search

behinger / etcomp

Project to compare different eyetrackers
MIT License
34 stars 7 forks source link

Generating smooth pursuits from Pupil labs data #22

Closed Jovasa closed 5 years ago

Jovasa commented 5 years ago

Hi,

Based on your comment on the Pupil labs' Discord server it's possible to generate smooth pursuits out of the Pupil labs' recording data but could you give a quick rundown how to do it?

Thanks in advance!

behinger commented 5 years ago

hey! I guess you mean how to detect smooth pursuit?

I currently see two ways to do it: use the hmm algorithm https://gitlab.com/nslr/nslr-hmm I don't use HMM anymore because it gave me way too many smooth pursuit episodes. One could retrain it as well, but I do not have the labeled dataset. What worked for me though is to label everything that was labelled as smooth pursuit as fixation, for my purposes (e.g. EEG analysis) this would be good enough.

Use the deeplearning algorithm https://www.biorxiv.org/content/early/2018/06/29/359018, but you need to retrain it (in principle I think its not difficult, but I got mixed results so far and do not have time to do it properly)

In our experiment (this repo) we know when to expect smooth pursuit and so we use a very specific solution (a bayesian change point model)

Does that help? Best, Bene

Jovasa commented 5 years ago

Yes thanks a lot.