The answer is to train an object detection model using Azure AI Custom Vision, export it as a Docker container, and deploy it on an on-premises server with GPU. This approach satisfies all requirements because Custom Vision natively supports continuous learning defect detection by allowing you to retrain the model with new images of emerging defect types without rebuilding from scratch, while the exported container runs locally to achieve the sub-100ms latency needed for 30 fps real-time inference and minimizes data transfer over limited bandwidth. On the AI-102 exam, this question tests your ability to match Azure AI services to specific industrial constraints, with a common trap being to confuse Azure AI Video Indexer (designed for offline video analysis, not real-time streaming) or Spatial Analysis (for people tracking) with Custom Vision’s object detection capabilities. Remember the memory tip: “Custom Container for Continuous Control” — Custom Vision’s exportable container gives you low-latency, offline inference with incremental retraining, making it the only fully managed service that fits a bandwidth-constrained, real-time factory line.
AI-102 Implement computer vision solutions Practice Question
This AI-102 practice question tests your understanding of implement computer vision solutions. Match the stated requirement to the specific cloud service, access model, or configuration option — many options are valid in isolation but not for this scenario. After answering, compare your reasoning against the explanation and wrong-answer breakdown below. Once you have made your selection, read the full explanation to reinforce the concept and understand why each distractor is designed to mislead on exam day.
You are a senior AI engineer at a manufacturing company. The company has a production line that uses cameras to capture images of assembled products. The current system uses a set of rule-based heuristics to detect defects, but it has high false-positive rates. You have been tasked to design a new computer vision solution using Microsoft Azure AI services. The solution must:
- Detect defects such as scratches, dents, and misalignments in real-time as products move on the conveyor belt (frame rate of 30 fps).
- Support continuous learning: when a new defect type is discovered, the model should be updated without retraining the entire model from scratch.
- Operate with low latency (<100 ms per inference) to keep up with the production speed.
- Use only fully managed services (no custom containers or edge devices).
- The factory network has limited internet bandwidth, so the solution must minimize data transfer.
Which approach should you recommend?
Clue words in this question
Noticing these words before you look at the options changes how you read each choice.
Clue: "minimum / minimize"
Why it matters: Asks for the least resource use — fewest addresses, smallest subnet, lowest overhead. Eliminate over-provisioned options even if they would technically work.
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Train an object detection model using Azure AI Custom Vision, export it as a Docker container, and deploy it on an on-premises server with GPU
Azure AI Custom Vision with object detection can be exported and deployed to a Docker container on-premises for low latency. Continuous learning is supported by retraining with new images. However, Option C suggests using Azure AI Video Indexer, which is for video analysis and not real-time defect detection. Option A uses Azure AI Vision Spatial Analysis, which is for people tracking. Option B uses Azure AI Document Intelligence, which is for documents. Option D is the only one that meets all requirements: Custom Vision allows retraining, container deployment for low latency, and minimal bandwidth (only model updates).
Key principle: NAT direction and interface roles matter as much as the IP address mapping. Inside/outside designation controls which traffic is translated.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
✗
Use Azure AI Vision Spatial Analysis to detect defects in real-time by analyzing video feeds
Why it's wrong here
Spatial Analysis is for people and object tracking, not defect detection.
✗
Use Azure AI Video Indexer to index and search for defects in recorded videos
Why it's wrong here
Video Indexer is not real-time and adds latency.
✗
Use Azure AI Document Intelligence to analyze images of products
Why it's wrong here
Document Intelligence is for documents, not manufacturing defects.
✓
Train an object detection model using Azure AI Custom Vision, export it as a Docker container, and deploy it on an on-premises server with GPU
Why this is correct
Custom Vision supports export to container for low latency, and can be retrained incrementally.
Clue confirmation
The clue word "minimum / minimize" in the question point toward this answer.
Related concept
Static NAT maps one inside address to one outside address.
Common exam traps
Common exam trap: NAT rules depend on direction and matching traffic
NAT is not only about the public address. The inside/outside interface roles and the ACL or rule that matches traffic are just as important.
Detailed technical explanation
How to think about this question
NAT questions usually test address translation, overload/PAT behaviour, static mappings and whether the right traffic is being translated. Read the interface direction and address terms carefully.
KKey Concepts to Remember
Static NAT maps one inside address to one outside address.
PAT allows many inside hosts to share one public address using ports.
Inside local and inside global describe the private and translated addresses.
NAT ACLs identify traffic for translation, not always security filtering.
TExam Day Tips
→Identify inside and outside interfaces first.
→Check whether the scenario needs static NAT, dynamic NAT or PAT.
→Do not confuse NAT matching ACLs with normal packet-filtering intent.
Key takeaway
NAT direction and interface roles matter as much as the IP address mapping. Inside/outside designation controls which traffic is translated.
Real-world example
How this comes up in practice
A media company stores terabytes of video archives that are accessed once a year for audit purposes. Moving these objects to a cold storage tier (Azure Archive, S3 Glacier, or Google Nearline) costs a fraction of hot storage. Questions like this test whether you understand storage tiers, access frequency tradeoffs, and retrieval latency requirements.
What to study next
Got this wrong? Here's your next step.
Review the four NAT address types (inside local, inside global, outside local, outside global), PAT port overload, and static vs dynamic NAT use cases. Then practise related AI-102 NAT questions on configuration and troubleshooting.
Implement computer vision solutions — This question tests Implement computer vision solutions — Static NAT maps one inside address to one outside address..
What is the correct answer to this question?
The correct answer is: Train an object detection model using Azure AI Custom Vision, export it as a Docker container, and deploy it on an on-premises server with GPU — Azure AI Custom Vision with object detection can be exported and deployed to a Docker container on-premises for low latency. Continuous learning is supported by retraining with new images. However, Option C suggests using Azure AI Video Indexer, which is for video analysis and not real-time defect detection. Option A uses Azure AI Vision Spatial Analysis, which is for people tracking. Option B uses Azure AI Document Intelligence, which is for documents. Option D is the only one that meets all requirements: Custom Vision allows retraining, container deployment for low latency, and minimal bandwidth (only model updates).
What should I do if I get this AI-102 question wrong?
Review the four NAT address types (inside local, inside global, outside local, outside global), PAT port overload, and static vs dynamic NAT use cases. Then practise related AI-102 NAT questions on configuration and troubleshooting.
Are there clue words in this question I should notice?
Yes — watch for: "minimum / minimize". Asks for the least resource use — fewest addresses, smallest subnet, lowest overhead. Eliminate over-provisioned options even if they would technically work.
What is the key concept behind this question?
Static NAT maps one inside address to one outside address.
About these practice questions
Courseiva creates original exam-style practice questions with explanations and wrong-answer analysis. It does not publish real exam questions, exam dumps, or protected exam content. Learn why practice questions differ from exam dumps →
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. You need to design a computer vision solution that detects defects in manufactured parts on a conveyor belt. The solution must run in near real-time and adapt to new defect types without retraining from scratch. Which TWO approaches should you consider?
medium
A.Use Azure AI Face to detect anomalies
✓ B.Use Azure AI Custom Vision with object detection and retrain with new defect images
✓ C.Implement transfer learning with a pre-trained model and fine-tune on defect images
D.Use Azure AI Video Indexer to analyze video feeds
E.Use pre-built Azure AI Vision Image Analysis to classify images
Why B: Custom Vision with object detection can be retrained with new defect images. Transfer learning allows quick adaptation to new defect types.
Variation 2. You are a data scientist at a healthcare startup. You have deployed a custom object detection model using Azure Custom Vision to detect tumors in MRI scans. The model was trained on 10,000 labeled scans from a single hospital. After deployment, the model performs well on scans from that hospital but poorly on scans from a different hospital with a different MRI machine. The new hospital's scans have slightly different contrast and resolution. The model's precision drops from 0.92 to 0.65, and recall drops from 0.88 to 0.50. You have access to 500 labeled scans from the new hospital. You need to improve the model's performance on the new hospital's data as quickly as possible with minimal effort. What should you do?
hard
A.Collect more labeled scans from the new hospital and train a new model from scratch.
B.Create a new Custom Vision project and train only on the 500 new scans.
C.Apply image preprocessing to normalize the new hospital's scans to match the old hospital's style, then use the existing model.
✓ D.Use the existing model as a starting point and retrain it with the 500 labeled scans from the new hospital.
Why D: Option D is correct because Azure Custom Vision supports transfer learning, allowing you to take an existing trained model and retrain it with new labeled data. By using the 500 labeled scans from the new hospital as a training set, you can fine-tune the model to adapt to the different contrast and resolution characteristics without starting from scratch. This approach is the fastest and requires minimal effort, leveraging the previously learned features while incorporating domain-specific adjustments.
Last reviewed: Jun 20, 2026
Question Discussion
Share a tip, memory trick, or ask about the reasoning behind this question. Do not post real exam questions, leaked content, braindumps, or copyrighted exam material. Comments are moderated and may be removed without notice.
This AI-102 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-102 exam.
Question Discussion
Share a tip, memory trick, or ask about the reasoning behind this question. Do not post real exam questions, leaked content, braindumps, or copyrighted exam material. Comments are moderated and may be removed without notice.
Sign in to join the discussion.