Tensorflow：形状必须为5级，但'conv3D'为1级

Question

import tensorflow as tf
import numpy as tf
if __name__ == '__main__':
    with tf.Graph().as_default():
        inputs = np.zeros([5,10,10,10,3])
        print(inputs)
        outputs = tf.nn.conv3d(
            input=inputs,
            filter=[5, 5, 5, 3, 8],  # filter:[filter_depth, filter_height, file_width, in_channels, out_channels]
            strides=[1, 4, 4, 4, 1],
            padding='SAME'
        )
        # outputs = get_model(inputs, True)
        print(outputs)

When I run the plain code in pycharm, it return the error information:

ValueError: shape must be rank 5 but is rank 1 for 'conv3D'(op: 'Conv3D') with input shapes: [5,10,10,10,3],[5].

Answer 1

引用链接： https : //www.tensorflow.org/api_docs/python/tf/nn/conv3d过滤器应该是张量，具体来说是一个变量（可训练），因此将filter=[5, 5, 5, 3, 8] 5，5，5 filter=[5, 5, 5, 3, 8] filter=tf.get_variable(shape=[5, 5, 5, 3, 8], dtype=tf.float64, name='filter')然后运行它并可以正常运行。