Learn step-by-step how to create and train object detection models in Azure Custom Vision. Explore best practices for handling large datasets and achieving high mean Average Precision (mAP).
Table of Contents
Question
You need to build a Custom Vision model for Xerigon Corporation. The model would do the following:
- Identify compact pickup trucks.
- Identify where the compact pickup trucks are located on a street.
The model will have large datasets and require extensive training time. The results will not be deterministic.
You would expect a +/- 1% mean Average Precision (mAP) difference with the same training data provided.
Which project type and domain should you use?
Drag the appropriate option to the Project type and Domain.
Options:
- Classification
- Object detection
- General [A1]
- General
- Logo
- Products on Shelves
- General (compact) [S1]
- General (compact)
Answer
Project type: Object detection
Domain: General [A1]
Explanation
The following displays a screenshot of creating a new project.
When you create a new project, the project can be one of two types:
- Image Classification – This project type is referred to as image classification, which assigns one or more labels to an entire image. You would choose this project type if you need to know what is in an image. For example, you want to categorize an image into several classes. You could identify if the image contains a car, bus, or truck.
- Object Detection – This project type finds and locates multiple objects inside an image. Choose this project type if you need to know what is in the image and where the objects are located. For example, you could detect multiple trucks in the image and where they are on the street.
A project type can have several domains. The following lists the different domains that are available for the Image Classification project type:
- General: This domain is used for various image classification tasks, providing standard accuracy and inference time. If you do not know which domain to pick, select a General domain such as General, General [A1], or General [A2].
- General [A1]: This domain has higher accuracy with the same inference time as the General domain but requires longer training time. You would use this domain for a problematic user scenario or a larger dataset.
- General [A2]: This domain has higher accuracy with quicker inference time than either General [A1] or General and requires less training time. It could be used for most datasets.
- Food: As the name implies, this domain is optimized for images of food, such as restaurant dishes and pictures on menus. If you have pictures of different vegetables, you would choose the food domain to classify them.
- Landmarks: This domain is used to find landmarks, both real and artificial. You would choose this domain if the landmark is visible, not obscured. However, the domain can work if the landmark is slightly blurred.
- Retail: This domain is used to find products in catalogs or on retail websites. You would use this domain to distinguish between products, for example, different types of clothes such as pants, sweatshirts, polo shirts, T-shirts, and shorts.
- Compact domains: This domain is used for real-time classification on mobile devices. You would use this domain when you want faster performance and lower resource usage, but you can accept less accuracy than standard domains.
The following lists the different domains that are available for the Object Detection project type:
- General: This domain is used for various image classification tasks, providing standard accuracy and inference time. If you do not know which domain to pick, select a General domain such as General or General [A1].
- General [A1]: This domain has higher accuracy with the same inference time as the General domain but requires longer training time. You would use this domain for a problematic user scenario or a larger dataset. Results from this domain are not deterministic. You can expect a +/- 1% mean Average Precision (mAP) difference with the same training data provided.
- Logo: This domain is used to find logos in images. You would use this domain for applications to identify and find logos in marketing images.
- Products on shelves: This domain is used to find and classifying products on store shelves. This domain can be used in retail inventory management to locate and identify products on crowed shelves.
- Compact domains: This domain is used for real-time classification on mobile devices. You would use this domain when you want faster performance and lower resource usage, but you can accept less accuracy than standard domains.
Since this scenario wants to locate compact pickup trucks on the street, you would choose object detection as the project type because you want to know what is in the image (compact pickup trucks) and where that object is located in the image (on the street).
You would choose General [A1] as the domain because you are using a large dataset and expecting a +/- 1% mean Average Precision (mAP) difference with the same training data.
Microsoft Azure AI Engineer Associate AI-102 certification exam practice question and answer (Q&A) dump with detail explanation and reference available free, helpful to pass the Microsoft Azure AI Engineer Associate AI-102 exam and earn Microsoft Azure AI Engineer Associate AI-102 certification.