Convolution 2D with feature selection as input

vision
1

I have an input of the form 1000* 5* 1* 50 where 1000 is the batch size with 5 features, having 45* 1 input.
I have an output label of 1000* 1* 2. This is a regression task.
Now, I would like to take individual features into account for example using only feature 1 or only feature 2 or combination of feature 1 and 2 and so on.
How may I set the convolutions to be?
Until now, I am using Conv1d with 1000* 1* 50 input feature wise and then concatenating wherever required with another Conv1d.

What should I choose for the kernal size and how to do this in one model? is there any option to choose which features I can select while using Conv2d?

2

I’m not sure I understand the question correctly.
If you want to use a specific input channel only, you could slice the input at the corresponding channel.
On the other hand you could use grouped convolutions to process each input channel individually.

Could you explain a bit more, what you mean by selecting the feature?