如何将4D张量流重塑为2D

Question

I have an X_train image as:

X-train (37248, 32, 32, 3)

y_train (37248, 43)

I have a feed-dictionary as

train_feed_dict = {features: X_train, labels: train_labels}
with tf.Session() as session:
    session.run(init)
    session.run(loss, feed_dict=train_feed_dict)

My features is:

features = tf.placeholder(tf.float32, shape=[None, 32, 32, 3]) 
features = tf.reshape(features, [-1, 3072])

But when I run the code I get this error, reshaping does not seem to take place.

ValueError: Cannot feed value of shape (37248, 32, 32, 3) for Tensor 'Reshape_5:0', which has shape '(?, 3072)'

Answer 1

I believe the error was two prong:

I had to split the data into np.array() format and flat_features = tf.reshape(features, [-1, 3072])

having features=tf.reshape(features, [-1, 3072]), I had two problems so just changing the name t flat_features by itself did not work

so @drpng was also right in his comment