I have a tensor of [20, 3, 32, 32]
I want to increase the 3 to 64 ([20,64,32,32], where 20 is the batch size). I tried the repeat function. But that only gives me 63 or 66, because with repeat you can only tile (multiply) indices.
How can I solve this?
How would you like to increase the channel dimension?
As you’ve explained,
repeat won’t be possible.
If you want to increase is in a “trainable” way, you could use a conv layer with a kernel size of