Should I load optimizer state while doing progressive resizing?

With learn.save it saves the optimizer state too by default. With progressive resizing should I use this optimizer state or train the model with a new optimizer state from scratch when I increase the image size??