Tensorflow——Cannot reshape a tensor with 6553600 elements to shape的问题解决_cannot reshape tensor of 0 elements into shape-CSDN博客

本文链接：https://blog.csdn.net/qq_45704645/article/details/109693611

该博客介绍了在TensorFlow中遇到的`ValueError: Cannot reshape a tensor`问题，原因是尝试将具有6553600个元素的张量重塑为不匹配的形状。解决方案是调整卷积层的卷积核数量，从32改为64，以匹配张量的元素数量。这表明在构建神经网络模型时，确保张量维度的一致性至关重要。

摘要由CSDN通过智能技术生成

ValueError: Cannot reshape a tensor with 6553600 elements to shape [100,32,32,32] (3276800 elements) 问题解决

错误代码：

x = tf.placeholder(tf.float32, [100, 3072])
x_image = tf.reshape(x, [-1, 32, 32, 3])
# x = tf.placeholder(dtype=tf.float32, shape=[None, 32, 32, 3])

y_ = tf.placeholder(dtype=tf.float32, shape=[None, 10])
W_e_conv1 = weight_variable([5, 5, 3, 32], "w_e_conv1")
b_e_conv1 = bias_variable([32], "b_e_conv1")
h_e_conv1 = tf.nn.relu(tf.add(conv2d(x_image, W_e_conv1), b_e_conv1))#(?,32,32,64)
conv1 = attention(h_e_conv1, 32, sn=False, scope='attention', reuse=False)

可以看出是32个5*5的卷积核

完整错误形式：

ValueError:Cannot reshape a tensor with 6553600 elements to shape [100,32,32,32] (3276800 elements) for 'attention/Reshape_3' (op: 'Reshape') with input shapes: [100,1024,64], [4] and with input tensors computed as partial shapes: input[1] = [100,32,32,32].

解释：
ValueError:无法将具有6553600个元素的张量重塑为形状[100,32,32,32]（3276800个元素）的形状[100,32,32,32]（3276800个元素），输入形状为[1001024,64]，[4]且输入张量计算为部分形状：input[1]=[100,32,32,32]。

解决方案：
将卷积核的个数变成64即可

x = tf.placeholder(tf.float32, [100, 3072])
x_image = tf.reshape(x, [-1, 32, 32, 3])
# x = tf.placeholder(dtype=tf.float32, shape=[None, 32, 32, 3])

y_ = tf.placeholder(dtype=tf.float32, shape=[None, 10])
# x_origin = tf.reshape(x, [-1, 28, 28, 1])

#具体含义是[卷积核的高度，卷积核的宽度，图像通道数，卷积核个数]，要求类型与参数input相同，有一个地方需要注意，第三维in_channels，就是参数input的第四维
W_e_conv1 = weight_variable([5, 5, 3, 64], "w_e_conv1")
b_e_conv1 = bias_variable([64], "b_e_conv1")
h_e_conv1 = tf.nn.relu(tf.add(conv2d(x_image, W_e_conv1), b_e_conv1))#(?,32,32,64)
conv1 = attention(h_e_conv1, 32, sn=False, scope='attention', reuse=False)
# print(conv1.get_shape())