Official annotations for Anticipation

[tqvinhcs]

Actually, I think it should be better to provide the "official annotations" for action anticipation along with action recognition task. For example, it is the 1s time stamp starting and ending of the anticipation segment.

Of course, we are able to modify the code for extracting these frame segments. However, It would be fairer if we have an "exact frame segments" provided by the dataset.