Python – How to add skip joins between convolutional layers in Keras

I want to add a skip connection between the residual blocks of keras. This is my current implementation and it doesn’t work because tensors have different shapes.

The function is as follows:

def build_res_blocks(net, x_in, num_res_blocks, res_block, num_filters, res_block_expansion, kernel_size, scaling):
    net_next_in = net
    for i in range(num_res_blocks):
        net = res_block(net_next_in, num_filters, res_block_expansion, kernel_size, scaling)
    
# net tensor shape: (None, None, 32)
        # x_in tensor shape: (None, None, 3)
        # Error here, net_next_in should be in the shape of (None, None, 32) to be fed into next layer
        net_next_in = Add()([net, x_in]) 

return net

But I see

ValueError: Operands could not be broadcast together with shapes (None, None, 32) (None, None, 3)

How do I add or merge these tensors into the correct shape (None, None, 32)? If this is not the right approach, how can you achieve the desired result?

This is what res_block looks like:

def res_block(x_in, num_filters, expansion, kernel_size, scaling):
    x = Conv2D(num_filters * expansion, kernel_size, padding='same')(x_in)
    x = Activation('relu')(x)
    x = Conv2D(num_filters, kernel_size, padding='same')(x)
    x = Add()([x_in, x])
return x

You cannot add tensors of different shapes. You can concatenate them with keras.layers.Concatenate, but this will leave you with tensors of shape [None, None, 35].

Or, take a look
Resnet50 Implemented in Keras. Their residual block has 1x1xC convolution in the shortcut and is suitable for cases where the dimensions to be added are different.