I want to use the pre-trained fasterrcnn_resnet50 model for my own custom object detector. My own images are greyscale and I have about 100+ images of those. Using the pre-trained network, I have to normalise the images by the mean and std of image net. However, with greyscale images, I am not sure whether I should use the same normalisation values of image net. Since all my channels for greyscale all have the same values?