Question about model #4
Description
After conv4 (encoder) we have an output shape of XY with 256 channels.
But the input in fc1 is 4096, shouldn't the input be XY256 or can this model only work with input images of size 6464 (that's the one case when the input is 4096)