when you go to train or infer with a feature extractor, it is necessary to know the flow window for the optic flow of the flow generator going to be used
a checkpoint needs to store this information in the state_dict so that when a model checkpoint is loaded for training the feature extractor that the correct flow window, wouldn't matter for inference because a user would simply specify the flow window to use in inferring optic flow
when you go to train or infer with a feature extractor, it is necessary to know the flow window for the optic flow of the flow generator going to be used
a checkpoint needs to store this information in the state_dict so that when a model checkpoint is loaded for training the feature extractor that the correct flow window, wouldn't matter for inference because a user would simply specify the flow window to use in inferring optic flow