Using on streaming video data

How can I use the pre-trained model on video data that is being streamed? Can live predictions about the actions being performed be made?
If not, how can I process the data of a saved video stream? (Suppose it is 8 hours worth of footage)?

Start by reviewing the repository's existing model inference entry points and video-input handling, since no specific file or test is named. Determine whether the pre-trained model supports live streams or long saved videos, and clarify the expected prediction workflow before defining what completion would mean.

Written by the indexing model from the issue text.