799aed2e658f677225d4f796e1251f26b52399aa,examples/next-frame.py,NextFrameGAN,forward_loss,#NextFrameGAN#,306

Before Change

                self.xs = []
                for i in range(self.frames-self.per_sample_frames+1):
                    self.xs += [torch.cat(current_inputs[i:i+self.per_sample_frames], dim=1)]
            d_loss, g_loss = self.forward_pass(current_inputs, self.xs, cs, gs, gcs, rgs, rcs, loss)

        if self.config.image_discriminator:
            self.ix = self.last_x

After Change

        self.zs, self.gzs = zs, gzs

        d_loss = torch.tensor([0.0]).cuda()
        g_loss = torch.tensor([0.0]).cuda()

        if self.config.vae:
            g_loss += vae_loss
            self.add_metric("vae", vae_loss)

        if self.config.mse:
            mse = nn.MSELoss()
            mse_loss = mse(self.mse_g, self.last_x)
            d_loss += mse_loss
            g_loss += mse_loss
            self.add_metric("mse", mse_loss)

        if self.config.regularize_c:
            all_cs = cs + gcs
            c_mean = sum([c.mean() for c in all_cs])/len(all_cs)
            self.add_metric("c_mean", c_mean)
            g_loss += c_mean

        if self.config.discriminator:
            if self.config.discriminator3d:
                self.x = torch.cat([x[:, :, None, :, :] for x in current_inputs], dim=2)
            else:
                self.x = torch.cat(current_inputs[:self.per_sample_frames], dim=1)
                self.xs = []
                for i in range(self.frames-self.per_sample_frames+1):
                    self.xs += [torch.cat(current_inputs[i:i+self.per_sample_frames], dim=1)]
            _d_loss, _g_loss = self.forward_pass(current_inputs, self.xs, cs, gs, gcs, rgs, rcs, loss)
            g_loss += _g_loss
            d_loss += _d_loss

        if self.config.image_discriminator:
            self.ix = self.last_x

Instances