I am trying to replicate the MSK+CNN model in this research paper: https://ieeexplore.ieee.org/document/9760210/authors#authors But I am not sure how to implement it correctly. It is suggesting that the first layer has 15 filters each with length of 100. And the second layer has 225 filters also with length of 100 each.
- What does it mean by saying the range of the kernels at lth layer? I am assuming it means the kernel size.
- I am having trouble to get the exact shape and length as in Fig.5 in this research paper by using kernel bank formulas. For example, my fi is from 1 to 15 and my yl is from 1 to 100. But when I ploted out the SK8 where fi = 8, I cannot get the straight line. Also, in the paper it suggests that the generated kernel should be half-wave and apply convolution operation to the input features to the convolution layer at the spatial position "k".
- Also how choose the value of fi in the second layer where the number of filters are 225?
So I try to get the SKli to match the shape of the weights shape for my first conv1 first layer. And then since I am using PyTorch, then I just layer.weight.copy_ in a initialize function. and then just model.conv1.apply(lambda layer: initialize(layer))
