Closed guanfuchen closed 6 years ago
I am facing the same question, have you figured it out? : )
Typically the model will perform best when the testing distribution reflects the training distribution, but there may be some degree of extrapolation if the testing samples share similar properties to training.
I want to use the model to predict other video, should this model be trained for every different video, or just use one trained model to predict other videos?