models.py 文件源码

def forward(self, x):
        xs = []
        for i, down in enumerate(self.down):
            if i == 0:
                x_in = x
            elif i == 1:
                x_in = self.pool_top(xs[-1])
            else:
                x_in = self.pool(xs[-1])
            x_out = down(x_in)
            x_out = self.dropout2d(x_out)
            xs.append(x_out)

        x_out = xs[-1]
        for i, (x_skip, up) in reversed(list(enumerate(zip(xs[:-1], self.up)))):
            upsample = self.upsample_top if i == 0 else self.upsample
            x_out = up(torch.cat([upsample(x_out), x_skip], 1))
            x_out = self.dropout2d(x_out)

        x_out = self.conv_final(x_out)
        b = self.hps.patch_border
        return F.sigmoid(x_out[:, :, b:-b, b:-b])