0389fda574d618ec22b488208a1d597a18d7ce72,examples/next-frame.py,NextFrameGAN,forward_pass,#NextFrameGAN#,195

Before Change

        d_real = D(x, context={"c": c})
        self.c = c
        self.d_fake_inputs = []
        rems = [None]+gs[:self.frames-1]
        for g, c in zip(gs[self.frames-1:], gcs):
            rems = rems[1:] + [g]
            d_fake_input = torch.cat(rems, dim=1)
            self.d_fake_inputs.append(d_fake_input)

After Change

    def forward_pass(self, frames, x, cs, gs, gcs, rgs, rcs, loss):
        d_fakes = []
        d_losses = []
        g_losses = []

        D = self.discriminator
        if self.config.discriminator3d:
            if self.config.gcsf:
                c = gcs[0][:,:,None,:,:]
            else:
                c = cs[-1][:,:,None,:,:]
        else:
            c = cs[-1]

        d_real = D(x, context={"c": c})
        self.d_real = d_real
        self.c = c
        self.d_fake_inputs = []
        rems = frames//gs[:self.frames]
        for g, c in zip(gs, gcs):
            rems = rems[1:] + [g]
            d_fake_input = torch.cat(rems, dim=1)
            self.d_fake_inputs.append(d_fake_input)
            d_fake = D(d_fake_input, context={"c": c})
            d_fakes.append(d_fake)
            _d_loss, _g_loss = loss.forward(d_real, d_fake)
            d_losses.append(_d_loss)
            g_losses.append(_g_loss)

        if len(rgs) > 0:
            grems = rgs[:len(rems)]
            rc = rcs[len(rems)-1]
            if config.discriminator3d:
                grems = [g[:,:,None,:,:] for g in grems]
                rc = rc[:,:,None,:,:]
                d_fakes.append(D(torch.cat(grems, dim=2), context={"c":rc}))
            else:
                d_fakes.append(D(torch.cat(grems, dim=1), context={"c":rc}))
            for rg, rc in zip(rgs[len(rems):], rcs[len(rems):]):
                grems = grems[1:] + [rg]
                d_fakes.append(D(torch.cat(grems, dim=1), context={"c":rc}))

        d_loss = sum(d_losses)/len(d_losses)
        g_loss = sum(g_losses)/len(g_losses)
        return d_loss, g_loss

    def forward_video_discriminator(self, cs, gcs, rcs):

Instances