Oxford Vision Library for Video RRS feed

  • Question

  • Hi,

    I'm using EmguCV on top of simple video recorded of Kinect stream.

    Is there best practicies to use Oxford Vision Library on Video without spaming the server ?

    - may be good offline algorithm,  before querying the server ?
    - or uploading the video on Azure then running a Oxford vision code ? (what about security, bandwidth, ...)

    Saturday, June 27, 2015 1:00 PM

All replies

  • Hi JP,

    One approach you can take today is to capture still images periodically and send those to the Computer Vision API.

    The Computer Vision API is throttled at 20 calls per minute, so you can submit a still image every 3 seconds to the API. Another approach that I've seen used is to send images less frequently than that (say one every 10 seconds) and then after a certain state change is detected to capture images more frequently.

    One additional tip is that you can scale down the size of the image before sending over the wire, as that may improve performance.

    Thank you for the info on your setup, I'll take this back to the team as we prioritize our next set of updates.

    Tuesday, June 30, 2015 8:06 PM
  • Hi,

    Yes I was thinking of a smart algorithm to detect
    - state change 
    - find a not blurry image

    The problem is old Computer Vision (EmguCV) works well to detect faces and sent it to Microsoft but Oxford will tell me yes it a face.

    If I take an other example:
    - I put a camera in the kitchen
    - Motion detection record short film of few seconds while cooking
    - If I want to undestand what is cooked

    => I'll have to magically find good moment to send screenshots
    => I assume you will have Food 101 later

    So hard to do magic :-)
    The other big question is privacy, I send picture on an US cloud that's a little bit "hard" for an Europeen company (but it's a subject for out of beta)

    Wednesday, July 1, 2015 8:35 AM