AI-103 Computer Vision — 30 Questions

Question 1

You are training a Custom Vision object detection model to identify specific manufacturing defects on a production line. You have uploaded 50 images and tagged the defects. When you attempt to train the model, the training fails. What is the most likely cause?

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

Accepted Answer

Object detection requires a minimum of 15 tagged images per tag.. Custom Vision object detection requires a minimum of 15 tagged images per tag before training can succeed. With only 50 images total and presumably fewer than 15 per defect tag, the training job fails validation. This is a documented hard requirement for object detection projects, distinct from classification which needs fewer images per tag.

Answer

You must have at least 5 tags defined in the project.

Answer

You must use multi-class classification instead of object detection for manufacturing defects.

Answer

Training images must be smaller than 256x256 pixels.

Question 2

You are building a content moderation pipeline. You need to detect if user-uploaded images contain adult content, racy content, or gore using Azure AI Vision. Which feature should you include?

Accepted Answer

features=adult. The Azure AI Vision Analyze Image API supports a features parameter, and 'adult' is the correct value to detect adult, racy, and gory content. This feature returns scores for adult content, racy content, and gore, enabling moderation decisions.

Answer

features=safety

Answer

features=moderation

Answer

features=image-analysis safety check

Question 3

You are writing a Python script using the Azure AI Vision SDK to analyze an image from a local file path. Which client class should you instantiate first?

Accepted Answer

ImageAnalysisClient. The Azure AI Vision SDK (version 4.x, the current Image Analysis API) exposes the ImageAnalysisClient class as the entry point for analyzing images. You instantiate it with an endpoint and credential, then call methods like AnalyzeImageFromUrl or AnalyzeImageFromFile. This class replaces the older ComputerVisionClient used by the legacy Computer Vision SDK.

Answer

ComputerVisionClient

Answer

VisionServiceClient

Answer

CustomVisionPredictionClient

Question 4

You need to retrieve coordinates of people detected in an image using Azure AI Vision 4.0. Which visual feature should you request?

Accepted Answer

features=people. In Azure AI Vision 4.0, the 'people' visual feature is specifically designed to detect people in an image and return their bounding box coordinates. The 'objects' feature detects generic objects (including some people as objects), but 'people' is the dedicated feature that returns person-specific detection with coordinates. This makes features=people the correct choice for retrieving coordinates of detected people.

Answer

features=objects

Answer

features=faces

Answer

features=tags

Question 5

You are testing a Custom Vision object detection model using the test image feature in the portal. You notice that objects are detected with low confidence scores. You want to improve the model's performance without adding more training images. What can you adjust in the training settings?

Accepted Answer

Increase the training time budget in advanced training settings.. In Custom Vision, the training time budget is a settable parameter in advanced training settings that controls how long the service spends optimizing the model. Increasing it allows more iterations and hyperparameter exploration, which can improve accuracy and confidence scores without adding images. This is the only listed option that directly affects model training quality.

Answer

Switch from object detection to multiclass classification.

Answer

Decrease the image resolution setting in project properties.

Answer

Disable data augmentation.

Question 6

Your organization requires that all Azure AI Vision API calls are audited and that request payloads (including images) are logged for compliance investigations. Which Azure service feature should you configure alongside Azure AI Services?

Accepted Answer

Enable Diagnostic Settings on the Azure AI Services resource to stream audit logs and telemetry to Azure Monitor Log Analytics.. Diagnostic Settings on an Azure AI Services resource allow you to stream audit logs and telemetry to destinations such as Azure Monitor Log Analytics, Storage, or Event Hubs. Enabling Diagnostic Settings and routing to Log Analytics captures request metadata and, depending on the category, can include request payloads for compliance auditing and investigation.

Answer

Configure Custom Vision training iterations export.

Answer

Enable Azure Bastion session recording.

Answer

Configure Azure Traffic Manager with logging rules.

Question 7

You are examining the metadata of an image analysis result from Azure AI Vision 4.0. Which property contains the dimensions (width and height) of the analyzed image?

Accepted Answer

metadata. In Azure AI Vision 4.0 image analysis results, the 'metadata' property contains the image's width and height dimensions. This metadata object is returned alongside the analysis results and provides basic image properties. The 'dimensions' property does not exist as a top-level field; the correct location is within 'metadata'.

Answer

dimensions

Answer

modelVersion

Answer

requestId

Question 8

You are deploying a Custom Vision model as a Docker container. You need to verify that the container is running and healthy before routing production traffic to it. Which health check endpoint should your load balancer ping?

Accepted Answer

/ready. Cognitive Services containers provide a standard liveness/readiness health check endpoint at `/live` or `/ready` depending on container configuration.

Answer

/status

Answer

/ping

Answer

/health

Question 9

You are developing a script to query a Custom Vision prediction endpoint. Which HTTP header must be included to authenticate your request using an API key?

Accepted Answer

Prediction-Key. The Custom Vision prediction endpoint requires the API key to be passed in the Prediction-Key HTTP header. This header authenticates the request when calling the prediction URL. Other headers such as Authorization: Bearer are used for token-based auth, not the prediction key.

Answer

Authorization: Bearer

Answer

X-Auth-Token

Answer

Vision-Api-Key

Question 10

You are developing a .NET application that interacts with Azure AI Vision Read API. Which TWO objects or properties would you inspect in the asynchronous response to retrieve extracted text lines and their confidence scores? (Choose two.)

Accepted Answer

text. Option E, lines, is correct because the Read API's asynchronous analyze result exposes a readResults array whose pages contain a lines collection, and each line object holds the recognized text along with bounding box coordinates and per-line confidence via its words. Option A, text, is correct because each line (and each word) in the Read response includes a text property containing the actual extracted string content, which is what you inspect to retrieve the recognized characters. Options B, C, and D are incorrect because dominantColors, adultScore, and smartCrops belong to other Azure AI Vision features (image color analysis, content moderation/adult classification, and thumbnail smart cropping respectively), not to the Read OCR result schema, so they contain no extracted text lines or confidence scores.

Answer

dominantColors

Answer

adultScore

Answer

smartCrops

Question 11

When designing an automated document processing pipeline using Azure AI Vision Read API, which THREE architectural considerations should you account for? (Choose three.)

Accepted Answer

Ensure document files adhere to supported page count and file size limits.. Option A is correct because the Azure AI Vision Read API enforces documented input constraints — for example, PDF/TIFF files are limited to 2,000 pages and files must be under 500 MB (and 50 MB for the synchronous Read call) — so a pipeline must validate page count and file size before submission to avoid rejected requests. Option C is correct because the Read API's initial POST returns HTTP 202 with an Operation-Location header pointing to the result endpoint; the pipeline must poll that URL until the operation status is 'succeeded' before retrieving the extracted text. Option E is correct because the service returns HTTP 429 when request rates exceed the tier's transactions-per-second quota, and Microsoft recommends retrying with exponential backoff (honoring the Retry-After header) to avoid throttling failures. Option B is not correct because the Read API is asynchronous for multi-page and large documents, returning 202 Accepted rather than a synchronous JSON payload with all text on the initial POST. Option D is not correct because extracted results are not stored permanently in local memory; results are retained temporarily on the service side and must be retrieved via the operation result endpoint, with pagination handled through the 'nextLink' value when results span multiple pages.

Answer

Expect synchronous JSON responses containing all extracted text directly on the initial POST call.

Answer

Configure the API to store extracted text permanently in local memory without pagination.

Question 12

You are integrating Azure AI Vision into a serverless Azure Functions architecture. A batch of 500 high-resolution images needs to be processed. To prevent function timeouts and optimize throughput, how should you architect the function triggers?

Accepted Answer

Use a Blob Storage trigger to enqueue image messages onto an Azure Storage Queue, and configure Queue-triggered functions to process image analysis asynchronously.. Decoupling ingestion from processing with a Blob Storage trigger that enqueues messages and Queue-triggered functions enables asynchronous, parallel processing of 500 images. This prevents the 230-second (Consumption) or configurable function timeout from being exceeded, and Queue triggers scale out automatically based on queue depth. It also provides built-in retry and poison-queue handling for failed image analyses.

Answer

Use Azure Logic Apps with a synchronous batch loop.

Answer

Trigger a single synchronous Azure Function that loops through all 500 images sequentially within a single HTTP request.

Answer

Configure an Azure Event Grid subscription to directly invoke Azure AI Vision REST endpoints without compute logic.

Question 13

You are using Azure AI Vision Image Analysis 4.0 to extract text from a photograph of a street sign. The sign contains text in both English and French. You want to ensure the OCR results include the correct language for each text line. Which capability should you use?

Accepted Answer

Read feature with automatic language detection enabled.. The Read feature includes automatic language detection, which identifies the language of each text line. For a sign with English and French, this ensures that both languages are recognized correctly without forcing a single language model. Other features like Dense Captions do not extract text, and custom OCR models are not supported in this service.

Answer

Dense Captions feature to describe the sign content in natural language.

Answer

OCR feature with a custom model trained on English and French text.

Answer

Read feature with the language parameter set to a specific language code.

Question 14

A media company runs an Azure AI Vision Image Analysis 4.0 deployment in a private virtual network. Their application must analyze images stored in a private Azure Blob Storage account that has public network access disabled. The application currently receives an HTTP 400 error when it passes the blob URL in the url parameter. You need to resolve the error with the least development effort while keeping the storage account private. What should you do?

Accepted Answer

Generate a user delegation SAS token with read permission for the blob, append it to the blob URL, and pass the resulting URL in the url parameter. When public network access is disabled, the analyze endpoint cannot anonymously fetch a blob URL. A user delegation SAS scoped to the blob supplies the missing authorization without exposing the account or requiring image downloads. It also works with private endpoints because the service reaches storage through its own network path using the signed URL.

Answer

Recreate the Image Analysis resource inside the same virtual network as the storage account and retry the original blob URL

Answer

Enable public network access on the storage account and restrict access with the storage firewall to the Image Analysis service tag

Answer

Download each blob to the application's local disk and submit the image bytes in the request body of the analyze call

Question 15

A security team monitors a parking garage with Azure AI Vision Image Analysis 4.0. They analyze one frame every five seconds and need to detect whether any person is present near a restricted door, ignoring cars and signage. The frames are 4K resolution and the people are small in the frame. They want to minimize false negatives. What should they do?

Accepted Answer

Request the people feature, crop the frame to the restricted-door region of interest, and lower the confidence threshold when evaluating results. The people feature is the correct detection capability, and its confidence values let you tune sensitivity. Cropping to the door region makes small figures occupy more pixels, which improves recall, while lowering the acceptance threshold trades some precision for fewer missed people. Together these changes target false negatives without changing the camera or network design.

Answer

Request the objects feature and alert when any object is detected in the frame

Answer

Request the tags feature and treat any tag containing person as an alert condition

Answer

Request the read feature and alert when any text is detected near the door

Question 16

You are integrating Azure AI Vision Read API into a high-throughput workflow. You receive an HTTP 202 Accepted response. What action should your application take next?

Accepted Answer

Poll the URL specified in the Operation-Location response header until the status is succeeded.. An HTTP 202 Accepted response from the Azure AI Vision Read API indicates that the request has been accepted for asynchronous processing, and the response includes an Operation-Location header containing a URL to poll for the result. The application must poll that URL (typically with a GET request) until the status field returns 'succeeded', at which point the extracted text is available in the response body. This is the documented asynchronous pattern for the Read API.

Answer

Switch to synchronous Read API endpoint.

Answer

Parse the response body directly for the extracted text results.

Answer

Immediately retry the POST request with an increased timeout value.

Question 17

You have trained a Custom Vision classification model. You need to export the model to run locally on an edge device using ONNX format. Which project domain must you select when creating the project to support export?

Accepted Answer

General (compact). To export a Custom Vision classification model to ONNX for edge deployment, the project must be created with a compact domain — specifically 'General (compact)' — because only compact domains support model export. The compact domain produces a smaller, optimized model suitable for local/edge inference.

Answer

Retail

Answer

Food (compact) for edge deployment

Answer

General

Question 18

You are developing a .NET application that extracts bounding boxes of text from scanned documents using Azure AI Vision 4.0. Which asynchronous API operation should you call first to initiate the OCR process?

Accepted Answer

POST /computervision/imageanalysis:analyze?api-version=2024-02-01. The Azure AI Vision 4.0 Image Analysis analyze endpoint is POST /computervision/imageanalysis:analyze with api-version=2024-02-01, and it supports the 'read' feature for OCR with bounding boxes. This is the correct modern endpoint for initiating asynchronous OCR in the 4.0 API surface.

Answer

POST /vision/v3.2/ocr

Answer

POST /vision/v2.1/recognizeText

Answer

POST /customvision/v3.3/Prediction

Question 19

Your company wants to generate descriptive captions for thousands of product images automatically using Azure AI Vision. Which visual feature parameter should you specify in the analyze request?

Accepted Answer

features=caption. The `caption` feature in Azure AI Vision's Analyze Image API generates a human-readable, one-sentence description of the entire image content. It is specifically designed for scenarios like automatically describing product images at scale. The API returns a `captions` array with `text` and `confidence` values, making it the correct choice for descriptive captioning.

Answer

features=read

Answer

features=smartCrops

Answer

features=objects

Question 20

You need to determine the primary dominant colors and whether an image is predominantly black-and-white using Azure AI Vision. Which visual feature parameter provides this information?

Accepted Answer

features=color. The features=color parameter in Azure AI Vision returns color-related analysis, including the dominant foreground and background colors and whether the image is black-and-white (isBWImg). This directly provides the dominant colors and monochrome determination requested. Other features return captions, people detection, or tags, not color metadata.

Answer

features=denseCaptions

Answer

features=people

Answer

features=tags