Custom Vision Prediction call error via Java SDK on Android studio - "Unable to Resolve Host"

Ashfak Ifham 1

Service: - Azure Custom Vision prediction service from android studio
Documentation link - https://learn.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?tabs=visual-studio&pivots=programming-language-java

Issue:
While using the java version of the SDK, and following the instructions on the documentation, Im not able to query a prediction service due to below error :
Unable to resolve host "findalgea-prediction.cognitiveservices.azure.com"
I have initiated the instances as required and the system has them active in the runtime. The attached images would provide further insight on to how I have called the prediction service, how the objects have been initiated and available at the runtime, and the subsequent error after it.

The prediction call is as done on the documentation and Im not able to zero in on the root cause of the matter. Please share some insights.

Attached images for your reference.

romungi-MSFT 43,696 Reputation points Microsoft Employee

2021-08-30T16:44:29.663+00:00

@Ashfak Ifham Did you get a chance to review below response and check with the suggested endpoint?

1 answer

romungi-MSFT 43,696 Reputation points Microsoft Employee

2021-08-27T10:37:38.353+00:00
@Ashfak Ifham I think the issue is you are trying to use the endpoint provided on Azure portal for a custom vision resource as the prediction endpoint. This is partly true because the key to your prediction endpoint request is the key of this resource. But, the endpoint is dependent on region in which your resource is created along with the project id and the iteration number which is actually published to this prediction resource.

A very simple way to get the prediction endpoint is to create it following the below convention for a image URL or a image file respectively:

https://<region_of_your_prediciton_resource>.api.cognitive.microsoft.com/customvision/v3.0/Prediction/<project_id>/classify/iterations/<published_iteration_name>/url https://<region_of_your_prediciton_resource>.api.cognitive.microsoft.com/customvision/v3.0/Prediction/<project_id>/classify/iterations/<published_iteration_name>/image

You can also lookup the prediction endpoint from Azure custom vision portal from the predictions tab for the published iteration:
Please sign in to rate this answer.
Ashfak Ifham 1 Reputation point

2021-08-31T04:29:06.707+00:00

Hi,

Thanks for the elaborated explanation. Let me check this right away and get back to you.

Best,
Ashfak

Ashfak Ifham 1 Reputation point

2021-08-31T14:28:37.16+00:00

Hi Team,

Thanks for your feedback on my query. I checked the given solution and noticed that the endpoint generated by both azure portal and the custom vision portal in my case are not with the location of the resource. I have attached 2 images for the concern. I have anyway created the resource in the EastUS location.

However, I manually edited the endpoint in the template that was shared by your team, as such :

https://eastus.api.cognitive.microsoft.com/customvision/v3.0/Prediction/aaead50a-95cd-403c-a3c2-9b774399e43b/classify/iterations/Iteration6/image

and attempted the query. I arrived at a resource not found error as attached. Im not sure why or where the root cause is.

Further more debugging and I located a null application variable from the microsoft.cognitive services classes as attached in the images. Im not too sure if this has anything to do with the error.

I also noticed that the expectation from the api maybe different since Im calling it through android studio , JAVA codes. But Im following the instructions on the documentations.

Is there any further insights on to the matter for me to resolve the concerned issue?

Thanks,
Ashfak

romungi-MSFT 43,696 Reputation points Microsoft Employee

2021-08-31T17:03:50.487+00:00

Thanks @Ashfak Ifham for the details. I have not used android studio before so I am unsure if there is any restriction blocking this call. If you are able to publish the iteration from this project then the predict call should work. As seen in the predict URL screenshot application name is not a required parameter for classify image so I am not sure if the application name is the issue. As per the Java documentation the application name seems to be the name of the class.

Apart from calling the predict API from this application are you able to use the REST API to classify an image, you could try to use the URL option of an image to check the prediction results with this endpoint.

Ashfak Ifham 1 Reputation point

2021-09-06T06:01:30.613+00:00

Hi,

In regards to the whole implementation of this custom vision project, the entire operation was done on the custom vision portal. Where I set up the model and training, etc. Upon publishing it via custom vision, I have used the endpoints to communicate from the android app for prediction service. So the only task that the app is trying to do from the android platform is the import of the project and the prediction. Im not sure if that is against a best practice instructions.

Im now trying to use the legacy method from REST API to check the service. Let me test it and get back to you.

Thanks,
Ashfak

Ashfak Ifham 1 Reputation point

2021-09-07T16:07:49.947+00:00

Hi,

I was able to test the webservice using REST and the endpoint worked out fine. Im not sure what was the cause of the failure. The Custom vision demo solution has a different way to use the API however with respect to the documentation.

So now Im proceeding to change the application to work with REST and proceed from there. Should be okay now.

Thanks for your assistance!

Best,
Ashfak
Sign in to comment

Share via

Custom Vision Prediction call error via Java SDK on Android studio - "Unable to Resolve Host"

1 answer