pytorch使用报错

    Unexpected key(s) in state_dict: \"bn1.num_batches_tracked\", \" 1.0.bn1.num_batches_tracked\", \" 1.0.bn2.num_batches_tracked\", \" 1.0.bn3.num_batches_tracked\", \" 1.0.downsample.1.num_batches_tracked\", \" 1.1.bn1.num_batches_tracked\", \" 1.1.bn2.num_batches_tracked\", \" 1.1.bn3.num_batches_tracked\", \" 1.2.bn1.num_batches_tracked\", \" 1.2.bn2.num_batches_tracked\", \" 1.2.bn3.num_batches_tracked\", \" 2.0.bn1.num_batches_tracked\", \" 2.0.bn2.num_batches_tracked\", \" 2.0.bn3.num_batches_tracked\", \" 2.0.downsample.1.num_batches_tracked\", \" 2.1.bn1.num_batches_tracked\", \" 2.1.bn2.num_batches_tracked\", \" 2.1.bn3.num_batches_tracked\", \" 2.2.bn1.num_batches_tracked\", \" 2.2.bn2.num_batches_tracked\", \" 2.2.bn3.num_batches_tracked\", \" 2.3.bn1.num_batches_tracked\", \" 2.3.bn2.num_batches_tracked\", \" 2.3.bn3.num_batches_tracked\", \" 3.0.bn1.num_batches_tracked\", \" 3.0.bn2.num_batches_tracked\", \" 3.0.bn3.num_batches_tracked\", \" 3.0.downsample.1.num_batches_tracked\", \" 3.1.bn1.num_batches_tracked\", \" 3.1.bn2.num_batches_tracked\", \" 3.1.bn3.num_batches_tracked\", \" 3.2.bn1.num_batches_tracked\", \" 3.2.bn2.num_batches_tracked\", \" 3.2.bn3.num_batches_tracked\", \" 3.3.bn1.num_batches_tracked\", \" 3.3.bn2.num_batches_tracked\", \" 3.3.bn3.num_batches_tracked\", \" 3.4.bn1.num_batches_tracked\", \" 3.4.bn2.num_batches_tracked\", \" 3.4.bn3.num_batches_tracked\", \" 3.5.bn1.num_batches_tracked\", \" 3.5.bn2.num_batches_tracked\", \" 3.5.bn3.num_batches_tracked\", \" 4.0.bn1.num_batches_tracked\", \" 4.0.bn2.num_batches_tracked\", \" 4.0.bn3.num_batches_tracked\", \" 4.0.downsample.1.num_batches_tracked\", \" 4.1.bn1.num_batches_tracked\", \" 4.1.bn2.num_batches_tracked\", \" 4.1.bn3.num_batches_tracked\", \" 4.2.bn1.num_batches_tracked\", \" 4.2.bn2.num_batches_tracked\", \" 4.2.bn3.num_batches_tracked\", \"seg_up_1.conv.1.num_batches_tracked\", \"seg_up_2.conv.1.num_batches_tracked\", \"seg_up_3.conv.1.num_batches_tracked\".

原因分析：使用pytorch0.4.1做训练,pytorch0.4.0做infer。主要问题是pytorch0.4.1添加了num_batches_tracked

修改位置：在batchNorm.py文件当中第28行添加如下两行

        if self.track_running_stats:
            self.register_buffer(\'running_mean\', torch.zeros(num_features))
            self.register_buffer(\'running_var\', torch.ones(num_features))
            self.register_buffer(\'num_batches_tracked\',torch.LongTensor([0]))#fix
        else:
            self.register_parameter(\'running_mean\', None)
            self.register_parameter(\'running_var\', None)
            self.register_buffer(\'num_batches_tracked\',None)#fix