Drop a dimension of a tensor in Tensorflow

Question

I have a tensor that have shape (50, 100, 1, 512) and i want to reshape it or drop the third dimension so that the new tensor have shape (50, 100, 512).

I have tried tf.slice with tf.squeeze:

a = tf.slice(a, [50, 100, 1, 512], [50, 100, 1, 512])
b = tf.squeeze(a)

Everything seem working when i tried to print the shape of a and b but when i start training my model this error came

tensorflow.python.framework.errors_impl.InvalidArgumentError: Expected size[0] in [0, 0], but got 50
     [[Node: Slice = Slice[Index=DT_INT32, T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"](MaxPool_2, Slice/begin, Slice/size)]]

Are there any problem with my slice. How can i fix it. Thanks

Answer 1

Generally tf.squeeze will drop the dimensions.

a = tf.constant([[[1,2,3],[3,4,5]]])

The above tensor shape is [1,2,3]. After performing squeeze operation,

b = tf.squeeze(a)

Now, Tensor shape is [2,3]

Answer 2

There are multiple ways to do it. Tensorflow has started supporting indexing. Try

a = a[:,:,0,:]

OR

a = a[:,:,-1,:]

OR

a = tf.reshape(a,[50,100,512])

Answer 3

I use the tf.slice wrong in this case, it's should be like this:

a = tf.slice(a, [0, 0, 0, 0], [50, 100, 1, 512])
b = tf.squeeze(a)

You can find out why by look at the tf.slice documentation