When fine-tuning a computer vision model, images must be preprocessed exactly as when the model was initially trained.