search

dattalab / keypoint-moseq

https://keypoint-moseq.readthedocs.io
Other
70 stars 28 forks source link

Question: what if not all behaviors are present in all videos? #159

Closed SusanL82 closed 2 weeks ago

SusanL82 commented 3 weeks ago

Hello,

I'm looking forward to trying out keypoint-moseq on some of my behavioral data (exploration in an object recognition task, and in some open field settings). I have tracked my rats using a top-view DLC models that has no paws, but that does have shoulders and hips instead.

I have previously tried to use VAME, but ran into trouble with my analyses when not all behavioral clusters were present in a given video. E.g. some rats rear and explore the arena walls a lot and others do not, or just don't do it in one of the video sessions. Is this a problem for keypoint-moseq?

Or could I just prevent issues by making sure that there are some examples of everything in the training dataset?

Best, Susan

calebweinreb commented 3 weeks ago

Hi,

Can you elaborate on why that was a problem with VAME?

The way keypoint-moseq works is to essentially concatenate the data from all your recordings and then model them simultaneously. So if a behavior is not present in a given video, then the corresponding syllable won't appear in the output for that video.