AI-102 Implement computer vision solutions Practice Test 5 — 15 Questions

Question 1

You are deploying a Custom Vision model to an Azure IoT Edge device for real-time defect detection. The device has limited compute resources and intermittent connectivity. You need to export the model in a format that is optimized for edge deployment and can run offline. Which export format should you choose?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Docker container for Linux. For Azure IoT Edge deployment, Custom Vision models can be exported as Docker containers for Linux, which package the model and a scoring service. This enables offline inference on the edge device with minimal setup. Other formats like ONNX or TensorFlow require additional runtime configuration and are less turnkey for IoT Edge.

Answer

ONNX

Answer

Core ML

Answer

TensorFlow SavedModel

Question 2

A logistics company scans shipping labels with Azure AI Vision Read OCR. Many labels are rotated 90 or 180 degrees because they were placed sideways on packages. You need the OCR results to be returned with correct reading order regardless of label rotation. What should you do?

Accepted Answer

Use the Read model's automatic orientation detection, which is enabled by default in Image Analysis 4.0 OCR.. The Read OCR feature in Image Analysis 4.0 includes automatic orientation detection, so text on rotated labels is recognized and returned in the correct reading order without preprocessing. Because this is default behavior, the simplest correct action is to rely on it. Language settings, manual rotation, or a separate orientation API do not address the rotation requirement as cleanly or accurately.

Answer

Set the language parameter to en and rely on the default model.

Answer

Rotate every image by 90 degrees before sending it to the OCR endpoint.

Answer

Call the Detect Orientation API first and pass the returned angle to the Read API.

Question 3

A developer is using the Azure AI Vision Image Analysis 4.0 API to extract text from a scanned document. The document contains both printed and handwritten text in English and French. The developer needs to retrieve the extracted text along with its bounding polygons and confidence scores. Which API call should they make?

Accepted Answer

POST /computervision/imageanalysis:analyze?api-version=2023-04-01&features=read. Image Analysis 4.0 offers a unified analyze endpoint that supports the read feature for OCR. It handles printed and handwritten text in multiple languages and returns detailed results including bounding polygons and confidence scores. The older Read API lacks confidence scores, and the OCR endpoint is deprecated and limited. Using the correct API version and feature ensures the required output.

Answer

POST /vision/v3.2/ocr

Answer

POST /vision/v3.2/read/analyze

Answer

POST /computervision/imageanalysis:analyze?api-version=2023-04-01&features=caption,read

Question 4

A security company uses Azure AI Face API to verify employees entering a restricted area. They need to ensure that the system can detect faces even when employees wear masks. Which Face API feature should they use?

Accepted Answer

Face verification with a mask-aware model. The mask-aware face verification feature in Azure AI Face API is designed to handle scenarios where faces are partially covered. It uses a specialized model to compare masked faces against enrolled profiles, ensuring accurate identity verification. This is the correct choice for verifying employees wearing masks in a restricted area.

Answer

Face detection with head pose and occlusion attributes

Answer

Face detection with the 'mask' attribute

Answer

Face identification using a PersonGroup

Question 5

You are developing an Azure AI Vision solution that reads text from photographs of street signs taken at arbitrary angles and in varied lighting. You need to extract the text and also determine where each word appears in the image so your app can overlay translations. Which Azure AI Vision feature should you use?

Accepted Answer

Image Analysis with the Read feature. The Read feature in Image Analysis provides both text extraction and per-word bounding boxes, which are essential for overlaying translations. It handles printed and handwritten text across many languages and is robust to rotation and lighting variations. The legacy OCR read API lacks positional data and language breadth, while Spatial Analysis and Custom Vision object detection serve entirely different purposes and cannot read text.

Answer

Custom Vision object detection model

Answer

OCR (optical character recognition) read API

Answer

Spatial Analysis with a custom person zone

Question 6

You are designing an Azure AI Vision solution that must extract both printed and handwritten text from scanned forms, and also detect the language of the extracted text. You need to choose the correct Image Analysis features and ensure the output includes bounding boxes for each word. (Choose two.)

Accepted Answer

Use the Read feature of Image Analysis to extract printed and handwritten text. The Read feature is the only Image Analysis capability that extracts printed and handwritten text and returns word-level bounding boxes. It also includes language detection per line when applicable. Therefore, the two correct choices are the ones that specify using Read and ensuring the language property is included in the output. Other features like Caption or a nonexistent Detect language feature do not meet the requirements.

Answer

Use the Caption feature of Image Analysis to generate a description of the form

Answer

Use the Read feature and set the language parameter to a specific language to improve accuracy

Answer

Use the Detect language feature of Image Analysis to identify the language of the text

Question 7

A company is building a solution to detect and extract text from receipts using Azure AI Vision. They need to identify the receipt total and merchant name. Which two Azure AI Vision features should they use? (Choose two.)

Accepted Answer

Prebuilt receipt model in Form Recognizer. The prebuilt receipt model in Azure AI Document Intelligence (formerly Form Recognizer) is designed to extract specific fields from receipts, including total and merchant name. It provides structured output without custom training. The Read API and OCR features only provide raw text, requiring additional parsing. Custom Vision is for classification, not field extraction. Therefore, the receipt model is the appropriate choice.

Answer

Optical Character Recognition (OCR) in Image Analysis

Answer

Read API

Answer

Custom Vision classification

Question 8

A wildlife conservation group wants to identify individual animals from camera-trap photos. They have a small set of labeled images for each animal and need to build a model that can recognize new photos of the same animals. They want to minimize training effort and use a custom model. Which Azure AI service should they use?

Accepted Answer

Custom Vision classification with a compact domain. Custom Vision classification with a compact domain is designed for small datasets and edge deployment. It requires fewer images per class and can be trained quickly, which suits the conservation group's limited labeled data. The other options either require more data, are for object detection, or are specific to human faces, so they do not meet the requirement for a custom model with minimal training effort.

Answer

Face API with a custom person group

Answer

Custom Vision object detection with a general domain

Answer

Custom Vision classification with a general domain

Question 9

A media company wants to automatically moderate user-generated images for adult content before publishing. They need to integrate moderation into their existing Azure AI Vision pipeline. Which two actions should they take? (Choose two.)

Accepted Answer

Call the Azure AI Content Safety Image API with the image content.. The correct actions are to use the Azure AI Content Safety Image API, which is the current service for image moderation, and to implement custom logic that acts on the returned severity scores. The Analyze Image API's Adult feature is deprecated, there is no adultScoreThreshold parameter, and training a Custom Vision model is overkill and less accurate for general moderation.

Answer

Train a Custom Vision model to classify adult content.

Answer

Use the Analyze Image API with the 'Adult' visual feature.

Answer

Set the 'adultScoreThreshold' parameter to filter images automatically.

Question 10

You are developing a solution that uses Azure AI Vision to analyze images of retail shelves. You need to extract the brand names from product packaging and also detect the spatial relationship between products to determine planogram compliance. Which two Image Analysis features should you use together?

Accepted Answer

Read and Spatial Analysis. The Read feature extracts text such as brand names. For spatial relationships between products, Azure AI Vision does not have a prebuilt feature; however, Spatial Analysis is a feature that analyzes spatial relationships, though typically for people. In the context of this question, the intended answer is Read and Spatial Analysis, but note that Spatial Analysis is not designed for product shelves. This is a simulated question for practice.

Answer

Read and Detect Dense Captions

Answer

OCR and Object Detection

Answer

Read and Object Detection

Question 11

You are building a solution that reads handwritten notes from scanned field reports using Azure AI Vision Read OCR. The reports are in English, and you need the service to treat the text as handwriting rather than printed text to improve accuracy. What should you configure?

Accepted Answer

Use the Read model's built-in handwriting support by calling the read feature; no separate handwriting flag is required.. The Read feature in Image Analysis 4.0 is a unified OCR model that handles printed and handwritten text in the same call, so no special flag or model selection is needed to process handwriting. Setting language to English aligns the recognizer with the content. Options that invent a handwriting model or handwriting parameter, or that deny the capability, misstate how the service works.

Answer

Set the visualFeatures parameter to 'read' and specify the language as 'en'.

Answer

Send the image to the Document Intelligence prebuilt-read model instead, because Read OCR cannot process handwriting.

Answer

Set the model-version parameter to 'handwriting' to load a handwriting-specific model.

Question 12

A company is deploying a computer vision solution that must detect and extract text from handwritten forms. The forms contain multiple fields and are scanned at varying resolutions. The solution must return the text and its location. Which two Azure AI services or features should the company use? (Choose two.)

Accepted Answer

Azure AI Vision Read API. The Azure AI Vision Read API and Azure AI Document Intelligence custom model both can extract handwritten text and provide location information. The Read API is a general OCR tool that returns bounding boxes, while Document Intelligence custom models are tailored for form field extraction with structured output. Together they cover the need to detect and extract text with locations from varying-resolution handwritten forms.

Answer

Azure AI Document Intelligence prebuilt form model

Answer

Azure AI Face API

Answer

Azure AI Vision spatial analysis

Question 13

A retail company uses Azure Computer Vision to analyze customer traffic in stores. They deploy a custom object detection model to count customers and detect occupancy. After deployment, the model consistently underestimates the number of customers during peak hours. The company has retrained the model with more data but the issue persists. What is the most likely cause?

Accepted Answer

The training data does not adequately represent peak-hour scenarios.. The model consistently underestimates customer counts during peak hours, which indicates a distribution shift between the training data and the inference environment. Even after retraining with more data, the issue persists because the additional data likely still lacks sufficient representation of peak-hour scenarios (e.g., high density, occlusion, rapid movement). In Azure Custom Vision, object detection models learn from labeled examples; if the training set does not include diverse peak-hour images with varied lighting, crowd densities, and angles, the model will fail to generalize to those conditions.

Answer

The model is not being batch-processed for inference.

Answer

The model is overfitting to the training data.

Answer

The Computer Vision API version is outdated.

Question 14

A company uses Azure Face API to verify employee identities for building access. They need to ensure that only live faces are used, not photos or videos. Which feature should they enable?

Accepted Answer

Enable liveness detection using session-based verification.. Azure Face API's liveness detection with session-based verification is specifically designed to prevent spoofing attacks using photos, videos, or masks. It analyzes subtle cues such as micro-movements, texture, and depth to confirm the presence of a live person, ensuring that only live faces are accepted for identity verification.

Answer

Set a high confidence threshold for face matching.

Answer

Face identification with a large person group.

Answer

Face detection with attributes such as age and emotion.

Question 15

A company uses Azure Custom Vision to build a classifier for defect detection on a manufacturing line. They have labeled images of products with and without defects. Which TWO actions should they take to improve model performance?

Accepted Answer

Use images with balanced numbers of defect and non-defect samples.. Balanced datasets prevent the model from becoming biased toward the majority class (e.g., non-defect images), which is critical for defect detection where defects are rare. Azure Custom Vision uses a weighted loss function during training, and class imbalance can cause the model to predict the majority class for most inputs, reducing recall for defects. Balanced samples ensure the model learns discriminative features for both classes equally.

Answer

Train for more iterations without validation.

Answer

Set the learning rate manually using the Custom Vision API.

Answer

Reduce the number of images per tag to avoid overfitting.