How to feed a convolutional neural network with images already split into three channels?

Greetings to all. I am newbie to this framework and therefore have no idea what to do. I have a set of images. Every three images in it are the same object, but in different channels. I want to build a classifier that receives similar images as input and then defines them as one object of a specific class. I don’t understand how to feed three separate images to the input so that the network understands that these are parts of one whole. If I just give her specially grouped images by me, and then concatenate the result, does it work?

Hey Saveman,
May I understand that your images are stored as 3 files, each of which storing the R / G / B channel ?
Are you planning to use an existing classifier architecture or make your own architecture ?
The answer will probably be concatenating your 3 tensors (r / g / b) together but just want to make sure depending on what you need / want to do :slight_smile:

Hi, Nicolas. Yes, one image is stored as three files, but not really RGB. These are images from a special instrument (imaging flow cytometer). Two images are Red and Green, and the third is the so-called brightfield image. As I understand it, an additional complication is that in all channels the images are grayscale. Let me explain. We tint the cage with special dyes and illuminate them with a laser. The device fixes the intensity of a particular color in different channels and the usual image in the optical range, but the pictures themselves remain gray. I will attach an example of how the image of one cell looks like this. Answering the second question - I would like to first use some modification of the standard architecture.