G(z)))
        //////////////////////////////////////////////////////
        index = 1
        while ((fabs((real_scores - fake_scores) / batch_size) > G_THRESHOLD) or g_update_first) and (
                    index <= G_STOP_THRESHOLD):
            // compute loss of fake_img
            g_loss = generator_criterion(fake_out, real_label, fake_img, real_img)
            // bp and optimize
            optimizerG.zero_grad()
            g_loss.backward()
            optimizerG.step()
            fake_img = netG(z)
            fake_out = netD(fake_img)
            fake_scores = fake_out.data.sum()
            g_update_first = False
            index += 1

        g_loss = generator_criterion(fake_out, real_label, fake_img, real_img)
        running_g_loss += g_loss.data[0] * batch_size
        d_loss_fake = discriminator_criterion(fake_out, fake_label)
        d_loss = d_loss_real + d_loss_fake
        running_d_loss += d_loss.data[0] * batch_size
        running_fake_scores += fake_scores

        train_bar.set_description(desc="[%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f"

After Change

        g_loss = generator_criterion(fake_out, fake_img, real_img)
        running_g_loss += g_loss.data[0] * batch_size
        d_loss = - torch.mean(torch.log(real_out) + torch.log(1 - fake_out))
        running_d_loss += d_loss.data[0] * batch_size
        running_fake_scores += fake_scores

        train_bar.set_description(desc="[%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f"

Instances