I have some input data that’s the same for every pixel of my input image, and I want the Conv2D layer to have access to it. Does it have to be in a channel, where that channel is the same value for every pixel? Or is there another approach?
Depending on the model, I would guess that you can pass the uniform input directly to your fully connected layer (after all the convolutions). It feels like it’s probably overkill to broadcast it out to the size of the entire input.
How you encode that input isn’t obvious, but you can start with just a Linear layer that connects the input value to the same fully connected layer that the convolutions connect to.