HoloLens 2 -Azure Custom Vision

dluong 10

I am working the following tutorial https://video2.skills-academy.com/en-us/training/modules/azure-cloud-services-tutorials/ Unfortunately, I couldn't upload the images captured by Hololens 2 to Azure Custom Vison sever.

In order to detail our problem, I assume that I created a Custom Vision resource as "ExampleCustomVision" with the both options of training and prediction, therefore, Azure server automatically generated 2 resources as "ExampleCustomVision" for training and "ExampleCustomVision-Prediction" for prediction. Then, I created a classification project as "ExampleProject" assigned to ExampleCustomVision resource.

Following the tutorial, here are credentials set for ObjectDetectionManage to take and upload images before deplying app to Hololens 2:

Azure setting

Azure Resource Subscription ID: My-Azure-account-ID
Azure Resource Group Name: My-Group-Name
Cognitive Service Resource Name: ExampleCustomVision
Resource Base End Point: End-Point-of-ExampleCustomVision

Project setting

API Key: Key-of-ExampleCustomVision
Project ID: ExampleProject-ID

I found the tag, which was set before taking the images, in my Azure storage and ExampleProject, but I couldn't train the model and all images were not uploaded to ExampleProject.

User's image

Could you help me fix this issue?

romungi-MSFT 45,961 Reputation points Microsoft Employee

2023-08-14T07:58:59.4533333+00:00

@dluong I am not sure if the tutorial that is used is updated to keep up with the limit of number of images or tags that are required for a custom vision project. But from the custom vision project perspective for object detection, the minimum number of images that are required for a tag to train the model is 15. See the limits page of custom vision here. The recommendation is lot more to train the model with good accuracy though.

Based on what you have seen the ObjectDetectionManager script that is mentioned in the tutorial is used to upload the images that are captured to the project based on the credentials you have used. I think you are able to see some images but not all of them in the custom vision project after you run this script. If this is the case, you can rule out the issue with credentials or project with the script. However, the number of images to be uploaded to the project might be something that needs to be sorted out. I do not have access to HoloLens or this script but if you can try to modify the script to upload atleast 15 images per tag and call the train API then I think you should be successful to run this step from the tutorial. I hope this helps!!
Matthew O'Leary 20 Reputation points

2023-09-06T23:38:10.2666667+00:00

Were you able to get this tutorial to work? I am having the same issues as you with the custom vision project/resources.
dluong 10 Reputation points

2023-09-08T07:28:59.7066667+00:00

The tutorial appears to be outdated as it instructs capturing only 6 pictures for training the Custom Visions model, whereas a minimum of 15 pictures is now required.

Share via

HoloLens 2 -Azure Custom Vision

Your answer