Rebecca

Forum Replies Created

Viewing 15 posts - 1 through 15 (of 33 total)
  • Author
    Posts
  • in reply to: Wekinator Machine #643

    Rebecca
    Keymaster

    That is super cool, Ryan! Thanks for sharing! Are you happy for me to put this on the Wekinator examples list (with credit to you) next time I do a site update?

    Rebecca

  • in reply to: GPU vs CPU fo DTW #630

    Rebecca
    Keymaster

    Hi Arthur,

    Thanks for the post. Currently I doubt you would see any benefit from running Wekinator DTW on a GPU. Implementation of the FastDTW matching could indeed be parallelised, but it’s not currently done in Wekinator. One could certainly refactor the code to try to get some performance gains, but if speed is a big concern then I might recommend using an existing DTW/FastDTW library and building something from scratch that’s well tailored to the type of sensor(s)/gesture(s)/etc. that you’re using. Wekinator’s DTW implementation makes certain assumptions about gesture lengths in the training set and during runtime, and you can tweak these in the interface right now to try to improve recognition time and/or accuracy for particular types of gestures. But you could almost certainly do even better if you build in assumptions that are more appropriate to a particular problem.

    By the way, if you’re using Unity, you may be interested in the new InteractML project: http://interactml.com/ We’re building a set of IML tools directly into Unity, using a visual programming paradigm. I’m really excited about it!

    Rebecca

  • in reply to: WekiMini.jar not displaying on small raspi screen #627

    Rebecca
    Keymaster

    Hi Viktor,

    Yes, unfortunately the Wekinator main GUI is about 350 x 850. It would take substantial redesign to make it fit on such a small screen.

    One thing I’ve wanted to do for a while is make a “headless” option so that it can run without a GUI, through control entirely by OSC messages (and perhaps commandline). If I get some more development cycles that is near the top of my list, though I’d be happy for someone else to take this on.

    Rebecca

  • in reply to: Multiple models in one project #621

    Rebecca
    Keymaster

    Hi Kalien,

    It’s not possible to combine DTW with classification and/or regression in a single project, due to how different the processes of recording examples and running models are. However, if you’d like to run DTW at the same time as classification and/or regression, you can run two Wekinator projects simultaneously (you’ll just need to set one up to listen for incoming features on a different port, and ensure you’re sending features to that port).

    Best
    Rebecca

  • in reply to: Problem connecting Processing with Wekinator #619

    Rebecca
    Keymaster

    Hi there,

    Thanks for posting. First, you won’t want to use Mouse_ForDTW_2Inputs with a classifier or continuous model; it’ll only work with dynamic time warping.

    So the issue seems mainly to be what is going wrong with Simple_Mouse-DraggedObject_2Inputs. The code here is only very slightly different from Simple_MouseXY_2Inputs, so I’m pretty certain it’s a problem with how it’s being set up and run rather than this particular piece of code not working on your computer. The easiest way to troubleshoot from scratch will be to restart your computer (thus killing any mystery OSC processes potentially in the background), then only start up Simple_Mouse_DraggedObject_2Inputs and Wekinator. If you’re on Windows, make sure you enable network traffic for Simple_Mouse_DraggedObject_2Inputs if you get a pop-up box asking you; otherwise it won’t have the permission to send OSC. Make sure Wekinator is set to use 2 inputs. Then click on the running Simple_Mouse_DraggedObject_2Inputs program, keeping an eye on Wekinator, and move the box on the screen around. You should see Wekinator’s indicator light for OSC Inputs turn green. Alternatively, if you’ve got a small screen, hit Record in Wekinator, then move the green box around in Simple_Mouse_DraggedObject_2Inputs, then return to Wekinator, and you should see some examples have been recorded.

    Rebecca

  • in reply to: DTW project with multiple inputs/outputs #609

    Rebecca
    Keymaster

    Hi Joao,

    DTW allows for multiple inputs — e.g., you could recognise a gesture made using all 10 fingers. However, it doesn’t allow for multiple parallel outputs — i.e., recognising a set of 5 gestures made with 1 finger, a set of 3 other gestures made with another finger, etc. You are correct that you’d need multiple instances of Wekinator running to implement this sort of thing, and if you’re recognising the same gestures then you could just load the same model into each one. However, due to the inefficiency of DTW you might find this *very* slow (and this wouldn’t change even if you could use the same model in a single project for each finger), so I’d recommend you think about another approach if you find this doesn’t work as accurately or quickly as you need.

    Rebecca

  • in reply to: Wekinator general questions #608

    Rebecca
    Keymaster

    Hi Joao,

    Thanks for posting. Briefly, there’s just one main developer of Wekinator (me) and I occasionally get funds to support other development, but I don’t currently have any funding to add more features. So it’s just a matter of being able to hack on it a few times a year at the moment, and during these times I prioritise changes that most seriously impact a lot of students (Kadenze or in-person students) or people using the software quite seriously in professional work. I’m aware of the issues you’ve mentioned but there are frankly higher-priority things in the queue, so I can’t promise I will be able to implement them. Please feel free to add any of these as issues on the project GitHub page so they are noted (and other users can voice their support for them as well). The reality is that these are mainly the by-products of having one main developer, and trying to efficiently support a cross-platform software. I wish these quirks weren’t there, but they are. If anyone would like to fix them, please feel free to fork on github and we can have a chat about whether/how to merge back in!

    Best
    Rebecca

  • in reply to: Wekinator general questions #607

    Rebecca
    Keymaster

    Hi Joao,

    Thanks for posting. Briefly, there’s just one main developer of Wekinator (me) and I occasionally get funds to support other development, but I don’t currently have any funding to add more features. So it’s just a matter of being able to hack on it a few times a year at the moment, and during these times I prioritise changes that most seriously impact a lot of students (Kadenze or in-person students) or people using the software quite seriously in professional work. I’m aware of the issues you’ve mentioned but there are frankly higher-priority things in the queue, so I can’t promise I will be able to implement them. Please feel free to add any of these as issues on the project GitHub page so they are noted (and other users can voice their support for them as well). The reality is that these are mainly the by-products of having one main developer, and trying to efficiently support a cross-platform software. I wish these quirks weren’t there, but they are. If anyone would like to fix them, please feel free to fork on github and we can have a chat about whether/how to merge back in!

    Best
    Rebecca

  • in reply to: Complex Gestures with Wekinator and Kinect #533

    Rebecca
    Keymaster

    Yes, this is certainly possible. I recommend especially paying attention to the Kadenze lectures on features and capturing change over time.

  • in reply to: Recieve OSC input over IP #531

    Rebecca
    Keymaster

    Wekinator will be perfectly happy to receive OSC from outside networks (it can’t even tell where a message is coming from). You will need to ensure that Muse Monitor is sending to the right IP address, that your network isn’t blocking OSC messages, that Muse Monitor is sending to the right port (6448 by default), that Muse Monitor is sending the right message name (if not /wek/inputs, Wekinator’s default, then you’ll need to change the OSC message name in Wekinator), and that Muse Monitor is sending a single OSC message with all its features sent in the same message (the same number of features every time) as floats. (See http://www.wekinator.org/detailed-instructions/#1_Setting_up_communication_between_inputs_outputs_and_Wekinator for more info.) If Muse Monitor isn’t capable of doing this, you’ll need to use a separate program like OSCulator or something you’ve written yourself to translate Muse Monitor messages into these types of messages.

  • in reply to: FaceOSC & Wekinator on Mac #519

    Rebecca
    Keymaster

    Hi Pratyush,

    Thanks for posting. There is an issue with openFrameworks-generated apps on some versions of OS X, specifically having to do with OS X Gatekeeper. This may be causing the problem you’re seeing.

    One easy fix is to move the app to another directory (e.g., to the desktop) and then back. Then try double-clicking on it again. If it’s the Gatekeeper issue causing your problem, this should fix it.

  • in reply to: faceOSC ——->> wekinator #514

    Rebecca
    Keymaster

    Hi there,

    Regarding the problem with the pre-compiled executable, I suspect you are experiencing the same problem as here: http://www.wekinator.org/forums/topic/mac-binaries-failing-for-webcam-inputs/

    There is an issue with openFrameworks-generated apps on High Sierra, specifically having to do with OS X Gatekeeper.

    One easy fix is to move the app to another directory (e.g., to the desktop) and then back. Then try double-clicking on it again. If it’s the Gatekeeper issue causing your problem, this should fix it.

    Rebecca

    • This reply was modified 1 year, 8 months ago by  Rebecca.
  • in reply to: faceOSC ——->> wekinator #512

    Rebecca
    Keymaster

    Hi there,

    I’m also on Sierra and FaceOSC works OK for me, so it’s hard to say what is going wrong. I recommend the following:
    1. Make sure you’re running the executable (http://www.doc.gold.ac.uk/~mas01rf/WekinatorDownloads/wekinator_examples/executables/mac/inputs/VideoInput_FacialExpression_oF_14Inputs.zip) rather than from source, in case something has changed in the source on Gene’s github to make it incompatible
    2. Make sure FaceOSC executable draws funny white lines on your face. If it doesn’t, that means it isn’t finding your face. Ensure the data/ directory (which was in the same zip as the executable) remains in the same directory as the executable. (I.e., unzip the file from step 1 and don’t move the executable without the data directory). If you still don’t see white lines, try adjusting your lighting or removing your glasses.
    3. Restart your computer and launch only the FaceOSC executable and Wekinator, and ensure Wekinator is set up to receive 14 inputs. Restarting ensures that there isn’t some other program also receiving OSC.

    Hope this helps.

  • in reply to: export model as C++ file #511

    Rebecca
    Keymaster

    Hi there, this option isn’t yet in the official “release” builds of Wekinator, but you can get it by either (1) building Wekinator from the current source (https://github.com/fiebrink1/wekinator), or (2) using Michael Zbyszyński’s fork at https://github.com/mzed/wekimini (which has release versions you can download if you don’t want to build from source). Using either of these versions, every time you save a project, Wekinator will automatically save the C++ versions of the trained models within the project directory.

  • in reply to: Not receiving class probabilities from classifier #502

    Rebecca
    Keymaster

    Hi Sergio,

    I’m afraid I’m not able to duplicate this problem. Have you tried receiving these messages using this Processing sketch? http://www.doc.gold.ac.uk/~mas01rf/WekinatorDownloads/wekinator_examples/all_source_zips/Processing_SimpleDistribution_1Classifier.zip

    Rebecca

Viewing 15 posts - 1 through 15 (of 33 total)