Azure Custom Vision Project Type and Domain Selection
Short answer
Project type: Object detection. Domain: General (compact).
Question
You have a factory that produces cardboard packaging for food products. The factory has intermittent internet connectivity.
The packages are required to include four samples of each product.
You need to build an Azure Custom Vision model that will identify defects in packaging and provide the location of the defects to an operator. The model must ensure that each package contains the four products.
Which project type and domain should you use? To answer, move the appropriate options to the correct targets. You may use each option once, more than once, or not at all. You may need to move the split bar between panes or scroll to view content.
NOTE: Each correct selection is worth one point.
Key Takeaways
The scenario tests the distinction between Image Classification and Object Detection, specifically identifying that spatial location data requires Object Detection, while offline constraints favor compact domains.
This question tests the correct configuration of Azure Custom Vision for defect detection and object counting in low-connectivity environments. The page establishes that Object Detection is required for locating defects, while General (compact) is optimal for intermittent connectivity.
Common mistake: Learners often select 'Image Classification' because they focus only on identifying defects, failing to recognize that the requirement to provide 'location' necessitates bounding box coordinates provided by Object Detection.
Expert Analysis
Why the Answer Is Correct
To satisfy the requirement of providing the location of defects, you must use an Object Detection project type, as it outputs bounding boxes around identified items. Furthermore, the factory has intermittent internet connectivity, which implies a need for edge deployment or offline inference; Azure Custom Vision's General (compact) domain is specifically optimized for smaller models suitable for such constrained environments. Therefore, selecting Object Detection and General (compact) ensures both functional accuracy and operational feasibility.
Why Other Candidate Answers Are Wrong
Image Classification cannot determine the specific location of defects within an image, only the overall category of the image content. Standard General domains produce larger models that may not perform well or fit into edge devices used in factories with limited resources compared to the compact variants.
Community Comment Notes
Community consensus strongly supports Object Detection due to the explicit need for spatial coordinates. As Srikantha noted, "Object detection: The model must identify specific items... and provide the location." Comments highlight that classification fails the location requirement entirely.
Exam Strategy
Always map specific output requirements (like 'location' or 'count') to the correct project type. When infrastructure constraints like 'offline' or 'low bandwidth' are mentioned, prioritize compact or optimized model domains over standard ones.
Frequently Asked Questions
Why can't I use Image Classification for defect location?
When should I choose General (compact) over General?
Discussion 1 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.
From public community threads 1
-
Srikantha 👍 1Project type: Object detection Domain: General (compact) Key Reasons Object detection: The model must identify specific items (verifying that exactly four products are inside each package) and provide the location (bounding box coordinates) of defects to an operator. Image classification can only categorize an entire image as a whole and cannot return spatial locations or count individual items. General (compact): The factory has intermittent internet connectivity, meaning the model must be exported to run locally on an edge device or container offline. In Azure Custom Vision, only domains designated as (compact) support model export (ONNX, TensorFlow, Docker, etc.).
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.
Related Analysis
Practice All AI-103 Questions
Access 88 questions with complete answers and detailed explanations.
View Full AI-103 Practice Test →