TensorFlow：值错误形状和排名不匹配：ValueError：形状（？，128，128，2）必须有排名2

Question

我在使用反卷积神经网络运行预测时遇到此异常。等级和形状似乎相同，所以我不确定是什么问题。

File "/home/Workspace/image-recognition/app/model/per_pixel_deconv.py", line 141, in Model
    softmax = tf.nn.softmax(output, name=None)

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 1396, in softmax
    result = _op_def_lib.apply_op("Softmax", logits=logits, name=name)

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 703, in apply_op
    op_def=op_def)

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2319, in create_op
    set_shapes_for_outputs(ret)

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1711, in set_shapes_for_outputs
    shapes = shape_func(op)

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 45, in _ShapeFunction
    return [op.inputs[0].get_shape().with_rank(rank)]

File "/home/anaconda2/envs/image-recognition/lib/python2.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 641, in with_rank
    raise ValueError("Shape %s must have rank %d" % (self, rank))
ValueError: Shape (?, 128, 128, 2) must have rank 2

Answer 1

形状 (?, 128, 128, 2) 是 4 阶张量，但 softmax 运算符显然期望 2 阶张量。

在我看来，您使用的是旧版本的 Tensorflow。在 0.10 版中，tf.nn.softmax 要求其输入为 2 级： https://www.tensorflow.org/versions/r0.10/api_docs/python/nn/classification#softmax

在当前版本的 Tensorflow 中，tf.nn.softmax 接受任何等级的输入： https://www.tensorflow.org/api_docs/python/nn/classification#softmax

尝试使用更新版本的 Tensorflow。

希望有帮助！