Vertex AI video action recognition - can it return action timeframes instead of a timestamps?

[shaharsol]

My problem is that my usecase requires the AI engine I use to provide predictions with the entire duration of the action. It seems to me that vertex AI picks a random frame in the span of the action and return it as the same start/end values. Here's an excerpt from an actual response

 

Can I make it work the way I need it to? Maybe I'm annotating in a wrong manner?

Here's a mockup of what I need. Notice how timeSegmentStart and timeSegmentEnd represent a duration now:

[ricconoel]

Hi,

Having the equal values for timeSegmentStart and timeSegmentEnd is an expected behavior as seen in this sample prediction response.

What I could suggest is to create a feature request in GCPs public issue tracker regarding your request. Please keep in mind that when you create the feature request, it still needs to be analyzed and considered by the product team and a definite ETA is not guaranteed. 

[shaharsol]

@ricconoel thanks for your reply. I agree that according to the sample this behavior is expected, however i wanted to know if it can be altered or configured. Looks weird to me that action recognition AI would stop at giving an arbitrary timestamp and not the whole duration of the action. Is there a way to find out if this feature is *already* implemented?

[ricconoel]

Unfortunately it cannot be altered or configured as of now. Hence my suggestion to create a feature request.