AI-102 Vision and Document AI

The third choice should be General compact, in other that the model can be exported to be used in iOS device

Classification
Multiclass
General (compact)

General (Compact) is the right choice for iOS: https://learn.microsoft.com/en-us/azure/ai-services/custom-vision-service/export-your-model

I chose:
Classification
Multiclass
general (compact)

Classification
Multiclass
General (compact)

classification
multiclass
General (compact): Since the model will be used in an iOS app, a compact model is preferred for performance and size reasons. The "General (compact)" domain is suitable for a wide range of image classification tasks and is optimized for mobile and edge devices.

1. Classification
2. Multiclass
3. General (compact)

1. Classification
2. Multiclass
3. General (compact)

1. Classification
2. Multiclass
3. General (compact)

Final Answer:
Classification
Multiclass
General (compact)

I also think that General(compact) https://docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/export-your-model
1. It can be running offline
2. Real time locally

Correct

Microsoft copilot:

For expenditure request authorization forms, you should choose the Prebuilt layout model. This model is designed to handle structured or semi-structured documents, making it suitable for forms with fields and values.

For structured employment application forms, you should choose the Custom template model. This model is well-suited for documents with a consistent and predictable layout, making it ideal for structured forms like employment applications.

For structured and unstructured survey forms, you should choose the Custom neural model. This model is designed to handle a variety of document types, including those with mixed structures, making it ideal for extracting data from both structured and unstructured survey forms.

Remember, the data needs to end up in a DB - structured, consistent.
None of the prebuilt model allow for this request.
a Custom model will allow this level of tuning.
custom Neural for the structured/unstructured surverys - and custom template for the others

A and B are both necessary steps...but we're only allowed to pick one. Before you can create person objects (with up to ~250 images for each person object) you have to create the person model. However, the person model is created for you automatically -- it's not a separate step. Follow the guide here: https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/quickstarts-sdk/identity-client-library?tabs=linux%2Cvisual-studio&pivots=foundry-portal

Option A alone is incomplete. Creating a person model is a step, but you must also create person objects and upload face images so Video Indexer can match faces in videos to named people; that’s why option B is the correct choice.

In Video Indexer, you create a Person Model.

You upload a few sample images of your CEO, HR manager, Sales lead, etc.

Video Indexer then matches those faces across all your videos.

Now, if you search "CEO," it instantly finds all 20 videos where the CEO appears.

B. Create person objects and provide face images for each object
Why B is the best choice:
Person objects and their face images are what enable face recognition and video search in practice.

Without creating person objects and supplying reference face images, the person model (Option A) remains empty and cannot recognize anyone.

Azure Video Indexer automatically creates a default person model if you don’t create a custom one—so the system can still function with just person objects and faces

Why not A?
Structural container (optional if you use the default).

So, while Option A sets the structure, Option B does the real work of enabling searchable face-based indexing in videos. Choose B as the best single answer

I say this answer is A.

A is right.

You can use a Person model to index your new video by assigning the Person model during the upload of the video.

You can use a Person model to index your new video by assigning the Person model during the upload of the video.

https://learn.microsoft.com/en-us/azure/azure-video-indexer/customize-person-model-with-website

Customers can build custom Person models and enable Azure AI Video Indexer to recognize faces that aren't recognized by default. Customers can build these Person models by pairing a person's name with image files of the person's face. You can use the default model if you like but the point is to add persons and their associated image files to the model : https://learn.microsoft.com/en-us/azure/azure-video-indexer/customize-person-model-with-website.

Should be A

seems right
https://docs.microsoft.com/en-us/azure/azure-video-indexer/customize-person-model-with-website

The current name for Form Recognizer (as of May 2024) is Document Intelligence.

1. Document Intelligence(Form Recognizer)
2. "S0", "eastus"

https://azure.microsoft.com/en-us/pricing/details/ai-document-intelligence/

Is it asking me to know it is eastus and not useast?? In this era this is important knowledge for Microsoft?
What does knowing these things prove? Microsoft should do an internal audit and shutdown in shame

To be honest if you're a developer you notice one parameter missing (name). So the code is not valid. But for the sake of answer we go with tier before location,

1. Document Intelligence (Form Recognizer)
2. "S0", "eastus"

1. FormRecognizer -> Document Intelligences service.
2. S0, "eastus"

Forms recognizer comes in 2 tiers one is F0 and another is S0. Apologizes on listing out S1 Ties as it doesnt exists with Document Intelligences service. Sorry for the confusion

Tier: “S1”
Location: “eastus”

To create a Standard tier resource that converts scanned receipts into text using Form Recognizer, you should call the provision_resource method with the following parameters:

Resource Name: “res1”
Kind: FormRecognizer

Document Intelligence (formerly known as Form Recognizer) is an Azure AI service that specializes in extracting structured data from unstructured documents, including receipts.

A. Face. The solution link explains:

Embed facial recognition into your apps for a seamless and highly secured user experience. No machine-learning expertise is required. Features include face detection that perceives facial features and attributes—such as a face mask, glasses, or face location—in an image, and identification of a person by a match to your private repository or via photo ID.

Why Computer Vision is the best choice
Azure Computer Vision provides prebuilt image analysis capabilities, including:

✅ Detecting people and objects
✅ Identifying whether a face is wearing a mask
✅ Image-based analysis (snapshots every 15 minutes fit perfectly)
✅ Simple REST API → minimal development effort
✅ Lower cost compared to video analytics services

You can periodically capture still images from cameras and analyze them for PPE compliance, which is much cheaper and simpler than processing continuous video.

Why not face ?
Focuses on face detection and recognition
Limited PPE detection (primarily face attributes)
Not designed for general safety gear detection
More restrictive and higher compliance/privacy concerns

asked chatGpt and its suggest computer vision too

If it weren't for the "Identify staff .." requirement it would be B. Because it needs to identify staff who removed PPE, its A.

Focuses on facial recognition and emotion detection—not designed for detecting PPE like masks or glasses.

B is the right one

Computer Vision is the best choice for your needs:
Why: It offers a flexible, cost-effective solution for detecting PPE compliance (masks and safety glasses) using custom-trained models. Its API-driven approach minimizes development effort, and it can be integrated into a system that processes video or image snapshots every 15 minutes. Costs are manageable since you’re only analyzing periodic snapshots rather than continuous video streams.

Yes, Azure Computer Vision can be used to monitor staff compliance with personal protective equipment (PPE) requirements by detecting whether employees are wearing the necessary safety gear, such as helmets and masks, in real-time. This technology helps ensure workplace safety and compliance with safety standards.

For monitoring staff compliance with PPE requirements in your factory, Azure Video Analyzer for Media (formerly Video Indexer) is the most suitable option. This service can identify staff who have removed masks or safety glasses, perform compliance checks every 15 minutes, and minimize both development effort and costs

Video indexer is the correct answer as it has built-in mask detection.

https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/intro-to-spatial-analysis-public-preview?tabs=sa#social-distancing-and-face-mask-detection

Face meets the requirements of identifying staff and checking PPE compliance effectively while keeping both development effort and costs low.

To meet the requirements of identifying staff who have removed masks or safety glasses, performing a compliance check every 15 minutes, and minimizing both development effort and costs, you should use:

B. Computer Vision

Azure Computer Vision can analyze images and videos to detect whether staff are wearing masks and safety glasses. It can be set up to perform checks at regular intervals, meeting the compliance requirement.

For more information, you can refer to the official - https://learn.microsoft.com/en-us/azure/cognitive-services/computer-vision/overview

The Face API does indeed detect human faces and can return various attributes, which is essential for applications where face recognition or verification is needed.

However, in the specific context of monitoring PPE compliance, while face detection is a crucial step, it might not be the most cost-effective and efficient choice compared to using the Computer Vision API directly, especially if the primary task is identifying PPE like masks and safety glasses. The Computer Vision API can perform both object detection (including PPE) and OCR, which might streamline the process with minimal development effort and cost.

Because you have to IDENTIFY staff, it is definitely Face.

"A" makes sense. The task is to recognize people, that DO NOT wear their equipment, not to see who is wearing.
Why check for social distancing, identifying masks, goggles, etc. if we only need to find the people that are not wearing that stuff and therefore are easily detected by Face.

Copilot: B

To build a monitoring solution for staff compliance with PPE requirements, you should use B. Computer Vision. This service can analyze images and videos to detect whether staff are wearing masks or safety glasses. It is cost-effective and requires minimal development effort compared to other options

C is the right option

Answer is correct:
As per https://learn.microsoft.com/en-us/azure/azure-video-indexer/customize-language-model-with-website

This step creates the model and gives the option to upload text files to the model.

B is the correct answer

Upload text file(s) to train the language model. The file can either contain a list of words as you would like them to appear in the Video Indexer transcript or the relevant words included naturally in sentences and paragraphs. As better results are achieved with the latter approach, it's recommended for the upload file to contain full sentences or paragraphs related to your content.

B is correct.

B is right.

I agree, B is the option

File 2 and 5 are excluded.

New service limits now goes up to 500MB so...
File 1, 3, and 6 are correct for "training the model", however if MSFT remove the word "training" from the question - be careful.

https://docs.microsoft.com/en-gb/learn/modules/work-form-recognizer/3-get-started
https://docs.microsoft.com/en-us/azure/applied-ai-services/form-recognizer/service-limits?tabs=v21

Correct answer: ACDF
https://learn.microsoft.com/en-gb/training/modules/work-form-recognizer/3-get-started

Understand Form Recognizer file input requirements
Form Recognizer works on input documents that meet these requirements:
- Format must be JPG, PNG, BMP, PDF (text or scanned), or TIFF.
- The file size must be less than 500 MB for paid (S0) tier and 4 MB for free (F0) tier.
- Image dimensions must be between 50 x 50 pixels and 10000 x 10000 pixels.
- The total size of the training data set must be 500 pages or less.

File1 PDF 20 MB✅ YesSupported format and under 50 MB
File2 MP4 100 MB❌ NoVideo files are not supported
File3 JPG 20 MB✅ YesSupported format and under 50 MB
File4 PDF 100 MB❌ NoExceeds 50 MB size limit
File5 GIF 1MB❌ NoGIF is not a supported training format
File6 JPG 40 MB✅ YesSupported format and under 50 MB

Form Recognizer works on input documents that meet these requirements:
- Format must be JPG, PNG, BMP, PDF (text or scanned), or TIFF.
- The file size must be less than 500 MB for paid (S0) tier and 4 MB for free (F0) tier.
- Image dimensions must be between 50 x 50 pixels and 10000 x 10000 pixels.
- The total size of the training data set must be 500 pages or less.

ACF is correct, for training the maximum limit is 50MB, for production the limit can go upto 500MB

Form Recognizer model was removed from AI-102 exam

1 3 and 6

1 3 6 YES!

For custom model training the total size is still 50MB. Answer is correct

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-custom-classifier?view=doc-intel-4.0.0

the question precises custom model ---- size limit 50MB
see this link https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/service-limits?view=doc-intel-3.1.0
Correct answers : ACF

I think A,C,F are correct

appeared in august exam

Based on filetype: PDF, JPG; and less than 50 MB of file size.

The supported file types for training a Form Recognizer model are PDF, PNG, and JPEG. Therefore, you can use only the files that have one of these formats for training the model. Based on the information provided, the files that can be used for training the model are:

A. File1 (PDF)
C. File3 (JPG)
F. File6 (JPG)

Therefore, the correct answers are:

A. File1
C. File3
F. File6

It should allow all except the GIF and MP4 on the paid tier, free tier supports file up to 4MB: https://docs.microsoft.com/en-us/azure/applied-ai-services/form-recognizer/concept-read#input-requirements

Supported file formats: JPEG/JPG, PNG, BMP, TIFF, and PDF (text-embedded or scanned). So there is more than 3 correct responses. PDF and JPG files (2 each, therefore 4 possible responses even though question asks for 3)

1 & 3 correct.
But GIFs are also allowed, if I remember right.

Box 3 is Yes, the stream will be generated from a local image!

Box 3 should be Yes, a stream is only a pathway for data. and in this case the data actually comes from a local file. The correct answer would be No, Yes, Yes.

No, Yes, Yes

The code will perform face recognition. No
The code will list tags and their associated confidence. Yes
The code will read a file from the local file system. Yes

1. **The code will perform face recognition.**
**Answer:** No
The code uses the `ComputerVisionClient` to analyze an image for visual features such as `Description` and `Tags`, not for face recognition.

2. **The code will list tags and their associated confidence.**
**Answer:** Yes
The code retrieves tags and their confidence scores from the `results.Tags` property and outputs them.

3. **The code will read a file from the local file system.**
**Answer:** Yes
The code uses `File.OpenRead(localImage)` to read an image file from the local file system.

N - Y - Y

The code will perform face recognition --> No
The code will list tags and their associated confidence --> Yes
The code will read a file from the local file system --> Yes

The code will perform face recognition.
No. The code is set to analyze image descriptions and tags (VisualFeatureTypes.Description, VisualFeatureTypes.Tags), but it does not include face recognition.


The code will list tags and their associated confidence.
Yes. The code includes a loop that iterates through results.Tags and prints each tag's name and confidence.

The code will read a file from the local file system.
Yes. The code uses File.OpenRead(localImage) to open and read an image file from the local file system.

So, the answers are:

The code will perform face recognition. No
The code will list tags and their associated confidence. Yes
The code will read a file from the local file system. Yes

https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/quickstarts-sdk/image-analysis-client-library?tabs=windows%2Cvisual-studio&pivots=programming-language-csharp#analyze-image

No
Yes
Yes

No Yes Yes
Image analysis
Return the category and confident score
Read local video streaming or streaming source

The fact that the parameter is named 'local', might be misleading - it could be anything.
HOWEVER, I'm not a C# expert, but File.OpenRead tells us that the file is on a filesystem to which the device has access.
SO my vote is NYY

Final Answer:
N Y Y

Box 1: No. The code generates description and tags. See line 3,4
Box 2: Yes. The code displays tag.Name and tag.Confidence
Box 3: Yes. File.OpenRead reads a local file. See https://docs.microsoft.com/en-us/dotnet/api/system.io.file.openread?view=net-6.0

No
Yes
Yes

Answer is NYY

https://docs.microsoft.com/en-us/java/api/com.microsoft.azure.cognitiveservices.vision.computervision.computervision.analyzeimageinstreamasync?view=azure-java-legacy

The anwser is correct
Create a project
Upload and tag images
Train a classifier model
https://learn.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?tabs=visual-studio&pivots=programming-language-csharp
For Type of model
https://azure.microsoft.com/en-us/use-cases/defect-detection-with-image-analysis/

I think the last box should be "Train the object detection model".

Object detection is better suited for quality control applications in manufacturing because it provides both the identification of the fault and its precise location.

1) Pinpointing defects: in a factory setting, knowing where the fault is on the component is critical for automated systems to discard or repair the part. Classification only tells you the part is 'defective', not the exact spot of the defect.
2) Multiple faults: a single component might have multiple, different faults. Object detection can identify and label each specific fault individually within the same image.
3) Actionable data: the bounding box output provides actionable data for downstream automation systems, like robotic arms programmed to interact with the component based on specific coordinates.

why not object detection?

Got a similar one in Oct 2024, some wording change from memory.

Answer is correct, it is a classifier since custom vision only builds an image classifier, detection is part of normal vision.

Got this in the exam, Jun 2024.

1. Create a project
2. Upload and tag images
3. Train a classifier model

1. Create a project
2. Upload and tag images
3. Train a classifier model

Final Answer:

Create a project
Upload and tag images
Train a classifier model

I think the key is recognizing it's a production line. That means the same component is coming down the line.

Then there are other lines producing other components.

Classification would be best suited for that scenario. If we were looking at an image that might have many different components in one image and want to find the location of different faults, then object would be more appropriate.

It's a classifier model because it's not detecting whether they objects are there or not, it's classifying them as faulty or not

I would pick object detection. Custom Vision functionality can be divided into two features. Image classification applies one or more labels to an entire image. Object detection is similar, but it returns the coordinates in the image where the applied label(s) can be found. If I were the users, I would certainly want know where the faults are located to make sure and have a second look and it’s quite useless by just telling me there are something wrong but can’t tell you where they are as I need to do extra work to find them!

Yes the anwser is correct
Create a project
Upload and tags images
Train the classifier model
https://learn.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?tabs=visual-studio&pivots=programming-language-csharp
and for type of model
https://azure.microsoft.com/en-us/use-cases/defect-detection-with-image-analysis/

Correct answer should be:
Create
Upload & Tag
Train the object detection model

The question was to help "detect" common faults. Detection means where the fault actually is in the image.

Create
Upload
Train the classifier

Quite confusing on the questions, since Object Detection technically can be correct IMO

Correct

Correct according to ChatGPT.

https://learn.microsoft.com/ja-jp/azure/ai-services/document-intelligence/prebuilt/business-card?view=doc-intel-4.0.0

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0#model-analysis-features

Correct answer is Correct Answer: A. Upgrade the business card model to v3.0 as per Copilot, this is the explanation given. The business card model v2.1 does not support QR code extraction.
Starting from v3.0, the prebuilt-businessCard model includes QR code support, allowing it to extract data encoded in QR codes on business cards.
Upgrading to v3.0 or v3.1 is a simple change that does not require retraining or custom model development, making it the least effort solution.

Correct answer is Correct Answer: A. Upgrade the business card model to v3.0 as per Copilot, this is the explanation given. The business card model v2.1 does not support QR code extraction.
Starting from v3.0, the prebuilt-businessCard model includes QR code support, allowing it to extract data encoded in QR codes on business cards.
Upgrading to v3.0 or v3.1 is a simple change that does not require retraining or custom model development, making it the least effort solution.

From AI:

To enable QR code interpretation with minimal administrative effort, the correct first step is A. Upgrade the business card model to v3.0.

Key rationale:
Prebuilt models like the business card model in v3.0 (or later) support add-on capabilities such as barcode/QR code extraction via the features=barcodes parameter.
The v2.1 business card model (likely referenced here) predates add-on features and lacks native QR code support without custom code.
Option B (read model) extracts text but does not inherently process QR codes.
Option C (custom model) requires training and introduces unnecessary complexity for QR code extraction.
Option D (contract model) is unrelated to QR codes or business cards.

Correct answer is Option A

Can technically read QR codes immediately however only returns raw text. You lose structured field extraction like Name, Company, Phone.App1 currently expects structured field output from the Business Card model, switching to the Read model would break the app logic unless you redesign App1.

Document Intelligence supports more sophisticated and modular analysis capabilities. Use the add-on features to extend the results to include more features extracted from your documents.
- barcodes

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept/add-on-capabilities?view=doc-intel-4.0.0&tabs=rest-api

Option A (upgrade the business card model to v3.0) isn't the right choice because even though the newer version of the business card model might include improvements, it doesn't specifically address the need to interpret QR codes.

Implementing the read model (option B) is the most direct way to ensure that App1 can interpret QR codes without requiring additional changes or custom models. The read model includes the capability to read barcodes and QR codes, providing the functionality you need with minimal administrative effort.

B is the answer

To ensure that App1 can interpret QR codes with minimal administrative effort, the best approach would be to:

A. Upgrade the business card model to v3.0.

The business card model version 3.0 includes enhanced capabilities, including the ability to interpret QR codes, which would directly address your requirement without needing to deploy additional models or implement complex configurations.

This upgrade will streamline the process, leveraging the existing capabilities of the business card model and minimizing the need for additional administrative tasks.

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept/add-on-capabilities?view=doc-intel-4.0.0&tabs=sample-code

My answer is based on the model v4.0.

According to my link:
https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0#model-analysis-features

Barcode isn't supported in the prebult Business card model. It's optional in the Read model and enabled in the Contract model. I go for the contract model because of the minimize administrative effort. I will cost more to set it up from the start but since Read can't ensure the QR-reading but if the QR-read fails there will be manual labour fixing it, which can potentially be a lot of work.

Add-on capabilities (includes QR codes extraction) are available within all models except for the Business card model
https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept/add-on-capabilities?view=doc-intel-3.1.0

To ensure that App1 can interpret QR codes with minimal administrative effort, you should implement the read model (Option B).
The read model in Azure AI Document Intelligence is designed to extract text from various types of documents, including QR codes12.
Upgrading the business card model to v3.0 (Option A) or deploying a custom model (Option C) would involve more administrative effort and may not specifically address the need to interpret QR codes.
Implementing the contract model (Option D) is not relevant to this requirement.

The question mentions "must minimize development effort". That means you should lean away from anything custom.

Answers are C and D.

Answers are C and D.

C&D are correct

A, D
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/concept-detecting-adult-content

Correct answers are C and D, verified using CoPilot:

The two tools you can use to ensure inappropriate content is blocked while minimizing development effort are:

C. Azure AI Content Safety and D. Azure AI Vision.

Azure AI Content Safety provides advanced algorithms for processing images and text to detect and block harmful content. Azure AI Vision offers computer vision capabilities to analyze images and detect inappropriate content

CD
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/concept-detecting-adult-content

Answer is C & D

Azure AI Vision & Content Safety

https://learn.microsoft.com/en-us/azure/azure-video-indexer/odrv-download

Copy the embed code and extract only the URL part including the key. For example:

https://onedrive.live.com/embed?cid=5BC591B7C713B04F&resid=5DC518B6B713C40F%2110126&authkey=HnsodidN_50oA3lLfk

Replace embed with download. You will now have a url that looks like this:

https://onedrive.live.com/download?cid=5BC591B7C713B04F&resid=5DC518B6B713C40F%2110126&authkey=HnsodidN_50oA3lLfk

Now enter this URL in the Azure AI Video Indexer website in the URL field.

Option D allows you to directly link the video file from OneDrive to Azure Video Indexer without needing to download and upload large files, saving time and bandwidth. This method provides a seamless way to access and process large files stored in OneDrive.

Options A, B, and C either involve unnecessary steps or do not directly integrate with Azure Video Indexer as efficiently.

Azure Video Indexer supports indexing videos from:

Local upload
Public URLs (including OneDrive sharing links)
Azure Blob Storage


For large files (like 20 GB), uploading manually is inefficient. Instead, providing a shared link from OneDrive allows Video Indexer to fetch the file directly.
A download link (option C) often expires or requires authentication, while a sharing link is designed for external access.

C is correct

not C. - because is only one-time only and temporary.

should be C. https://learn.microsoft.com/ja-jp/azure/azure-video-indexer/considerations-when-use-at-scale

Given this documentation:
https://learn.microsoft.com/en-us/azure/azure-video-indexer/upload-index-media

You can't:
- Use URL's from streaming services like Youtube.
- Use URL's where the file size is > 2 GB (in the troubleshooting section).

Only remaining option is B.

This method allows you to directly use the file from OneDrive without the need to download and re-upload it, saving time and bandwidth.

Tried it with a sharing link for a video in onedrive, it does not work :
"The URL is either invalid, not secured, or there was no file found in the URL"

Got this question in Oct 2024 exam, except the "download link" is replaced by "embed link".

There isn't a separate "download link" option; sharing links serve this purpose. OneDrive uses sharing links to provide access, which can also allow downloading.

D. From OneDrive, create a sharing link for File1.avi, and then copy the link to the Azure AI Video Indexer website.
Explanation:
Azure Video Indexer can index videos from URLs provided by users. For Azure Video Indexer to access the video, the link should be a public sharing link that does not require authentication. Creating a sharing link from OneDrive provides this functionality, allowing Azure Video Indexer to fetch the video for indexing.

These kind of naive questions do not consider that most people who do these exams are corporate employees (following illogical certifcation policies) and they do not use video indexing from one-drive data. Such thoughtless questions. Really, i should read how one-drive generates urls??

I think C and D are equivalent

Answer could be D, according to Microsoft Copilot.
Azure AI Video Indexer typically requires a sharing link to a video hosted online.
Azure AI Video Indexer doesn’t require a separate download link.

Azure AI Video Indexer has File Size Limit:
-If you’re uploading a file directly from your device, the maximum file size allowed is 2 GB.
-When uploading a video from a URL, the file size limit increases to 30 GB.

B. Download File1.avi to a local computer, and then upload the file to the Azure AI Video Indexer website.
Here's why this is the correct approach:
File size: The video file is 20 GB, which is a large file. Azure Video Indexer has file size limitations for direct uploads.
Direct access: Azure Video Indexer typically requires direct access to the video file for processing. Downloading the file to a local computer and then uploading it ensures that the Video Indexer has direct access to the entire file.
Compatibility: This method ensures that the file is in a format that Azure Video Indexer can directly process without relying on external services or links.
Control: Downloading and uploading manually gives you more control over the process and allows you to verify the file before indexing.

When I took the exam 22/06/24, option C has been removed, and replaced with "From Onedrive, create an embed link...". Very probably this means C was NEVER correct.

I also note that the link audlindr cites makes no mention of anything other than uploading (neither does the official az-102 MS course). Given there is no requirement to "minimise time/effort" or automate, I don't see how hacking and pasting URLs are necessary in this scenario when down then uploading will DEFINITELY work - everything else is a punt.

FWIW I chose option B and got 90% for the "image and video" component (i.e. got one image/video question wrong, and I'm 99% certain it wasn't this one). Make of that what you will.

The second one should be generate Thumbnail imho.

Both answers are incorrect.

The correct answers are:
https://contoso1.cognitiveservices.azure.com/
AND
generateThumbnail

westus.dev.cognitive.microsoft.com wouldn't be a correct Computer Vision endpoint if the resource name is contoso1.

Also, per the documentation, areaOfInterest "returns a bounding box around the most important area of the image", it doesn't return a different size photo (https://docs.microsoft.com/en-us/rest/api/computervision/3.1/get-area-of-interest).

https://westus.api.cognitive.microsoft.com/vision/v3.1/generateThumbnail?width={width}&height={height}&smartCropping=true

For this one, there is an update since 1st of July 2019: New resources will use custom subdomain names. therefore the endpoint will be contoso1.xxx

The correct answers are:
https://contoso1.cognitiveservices.azure.com/
AND
generateThumbnail

POST https://westus.api.cognitive.microsoft.com/vision/v3.2/generateThumbnail?width=500&height=500&smartCropping=True


{
"url": "{url}"
}

1. contoso1
2. gererateThumbnail

Final Answer:
1. https://contoso1.cognitiveservices.azure.com/
2. generateThumbnail

API URL:
The base URL for the Analyze Image 4.0 API is typically:
https://<region>.api.cognitive.microsoft.com/vision/v4.0/analyze

Replace <region> with the appropriate Azure region (in this case, West US).

was on exam 28/06/2023

I simulated this in Azure Portal:
1. endpoint is https://contoso1.cognitiveservices.azure.com/
2. thumbnail

https://contoso1.cognitiveservices.azure.com/ is correct.

Context from ChatGPT:
westus.api.cognitive.microsoft.com is also a valid endpoint for the Cognitive Services APIs, including the Computer Vision API. However, it is important to note that this endpoint is deprecated and will be retired on October 31, 2024.

Therefore, it is recommended to use the newer endpoint format https://<resource-name>.cognitiveservices.azure.com/ for any new development work. This endpoint format follows a more standard Azure resource URL pattern and is also more flexible in terms of geographic distribution and availability.

Hope it helps.

The first is https://contoso1.cognitiveservices.azure.com the second is generateThumbnail
POST https://*.cognitiveservices.azure.com/vision/v3.2/generateThumbnail?width=100&height=100&smartCropping=true&model-version=latest HTTP/1.1
Host: *.cognitiveservices.azure.com
Content-Type: application/json

{"url":"http://example.com/images/test.jpg"}

on my exam (2023-01-16 Passed)

My Answer:
https://westus.api.cognitive.microsoft.com
But I think this is wrong.Because Question request use contoso1!
So correct answer is :
https://contoso1.cognitiveservices.azure.com/

Westus
generateThumbnail

https://docs.microsoft.com/en-gb/azure/cognitive-services/computer-vision/how-to/generate-thumbnail#call-the-generate-thumbnail-api

curl -H "Ocp-Apim-Subscription-Key: <subscriptionKey>" -o <thumbnailFile> -H "Content-Type: application/json" "https://westus.api.cognitive.microsoft.com/vision/v3.2/generateThumbnail?width=100&height=100&smartCropping=true" -d "{\"url\":\"https://upload.wikimedia.org/wikipedia/commons/thumb/5/56/Shorkie_Poo_Puppy.jpg/1280px-Shorkie_Poo_Puppy.jpg\"}"

First one is contoso.cognitive services. Just checked my own script and cognitiive services uses ur rg name in the endpoint URI.

For first dropdown, 3rd option works with cognitive service key and computer vision key as well. whereas 2nd option works with only computer vision key. so answer 3rd works in both situation. therefor i'll go with https://westus.api.cognitive.microsoft.com/vision/v3.1/generateThumbnail?width=500&height=500&smartCropping=True

I think answer is correct.
PUT: puts a file or resource at a specific URI, and exactly at that URI.
If there's already a file or resource at that URI, PUT replaces that file or resource.
If there is no file or resource there, PUT creates one.
POST: POST sends data to a specific URI and expects the resource at that URI to handle the request.

Although in Web Programming, and API dev, PUT is an http(s) request method for an update operation. I can however create a resource when there is no resource to update. I don't know why precisely but the method used by Azure to actually make a REST API request to create a resource is actually "PUT". So, the answers are correct.
See Ref:
https://docs.microsoft.com/en-us/rest/api/resources/resources/create-or-update

1. PUT
2. CognitiveServices

The answer is
PUT
congnitiveservices
Why PUT? Because while POST can also create new resources like PUT but its not idempotent like PUT hence its always better to us PUT when creating services.

PUT and CognitiveSearvices

Cognitive Services are "AI parts" that mimic human cognition (Cognitive) and are immediately available as WebAPI. Embedded applications can build cognitive solutions for vision, speech, language, decision making, and search without the need for technical knowledge in AI or data science.

When i differed with ChatGPT about PUT being correct, this is what I got

When creating a new resource in Azure, the PUT method is used to update or create the resource with the provided configuration. In the case of Azure Cognitive Services, you typically use the PUT method to provision a new instance of the service with the specified settings.

Therefore, the correct HTTP request to create the new resource should use the PUT method, not POST.

Answer from ChatGPT, PUT and cognitiveservices
curl -X PUT -H "Authorization: Bearer {access_token}" -H "Content-Type: application/json" \
-d '{"kind":"TextAnalytics","sku":{"name":"F0"}}' \
"https://management.azure.com/subscriptions/{subscription_id}/resourceGroups/{resource_group_name}/providers/Microsoft.CognitiveServices/accounts/{api_name}?api-version=2021-04-30"

The given answers are CORRECT

PUT is correct.
https://learn.microsoft.com/en-us/rest/api/resources/resources/create-or-update

Yes, PUT
https://learn.microsoft.com/en-us/rest/api/cognitiveservices/accountmanagement/accounts/create?tabs=HTTP

1. PUT
2. CognitiveServices

PUT
CognativeServices

PATCH is eliminated (It is only used for update). I think we can use both POST and PUT (to create resources). But good to use PUT (just in case if API has been re-trigerred, So it will not fail.

And 2nd answer is "Cognitive services" which provides a lot of models. (So we can use computer vision as well). if we select only computer vision, then we can't use Sentiment analysis and OCR (for which we are trying to create a resource).

May be the reason to use PUT is because , PUT requests must be idempotent. If a client submits the same PUT request multiple times, the results should always be the same (the same resource will be modified with the same values). POST and PATCH requests are not guaranteed to be idempotent.

Shouldn’t the first one be POST? It says it’s a new resource created, not an existing one updated.

I think the answer is as follows.

Create a custom model project and link the project to sa1
Upload five sample documents.
Apply labels to the sample documents.
Train and test the model.

Given answer seems correct since documents would be uploaded in SA already. https://learn.microsoft.com/en-us/training/modules/work-form-recognizer/9-form-recognizer-studio

If you did not create custom model project , where do you upload images ? so for me:
Create a custom model project and link the project to sa1
Upload five sample documents.
Apply labels to the sample documents.
Train and test the model.

This came in the exam 28/06/2023

"Starting on the 20th of September, 2023 you won’t be able to create new Metrics Advisor resources. The Metrics Advisor service is being retired on the 1st of October, 2026."

source: https://learn.microsoft.com/en-us/azure/ai-services/metrics-advisor/overview

Azure Anomaly Detector (D): While it supports multivariate detection across up to 300 signals , it focuses primarily on detection rather than providing built-in root cause analysis and alerting capabilities, requiring additional development work

https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/detecting-methane-leaks-using-azure-metrics-advisor/3254005

Azure AI Anomaly Detector is the optimal choice for this scenario due to its prebuilt models for multi-sensor correlation analysis and low-code implementation.
Why Not Other Options?
A. Metrics Advisor: Focuses on business metrics (e.g., revenue, web traffic), not IoT sensor data, and is being deprecated.
B. Form Recognizer: Extracts text/data from documents, irrelevant for sensor analysis.
C. Azure Machine Learning: Requires custom model development, increasing time/resources compared to Anomaly Detector’s prebuilt APIs.

Azure Metrics Advisor was retired on September 20, 2023 and replaced by Azure AI Anomaly Detector. If the Exam is updated, is D

To build an app that performs anomaly detection across multiple correlated sensors, identifies the root cause of process stops, and sends incident alerts while minimizing development time, you should use Azure Metrics Advisor.

Azure Metrics Advisor is designed for monitoring and detecting anomalies in time-series data, making it suitable for handling multiple correlated sensors. It also provides root cause analysis and incident alerting capabilities

Should be A, but it is retired. D

A. Azure Metrics Advisor. It’s designed for anomaly detection across multiple data streams, can identify root causes, and send incident alerts—all while minimizing development time

Should be A, But starting on the 20th of September, 2023 you won’t be able to create new Metrics Advisor resources. The Metrics Advisor service is being retired on the 1st of October, 2026.
Source: https://learn.microsoft.com/en-us/azure/ai-services/metrics-advisor/overview

To build an app that performs anomaly detection across multiple correlated sensors, identifies the root cause of process stops, and sends incident alerts while minimizing development time, you should use:

D. Anomaly Detector

The Anomaly Detector service is designed to handle multivariate anomaly detection, which means it can analyze multiple correlated signals and identify anomalies. It also helps in pinpointing the root cause of anomalies and can be integrated with alerting mechanisms to send incident alerts⁵⁶. This service is well-suited for your requirements and minimizes development effort by providing pre-built models and easy integration.

Azure AI Metrics Advisor

https://learn.microsoft.com/en-us/azure/ai-services/metrics-advisor/overview

Streaming data Metric Advisor where as batch data Anamoly detector.

A is the answer.

D is Correct for anomaly detection

A is right. From Takedajuku perspective, if you study for 4 days and spend 2 days reviewing, you will have a better chance of passing the exam.

Why is A?

Actually the model should be retrained prior to publishing:

"From the top of the page, select Train to retrain using the new domain."

So it should be:
1. Change the model domain
2. Retrain
3. Publish

Change the model domain {Retail(compact)}
Retrain the model
Export the model

https://docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/export-your-model

Why testing part is not required?

1. Change the model domain
2. Retrain the model
3. Export the model

Testing the model is important for quality, but not strictly required for deployment preparation

There is no mention of whether or not they will actually deploy it, although they plan to deploy it.
If they only plan to deploy it, isn't the current answer correct?

1. Change the model domain.
2. Retrain the model.
3. Test the model.

Change the model domain Retail(compact)
Retrain the model
Export the model

But why they ask 3 and answer is 4.
Btw...
1. Change the model domain
2. Retrain
3. Publish

You train a Custom Vision model to identify a company's products by using the Retail domain.

You are already using the retail domain, so you technically don't need to change the domain. You are only preparing it for deployment on a mobile app.

So ...

To prepare the Custom Vision model for deployment as part of an app for Android phones, you should perform the following three actions in sequence:

Retrain the model. - Ensure the model is trained on the latest data to improve accuracy.
Test the model. - Validate the model's performance to ensure it meets the required standards.
Export the model. - Export the trained and tested model for integration into the Android app

Change the model domain
Retrain the model
Export the model

1. Change the model domain
2. Retrain
3. Export

1. Change the model domain
2. Retrain
3. Export

The Android App has Internet Connectivity, so there is no need for the Compact Version and also no need (and not even the option) for an Export. Therefore.
1. Change the Model
2. Retrain the Model
3. Test the Model.

Final Answer:
Change the model domain {Retail(compact)}
Retrain the model
Export the model

was on exam 28/06

Change the model domain: Since you trained the model using the Retail domain, you need to switch the domain to one that is optimized for mobile devices such as the General (compact) domain.

Retrain the model: After changing the domain, you need to retrain the model using the new domain settings.

Export the model: Once the model is retrained, you can export it in the format that is compatible with your Android app. The model can be exported as a TensorFlow or Core ML model for deployment on Android.

The model was trained, we must to test it, chage the model domain and export it.
https://learn.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/test-your-model (How-to guides) the first is test, second change domain in Export models finaly export model

on my exam (2023-01-16 Passed)

My Answer:
1. Change the model domain
2. Retrain
3. Publish

as per link in solution

this appeared in the exam 28/06

while (true)
{
var results = await client.GetReadResultAsync(operationId);

if (results.Status == OperationStatusCodes.Succeeded)
{
// Process the results
break;
}
else if (results.Status == OperationStatusCodes.Failed)
{
// Handle failure
break;
}

// Delay to avoid hammering the service with requests
await Task.Delay(1000);
}

as per link in solution

results.Status
GetReadResultAsync

was on exam 7 Jun 2022

C and D are the correct answers.

Keyword "jargon", so choose "A. Create a custom language model" .

A https://learn.microsoft.com/en-us/azure/azure-video-indexer/customize-language-model-overview

This question was in the test I took on August 2024

Sorry, A is the correct answer.

B is the correct answer.

Is this question still available on May 21, 2024?

Question is incomplete and needs more context to determine the answer.

A. Create a custom language model: This option allows for the customization of the language model used for transcribing audio and video content. By creating a custom language model, you can train it to understand and transcribe specialized jargon and industry-specific terminology with high accuracy. This directly addresses the requirement to transcribe jargon with high accuracy and supports searches for equivalent terms by ensuring that the transcriptions are as accurate and relevant as possible.

The requirement "Can transcribe jargon with high accuracy" wouldn't be met with the answer B.

--- Can transcribe jargon with high accuracy
Video Indexer (VI), the AI service for Azure Media Services enables the customization of language models by allowing customers to upload examples of sentences or words belonging to the vocabulary of their specific use case. Since speech recognition can sometimes be tricky, VI enables you to train and adapt the models for your specific domain. Harnessing this capability allows organizations to improve the accuracy of the Video Indexer generated transcriptions in their accounts.

https://azure.microsoft.com/en-us/blog/new-ways-to-train-custom-language-models-effortlessly/

Leaning towards A on this.
https://azure.microsoft.com/en-us/blog/new-ways-to-train-custom-language-models-effortlessly/
https://docs.microsoft.com/en-us/azure/azure-video-indexer/video-indexer-overview

I think this is B... because even A makes sense... but u have to extract this information from video somehow. So B

should be A