How to set the number of output channels for a tensor？

[SkyWxxk]

Firstly, thanks to authors' KAN-CNN works. But I was confused that the output channel number of the tensor was directly multiplied by the n_convs parameter, rather than the expected output channel number. Here is the demo:

import torch
from torch import nn
import torch.nn.functional as F
from kan_convolutional.KANConv import KAN_Convolutional_Layer

class KANC_MLP(nn.Module):
    def __init__(self,device: str = 'cpu'):
        super().__init__()
        self.conv1 = KAN_Convolutional_Layer(
            n_convs = 3,
            kernel_size= (3,3),
            padding =(1,1),
            device = device
        )
        self.conv2 = KAN_Convolutional_Layer(
            n_convs = 5,
            kernel_size = (3,3),
            padding =(1,1),
            device = device
        )
        self.pool1 = nn.MaxPool2d(
            kernel_size=(2, 2)
        )
        self.flat = nn.Flatten() 
        self.linear1 = nn.Linear(3675, 256)
        self.linear2 = nn.Linear(256, 10)

    def forward(self, x):
        x = self.conv1(x)
        print(f'After conv1: {x.size()}')  # Print size after conv1
        x = self.pool1(x)
        print(f'After pool1: {x.size()}')  # Print size after pool1
        x = self.conv2(x)
        print(f'After conv2: {x.size()}')  # Print size after conv2
        x = self.pool1(x)
        print(f'After pool2: {x.size()}')  # Print size after pool2
        return x

def main():
    model = KANC_MLP(device="cpu")
    x = torch.randn(1, 3, 32, 32)
    output = model(x)
if __name__ == "__main__":
    main()

The result shows here:

After conv1: torch.Size([1, 9, 32, 32]) After pool1: torch.Size([1, 9, 16, 16]) After conv2: torch.Size([1, 45, 16, 16]) After pool2: torch.Size([1, 45, 8, 8])

We can see that the channel was just multiplied by "n_convs". How can we get the desired output channel? Thanks!

[AlexBodner]

Hello, the responsible for it here. In our implementation, you especify the ammount of convolutions that you want in a layer, so it applies each conv to each entering channel, resulting in a channels * convs output. Hope that this clarifies it. Regards, Alex