Yujia-Yan
commented
1 year ago Hi,
This probability can be computed as long as there is a clear definition. In defining p(event), we need to know what this probability is normalized against, e.g., event v.s. nonevent for the entire duration of the event, or event v.s. all the competing candidates that are overlapping with the event. Actually we can have a dozen of definitions that may seem useful from some aspects.
The unnormalized score can be read directly from the score tensor.
is anyway to filter noise? for example: some low probability events.