Introducing Azure AI Vision

- [Instructor] Azure AI Vision is Microsoft's pre-trained computer vision offering. Using Azure AI Vision, you don't need to train a computer vision model of your own. Instead, you provide images, and Microsoft uses its pre-trained image model to give you insights based on those images. Now, how this works is you will upload an image, either by providing the raw bytes of that image or by giving Azure a URL to that image where it can access it on the internet. Next, you need to tell Azure which image analysis task it should perform, such as crop suggestions, object detection, tagging, captioning, dense captioning. There are many other options available to you, and you can select multiple of these if you want. You make this call out to Azure, and Azure will respond with insights about your image in JSON format, such as the captions, objects detected, tags, et cetera. And then you can analyze your images using these responses. Now, the really nice thing about all of this is that you're not paying for a constantly running service, such as a web application. Instead, this is priced at a per usage level. So each image analysis task that you want to perform and each time you perform it is when you get billed. Okay? Now, you may have heard Azure AI Vision referred to prior names prior to summer of 2023. Azure AI Vision was previously called Azure Computer Vision, and it also had Azure Custom Vision wrapped into that. Microsoft rebranded things related to Azure Cognitive Services to Azure AI Services, and so you'll see a lot of older documentation and things referencing older names. But nowadays we call this Azure AI Vision to reference Microsoft's pre-trained image models. There are two ways of working with Azure AI Vision. The first is to provision an Azure AI services resource, and the second is to use an Azure AI Vision resource. Both of these approaches will allow you to analyze your images, but they have different advantages and disadvantages. Using an Azure AI services resource, you can use other Azure AI services, such as speech, language, and other features, that might not be related to images. Now, this is easier for developers since you really only need to use one key and one endpoint in order to interact with all these AI services. However, the business gets less granular billing and controls. So if you didn't want your developers to be able to use speech but you wanted them to use images, well, Azure AI services wouldn't be what you want to use. Now, you can also use an Azure AI Vision resource, and that's only usable for computer vision purposes. This helps prevent people from accidentally using this Azure AI Vision resource for other purposes that the business did not want to allow. It also helps the business break down costs at a more granular level so you can see what's associated with the Vision versus other services that might exist throughout Azure.