How to create video dataset without loading from the harddrive?

Hello!
I’m a beginner with pytorch but fairly good with python. I am creating a dataset where simple, short videos are catagorized with labels from 1 to 10.

I am trying to make a dataset object and have previously been using the built in LabeledVideoDataset class. However now I am using a dataset that I am not allowed to download because it is sensitive data. I can only scan the videos from my lab’s site without them being downloaded to my computer. I have data variables that I want to train my model on.

I am used to TensorFlow where you load the data and make it into a dataset from there. However for some reason it looks like every dataset loader i can find, you need to put in a path variable that tells you where to load the data from your file system. If I put the dataset on my hard drive I will be in legal trouble. If anybody could tell me how to do the tensorflow equivalent of model.fit(x,y) where the x and y are just variables, it would be much appreciated!

sorry for the beginner question

Tensorflow would also need to load the data eventually, so could you describe how it’s done?
PyTorch datasets often use lazy data loading, so storing only the path and allowing the actual sample to be loaded and processed in the __getitem__ once it’s needed will reduce the memory usage compared to preloading the entire dataset.