search

PsychoinformaticsLab / pliers

Automated feature extraction in Python
https://pliers.readthedocs.io/en/latest/
BSD 3-Clause "New" or "Revised" License
294 stars 68 forks source link

Add support for Microsoft Video Indexer and/or Google object tracking #329

Open tyarkoni opened 5 years ago

tyarkoni commented 5 years ago

Tracking objects/people in videos is a high priority. Of the major available services, AWS Rekognition requires upload to an S3 bucket, so we can defer that. Azure's Video Indexer looks like it should be fairly easy to add. Google now has a beta Object Tracking mode in Video Intelligence. @qmac, is there a specific reason you never implemented Object Tracking, or was it just not available at the time and/or you didn't get to it before you left? If the latter, I'll probably add it myself, but just wanted to make sure you didn't already look into it and decide it was unworkable for some reason.

[Edit: Video Indexer is more complex than I initially thought.]

qmac commented 5 years ago

Yeah all I remember was it being complicated to do. IIRC for the Microsoft Face tracking, you had to get the IDs for an original set of faces and then query for those in subsequent requests. Not sure if all the tracking APIs operate that way, its all kind of hazy at this point.