The problem about input shape of conv3d

[dmkr001]

I am trying to use a combination of siamese and 3dcnn, and I set the parameters like that:

def create_base_network():

model=Sequential()
model.add(Convolution3D(
    kernel_size=5,
    input_shape=(1,10,28,28),#(channels,batch_size,h,w)
    filters=32, 
    strides=1,  
    padding='same',  
    data_format='channels_first'
))

and input shape like that:

tr_pair0=tr_pair0.reshape(10840,1,28,28,10) tr_pair1=tr_pair1.reshape(10840,1,28,28,10) tr_y=tr_y.reshape(10840,10,)

input_shape = tr_pair0.shape print(input_shape) input_a = Input(shape=input_shape) input_b = Input(shape=input_shape) print(input_a)

because we re-use the same instance base_network,

the weights of the network

will be shared across the two branches

processed_a = base_network(input_a) processed_b = base_network(input_b)

but the problem is:

Traceback (most recent call last): File "/home/xzh/PycharmProjects/test/3DCNN.py", line 144, in processed_a = base_network(input_a)

ValueError: Dimension must be 6 but is 5 for 'sequential_1/conv3d_1/transpose' (op: 'Transpose') with input shapes: [?,10840,1,28,28,10], [5].

I really do not know what's the problem,can anyone help me, thank you very much.

[connection-on-fiber-bundles]

It's explicitly stated here https://github.com/keras-team/keras/blob/369854e5da3f9c58eedd9b77e778cf87755bd6a4/keras/engine/topology.py#L1415 that the shape argument of Input should not include the batch size.

So how about

input_a = Input(shape=input_shape[1:])
input_b = Input(shape=input_shape[1:])

instead?

[dmkr001]

yeah，you are right

	ffbcf 邮箱：xuzihua001@163.com

签名由 网易邮箱大师 定制

On 03/22/2018 08:12, connection-on-fiber-bundles wrote:

It's explicitly stated here https://github.com/keras-team/keras/blob/369854e5da3f9c58eedd9b77e778cf87755bd6a4/keras/engine/topology.py#L1415 that the shape argument of Input should not include the batch size.

So how about

input_a = Input(shape=input_shape[1:]) input_b = Input(shape=input_shape[1:])

instead?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.