Resnet image input format


I am using a resnet50 model to classify facial expressions of the FER2013 dataset from kaggle.

The images are black&white. Is it preferable to load them as RGB images or as pure BW images with only one color channel?

In the examples I have seen, resnet models take RGB images. However, I tried with 1-channel BW images and the fitting proceeds as well (I only tried a few epochs so I don’t know what results I will achieve yet).

My question is, is there some reason to prefer RGB vs BW formats as input to pre-trained RESNET models?