A suitable application of OpenVibe?

[pwsnow]

Thanks Fabien!

I have tried to replicate the error but can't seem to replicate it. Now when I run the training scenario it just runs without completing or stopping due to the errors. I'll try to replicate the error today, I have been working on MatLab scripts to get around this so haven't looked at OpenVibe for a week.

Although off the top of my head it complianted about the k-fold cross-validation failing at the first test. Switching to SVM gave no errors but only 50% accuracy. The classification didn't work as a result due to I believe due to the 0.5 needing to be subtracted from the result due to the paper using LDA.

Below are my scenairos (taken from the report I posted a few posts ago), first the acquisition scenairo which is fairly straightforward:



Then the 2 class training scenairo (Using LDA as the classifier):



Finally the classification scenairo (This instance reading from a file):



Just realised that this version has one too many Simple DSP boxes, this was adding to correct the value by 0.5. The original didn't have this extra box!

I'll post some more details if I can replicate them.

Regards,

Peter

[pwsnow]

Ok, just a quick update.

I have managed to get everything working!

Although the online classification is very slow. I'm getting just over a second and a half to detect grasp and the same to detect release. I'm using the "stock" options for the classifier, any tips on how to get faster results in general?

I recently recieved a Myo EMG armband, the results are very good in terms of speed of classification! Although limited to 5 gestures, raw EMG will be available to access very soon.

I'm working on a VRPN plugin for Unreal Engine 4, at this point purely a client plugin which will allow UE4 to talk to OpenVibe via VRPN. I'll post some details when I have it working.

Peter