Log in or register to post comments

MT360

May 15, 2019 - 2:42am #1

Hello,

The model training is happening in Cloud or internally ? If cloud means, is model accessible ? We do not want to put our model in cloud. So is there any other way to train the model ? or if there any other method to train the model ? 

 

 

 

Thanks

MT360

June 16, 2019 - 11:19pm #10

Hello, 

Thank you for your reply. 

I will try those methods.  

 

Thanks 

MT360

June 13, 2019 - 8:27am #9

Hello,

One approach is to:

  • Store the latest reported pose (which is is world coordinates)
  • Turn the ObjectTracker off and the use the old pose to display the object

Or, in Unity, you can detach the augmentation from its parent trackable with Transform.setParent(null, true);

Thanks,

Vuforia Engine Support

MT360

June 13, 2019 - 12:05am #8

Hello,

We are facing nearly 3 to 4 cm offset after tracking. Means, 3D model and physical model is not matching properly. I could not able to rectify this issue. Is there any way to stop tracking and freeze the model in world space after tracking. So that it will not move if we move closer to object also. This is our aim to achieve. 

 

Thanks

 

MT360

June 11, 2019 - 9:11am #7

Hello,

Yes, MT360 is training-based so uploading to Vuforia cloud services is required to create a dataset.

Can you please explain more what you mean by "100% accuracy"?

For improving the registration accuracy of object overlay augmentations (when using Model Targets on HoloLens), it is recommended to use objects larger than “toys” (like the KTM bike or Viking Lander). Even in the current Unity Model Targets on HoloLens sample, the Viking Lander augmentation that is meant to directly overlay upon the physical target may not have exact registeration.

There are a variety of factors that contribute to this offset, including user calibration and performance differences between HoloLens units, but as the objects become larger the registration accuracy should also improve.

Thanks,

Vuforia Engine Support

MT360

June 10, 2019 - 11:39pm #6

Can you please help us to solve this issue ? 

 

Thanks in advance

MT360

May 21, 2019 - 9:14pm #5

Thank You for your response. 

MT360 also Training Based ? or we can do like normal model target ? Because we are struggling to achieve 100% accuracy on HoloLens.

Can you please suggest some solution to achieve this? Can we able to stop tracking or fix the model in static position , so that the model won't move if we move nearer to the physical object. Please help us to achieve this.

 

 

Thanks 

MT360

May 16, 2019 - 11:05am #4

Hello,

The product team is aware of similar requirements. At this time there is no roadmap for supporting Model Targets training on desktop or on a private cloud.

Unfortunately, there currently is no way to train Model Targets without the 3D model being uploaded to Vuforia's cloud.

Thanks,

Vuforia Engine Support

MT360

May 15, 2019 - 9:02pm #3

Hello,

Thank You for your reply.

Can we use our own cloud to train the models ? . Or can we expect that feature in Vuforia RoadMap ?

Since we do not want to publish our data to the cloud, is there any way to achieve this ? 

 

Thanks

MT360

May 15, 2019 - 3:38pm #2

Hello,

Model Target training is done on the cloud, so your 3D model will be uploaded to the cloud in support of that.

Untrained (or traditional) Model Targets are created on the local machine. No data is uploaded to the Cloud in this case.

Thanks,

Vuforia Engine Support

Log in or register to post comments