Custom Vision module exercise buggy code

[0-BSCode]

Describe the bug

In the Exercise - Upload the data portion of the module, one of the options provided to upload the images is to use Python and the Custom Vision SDK to upload and tag images. In the section discussing this, there is a repo that we may clone so we can download the code right away instead of coding the upload script from scratch. However, executing the downloaded code results in errors. Furthermore, there are a number of discrepancies between the downloaded code and the code provided in the instructions.

To Reproduce

Steps to reproduce the behavior:

1. Clone the repo mentioned in the instructions onto your local machine.
2. Navigate to the local repository and unzip the bird-photos zip file in the project directory.
3. Open the jupyter notebook file and replace the path provided in the os.chdir() function with the path to where your photos are stored.
4. Replace the ENDPOINT and training_key variables with the endpoint and training key provided by your Custom Vision resource.
5. Select the "Run All" cells option for the notebook.
6. Error is encountered (invalid arguments)

Expected behavior

Images should be uploaded onto your created Custom Vision project and the model can be trained, published, and tested using the downloaded script.

Fix

1. Update the CustomVisionTrainingClient function call in the "Create Custom Vision Project" section.

   # Provided code
   ENDPOINT = "<YourEndpointHere>"
   training_key = "<YourKeyHere>"
   trainer = CustomVisionTrainingClient(training_key, endpoint=ENDPOINT)

   # Functional code
   ENDPOINT = "<YourEndpointHere>"
   training_key = "<YourKeyHere>"
   credentials = ApiKeyCredentials(in_headers={"Training-key": training_key}) # Imported from msrest.authentication
   trainer = CustomVisionTrainingClient(ENDPOINT, credentials)

2. Update the create_images_from_files function call under the uploadImageList function in the "uploadImageList function to upload list of images" section.

   # Provided code
   upload_result = trainer.create_images_from_files(project.id, images=image_list)

   upload_result = trainer.create_images_from_files(project_id=project.id, batch=ImageFileCreateBatch(images=image_list)) # ImageFileCreateBatch is imported from azure.cognitiveservices.vision.customvision.training.models

3. Update the CustomVisionPredictionClient function call at the start of the "Test the published endpoint" section.

   # Provided code
   prediction_key = "<YourKeyHere>"
   endpoint = "https://<UpdateWithYourEndpoint>-prediction.cognitiveservices.azure.com/"
   base_image_url = "./"
   publish_iteration_name = "<YourIterationNameHere>"
   project_id = "<YourProjectIdHere>"
   predictor = CustomVisionPredictionClient(prediction_key, endpoint=endpoint)

   # Functional code
   prediction_key = "<YourKeyHere>"
   endpoint = "https://<UpdateWithYourEndpoint>-prediction.cognitiveservices.azure.com/"
   base_image_url = "./"
   publish_iteration_name = "<YourIterationNameHere>"
   project_id = "<YourProjectIdHere>"
   pCredentials = ApiKeyCredentials(in_headers={"Prediction-key": prediction_key})
   predictor = CustomVisionPredictionClient(endpoint, pCredentials)

Desktop (please complete the following information):

• OS: Windows 10
• Browser chrome
• Version 110.0.5481.104 (Official Build) (64-bit)

Additional context

Attached below are the details elaborating on how I discovered the issues and how I came about fixing them.

Under the "Create Custom Vision Project" cell, we run into an error saying we got multiple values for the 'endpoint' argument upon initializing CustomVisionTrainingClient.

Upon reviewing the instructions provided in the module, the following is the code that's supposed to be utilized, which is different from the downloaded code.

Updating the downloaded code to follow the code from the module results in the successful creation of a project.

However, another issue is encountered in the "Loop thru tags and call each function..." cell which comes from a faulty function call under the uploadImageList function.

Once again, another discrepancy is found between the function call from the downloaded code and the function call from the code provided in the instructions, the difference being the create_images_from_files function call.

However, changing this to the code provided by the instructions results in a different error.

According to this issue, the issue can be fixed by updating the create_images_from_files function call. Doing so stops the error from occurring.

---

Student, Student Ambassador

---

[BethanyJep]

Wow, great work on debugging, I will work to ensure this is implemented.