Concatenation of 2 different feature maps of different dimension

Hi all,

In CNN, 2 feature maps have different dimensions, one is 32x32x1024 and another one is 16x16x2048. If I want to merge them without any change in the feature map, what will be the procedure?

I'm using MATLAB 2020b.

Thanks in advance

Answer 1

[KALYAN ACHARJYA]

Have you refered the following links? (External Links)

https://ai.stackexchange.com/questions/2932/concatenate-convolution-layers-with-different-strides-in-tensorflow

https://stackoverflow.com/questions/64965116/keras-how-to-equalize-the-dimensions-of-2-layers-for-concatenation

If there is no other option left, you may have to choose from the following options -

• Pad the feature 1 vector with zero to make both x & y are equal.
• Make feature 2_modified=feature_2(1:2:end,1:2:end,:)
• Resize any one, so that both can be perfectly concatenate

Hope it helps

[Susama Bagchi]

Hi,

Thank you for your prompt reply. But I found that both the links are not for MATLAB.

In my case, I tried to upsample 16x16x2048 feature map. But to match the filter size if I use a convolution layer, it seems due to its weight the feature map is not the same. Is there any solution in this regard?

[KALYAN ACHARJYA]

Yes, these lines are not MATLAB, but you can get the idea, so that it can be implemented in MATLAB as well.