The Mixed Reality Forums here are no longer being used or maintained.
There are a few other places we would like to direct you to for support, both from Microsoft and from the community.
The first way we want to connect with you is our mixed reality developer program, which you can sign up for at https://aka.ms/IWantMR.
For technical questions, please use Stack Overflow, and tag your questions using either hololens or windows-mixed-reality.
If you want to join in discussions, please do so in the HoloDevelopers Slack, which you can join by going to https://aka.ms/holodevelopers, or in our Microsoft Tech Communities forums at https://techcommunity.microsoft.com/t5/mixed-reality/ct-p/MicrosoftMixedReality.
And always feel free to hit us up on Twitter @MxdRealityDev.
UWP10/HoloLens/.NET Core Compatible Image Recognition / Machine Learning Libraries
Hi,
Do any developers have recommendations for compatible libraries for image and scene recognition? E.g. a library that has worked for image classification and is compatible with HoloLens? I'm thinking classifying spatial data along with photos might help approve accuracy of meshes in scenarios where you are interested in acute details.
https://enoxsoftware.com/opencvforunity/ - I'm considering this one in Unity App Store that wraps opencv 2.4, hoping some one has already found it works.
Also looked at Accord.NET, which some modules might not want to use due to GPL: https://www.nuget.org/packages?q=portable.accord
Much appreciated
Best Answer
-
Options
https://enoxsoftware.com/opencvforunity/ - I'm considering this one in Unity App Store that wraps opencv 2.4, hoping some one has already found it works.
I loaded that on the hololens and it does work, he includes a UWP package of plugins for win10. It does suck performance.... I loaded some of these classifiers using it and was able to detect faces, just performance was painful https://github.com/Itseez/opencv/tree/master/data/haarcascades
I even loaded his AR example on hololens but the ar content seemed to drift more than just pinning things on the hololens.
If you don't show the live camera feed on the hololens, but instead just use the tracked data it returns, performance probably wouldn't be nearly as bad.
6
Answers
If I were doing it, I'd use a cloud based solution, such as Azure. I guess it depends on your app, but machine learning may be a little heavy handed for the for Hololens to handle without burning through the battery.
Yes Asp.net Azure Image recognition is at benfoster.io High performance image processing with Image Resizer and Azure
Thanks for suggestion. I have a scenario where trying to monitor the user's environment continuously, since they will actively be modifying it, so that would lead to hundreds of requests, since closer to monitoring video stills/deltas than an occasional image...but guess will see pricing.
See if could get spatial mapping/hand motion trigger to limit number of recognition requests... The initial training process of creating recognizer, I'd do off of the HoloLens (maybe smaller set of initial calibration in the field error correction user feedback or sampling), but to your point I don't have the background to know true run time cost, since probably GPU heavy.
Thanks, not quite what looking for, focusing on real time recognition, but I think this could be useful for other applications i have in the future.
@BotGreet Not sure if this helps, but both Microsoft and Google have open sourced their machine learning tookits. May be worth a look.
http://www.dmtk.io/index.html
https://www.tensorflow.org/
I loaded that on the hololens and it does work, he includes a UWP package of plugins for win10. It does suck performance.... I loaded some of these classifiers using it and was able to detect faces, just performance was painful https://github.com/Itseez/opencv/tree/master/data/haarcascades
I even loaded his AR example on hololens but the ar content seemed to drift more than just pinning things on the hololens.
If you don't show the live camera feed on the hololens, but instead just use the tracked data it returns, performance probably wouldn't be nearly as bad.
PCT Vuforia has a nice API for recognizing image targets and showing 3D objects and virtual buttons. They have a HoloLens early access program, but unfortunately it's now closed. You can try contacting their dev relations, YMMV
Anyone tried TensorFlow so far?