AI-102 Question 143
Select 3You are tasked with building a custom image classification model to identify different species of birds using Azure. You decide to use Azure Custom Vision for this purpose. After uploading a dataset of bird images and labeling them, you want to train the model. Which next steps should you take to ensure the model is properly trained and optimized for deployment?
- A
Configure the training parameters, such as the number of iterations and training domain, before starting the training process.
- B
Train the model using the provided dataset and evaluate its performance using precision and recall metrics.
- C
Deploy the model immediately after the first training run without evaluating the performance to save time.
- D
Use a separate validation dataset to test the model's accuracy and fine-tune it based on the results.
- E
Export the model in ONNX format directly after training without any further evaluation.
Show answer and explanation
Correct answers: A, B, D
Explanation
To successfully train and optimize a custom image classification model in Azure Custom Vision, you need to configure appropriate training parameters, evaluate the model's performance using precision and recall, and validate the model with a separate dataset. This ensures the model is accurate and reliable before deployment or export.
- A. Correct.
Correct: Configuring training parameters, such as iterations and the appropriate domain (e.g., General, Food, Retail), is essential for optimizing the model's performance.
- B. Correct.
Correct: Training the model and evaluating its performance using metrics like precision and recall ensures that the model meets the required accuracy standards.
- C. Incorrect.
Incorrect: Deploying the model immediately without evaluation is not recommended, as the model's performance might not meet the required standards or may need fine-tuning.
- D. Correct.
Correct: Using a validation dataset to test the model's accuracy is a best practice to ensure it performs well on unseen data. Fine-tuning based on validation results helps improve the model.
- E. Incorrect.
Incorrect: Exporting the model without evaluation can lead to suboptimal performance in real-world scenarios. Evaluation is a critical step before deployment or export.