Transfer learning-Preprocess Necessary?

I am trying to use Resnet 50 for CIFAR-10 classification. I was wondering that is the step below necessary? X_Cifar has dimensions [1,3,32,32] and I believe Resnet50 would only process [1,3,224,224]? However, I have seen implementation without the above step. Not sure what is the right way.

Step:
weights = ResNet50_Weights.DEFAULT
preprocess = weights.transforms()
X_resnet = preprocess(X_Cifar10)