Define the path to the pretrained weights from the PerceptualLoss

[Warvito]

In some systems, it will not be possible to access .cache/ or download the models on the fly (for example, FLIP platform and other federated tools). Having an argument like model_path:Optional[str]=None where you can optionally add a path to the model weights might be helpful.